Practical_3_Recurrent_Neural_Networks.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Practical 3: Recurrent Neural Networks",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python2",
      "display_name": "Python 2"
    }
  },
  "cells": [
    {
      "metadata": {
        "id": "9jDBz0IbW3Xy",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Practical 3: Recurrent Neural Networks (RNNs)\n",
        "\n",
        "\n"
      ]
    },
    {
      "metadata": {
        "id": "-0F3Ao8BKa0g",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Introduction\n",
        "\n",
        "Feedforward models (eg deep MLPs and ConvNets) map fixed-size input-data (vectors of a fixed dimensionality) to their output labels. They're very powerful and have been successfully used for many tasks. However, a lot of data is not in the form of fixed-size vectors, but exists in the form of **sequences**. Language is one good example, where sentences are sequences of words. In some way, almost any data types can be considered as a sequence (for instance an image consists of a sequence of pixels, speech a sequence of phonemes, and so forth). \n",
        "\n",
        "Recurrent neural networks (**RNNs**) were designed to be able to handle sequential data, and in this practical we will take a closer look at RNNs and then build a model that can generate English sentences in the style of Shakespeare!"
      ]
    },
    {
      "metadata": {
        "id": "otAAvBVFSZy8",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Learning Objectives\n",
        "* Understand how RNNs model sequential data.\n",
        "* Understand how the vanilla RNN is a generalization of feedforward models to incorporate sequential dependencies.\n",
        "* Understand the issues involved when training RNNs.\n",
        "* Know how to implement an RNN for time-series estimation (**regression**) and an RNN language model (character-level **classification**) in Tensorflow using Keras."
      ]
    },
    {
      "metadata": {
        "id": "gfKcEFUxa--9",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "##Imports\n"
      ]
    },
    {
      "metadata": {
        "id": "h8glXxcyew17",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "6dee02e2-e3b9-4ee2-f7a7-2006dd9979e0"
      },
      "cell_type": "code",
      "source": [
        "#@title Imports (RUN ME!) { display-mode: \"form\" }\n",
        "\n",
        "#!pip -q install pydot_ng\n",
        "#!pip -q install graphviz\n",
        "#!apt install graphviz > /dev/null\n",
        "\n",
        "from __future__ import absolute_import, division, print_function\n",
        "\n",
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "import math\n",
        "import random\n",
        "import ssl\n",
        "import sys\n",
        "import urllib2\n",
        "from IPython import display\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "\n",
        "print('Running TensorFlow version %s' % (tf.__version__))\n",
        "try:\n",
        "  tf.enable_eager_execution()\n",
        "  print('Eager mode activated.')\n",
        "except ValueError:\n",
        "  print('Already running in Eager mode')"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Running TensorFlow version 1.10.1\n",
            "Eager mode activated.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "yVL1OwL7aH8c",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "##From Feedforward to Recurrent Models\n",
        "\n",
        "### Intuition\n",
        "RNNs generalize feedforward networks (FFNs) to be able to work with sequential data. FFNs take an input (e.g. an image) and immediately produce an output (e.g. a digit class), something like this:"
      ]
    },
    {
      "metadata": {
        "id": "NqZsIaRU6-WK",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def ffn_forward(x, W_xh, W_ho, b_hid, b_out):\n",
        " \n",
        "    # Compute activations on the hidden layer.\n",
        "    hidden_layer = act_fn(np.dot(W_xh, x) + b_hid)\n",
        "    \n",
        "    # Compute the (linear) output layer activations.     \n",
        "    output = np.dot(W_ho, hidden_layer) + b_out\n",
        "    \n",
        "    return output"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Kb3Tjms06_XL",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**NOTE**: You don't have to run this cell, it's just shown to illustrate the point.\n",
        "\n",
        "RNNs, on the other hand, consider the data sequentially, and can remember what they have seen earlier in the sequence to help interpret or contextualize elements from later in the sequence when making predictions, something like this:\n"
      ]
    },
    {
      "metadata": {
        "id": "ACx_wHGB7AWc",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def rnn_forward(data_sequence, initial_state):\n",
        "\n",
        "    state = initial_state  # Reused at every time-step\n",
        "    all_states, all_ys = [state], []  # Used to save all states and predictions\n",
        "\n",
        "    for x, y in data_sequence:\n",
        "      \n",
        "      # recurrent_fn() takes the current input and the previous state and produces a new state\n",
        "      new_state, y_pred = recurrent_fn(x, state)\n",
        "      \n",
        "      all_states.append(new_state)\n",
        "      all_ys.append(y_pred)\n",
        "      \n",
        "      # Update state for the next time-step\n",
        "      state = new_state\n",
        "\n",
        "    return all_states, all_ys"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "-Crh_ViE7Ave",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "To understand the distinction between FFNs and RNNs, imagine we want to label words as the part-of-speech categories that they belong to: E.g. for the input sentence \"I want a duck\" and \"He had to duck\", we want our model to predict that duck is a `Noun` in the first sentence and a `Verb` in the second. To do this successfully, the model needs to be aware of the surrounding context. However, if we feed a FFN model only one word at a time, how could it know the difference? If we want to feed it all the words at once, how do we deal with the fact that sentences are of different lengths?\n",
        "\n",
        "RNNs solve this issue by processing the sentence word-by-word, and maintaining an internal **state** summarizing what it has seen so far. This applies not only to words, but also to phonemes in speech, or even, as we will see, elements of a time-series.\n",
        "\n"
      ]
    },
    {
      "metadata": {
        "id": "zUqww79L6Ot-",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Unrolling the network\n",
        "\n",
        "Imagine we are trying to classify a sequence $X = (x_1, x_2, \\ldots, x_N)$ into labels $y$ (for now, let's keep it abstract). After running the `rnn_forward()` function of our RNN defined above on $X$, we would have a list of internal states and outputs of the model at each sequence position. This process is called **unrolling or unfolding in time**, because you can think of it as unrolling the *computations* defined by the RNN loop over the inputs at each position of the sequence.  RNNs are often used to model **time series data** (which we will do in this practical), and therefore these positions are referred to as **time-steps**, and hence we call this process \"unrolling over time\". See the visualization below from this great [blog post](http://www.wildml.com/2015/09/recurrent-neural-networks-tutorial-part-1-introduction-to-rnns/) (it uses slightly different notation, but the idea should be clear):\n",
        "\n",
        "![Unrolled Network](http://d3kbpzbmcynnmx.cloudfront.net/wp-content/uploads/2015/09/rnn.jpg)\n",
        "\n",
        "> **We can therefore think of an RNN as a composition of identical feedforward neural networks (with replicated/tied weights), one for each moment or step in time. **\n",
        "\n",
        "These feedforward functions that make up the RNN(i.e. our `recurrent_fn` above) are typically referred to as **cells**, and the only restriction on its API is that the cell function needs to be a differentiable function that can map an input and a state vector to an output and a new state vector. What we have shown above is called the **vanilla RNN**, but there are many more possibilities. One of the most popular variants is called the **Long Short-Term Memory (LSTM)** cell, which we'll use later to create our Shakespeare language model.\n",
        "\n",
        "### Putting this together \n",
        "\n",
        "In the feedforward models we've seen before, the input $x$ is mapped to an intermediate hidden layer $h$ as follows:\n",
        "\n",
        "\\begin{equation}\n",
        "  h = \\sigma(\\underbrace{W_{xh}x}_\\text{current input (per-example)} + b)\n",
        "\\end{equation}\n",
        "\n",
        "where $\\sigma$ is some non-linear activation function like ReLU or tanh.  We can then make a prediction $\\hat{y} = \\sigma(W_{hy}h + b)$ based on $h$, or we can add another layer, etc.  **NOTE**: We use the weight subscript $W_{xz}$ to indicate a mapping from layer $x$ to layer $z$.\n",
        "\n",
        "RNNs generalize this idea to a sequence of inputs $X = {x_1, x_2, ...}$ by maintaining a sequence of state vectors $h_t$, one for every time-step $t$, as follows:\n",
        "\n",
        "\\begin{equation}\n",
        "  h_t = \\sigma(\\underbrace{W_{hh}h_{t-1}}_\\text{previous state} + \\underbrace{W_{xh}x_t}_\\text{current input (per time-step)} + b)\n",
        "\\end{equation}\n",
        "\n",
        "Feedforward models map one input to one output. On the other hand, RNNs can give **many-to-many** (many inputs, many labels), **many-to-one**, and **one-to-many** mappings. For example, we could use each $h_t$ to predict the part of speech for that time-step ($y_t= \\sigma(W_{hy}h_t+b)$) or we could use the last state $h_T$ to predict the topic of the whole document. Which task the RNN performs is based on the training data (of course)! We can visualize these as follows (fromt this [excellent blog post](http://karpathy.github.io/2015/05/21/rnn-effectiveness/) by Andrej Karpathy):\n",
        "\n",
        "![](http://karpathy.github.io/assets/rnn/diags.jpeg)\n",
        "\n",
        "**QUESTIONS**\n",
        "* How are FFNs and RNNs **similar**?\n",
        "* How are they **different**?\n",
        "* Why do we call RNNs \"recurrent\"?\n",
        "* Can you think of a one-to-many task?"
      ]
    },
    {
      "metadata": {
        "id": "9wuobtH3aB59",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "##Modeling General Time-Series\n",
        "\n",
        "We will train an RNN to model a time-series as a first step. A **time-series** is a series of data-points ordered over discrete time-steps. Examples include the hourly temperature of Stellenbosch over a month or a year, the market price of some asset (like a company's stock) over time, and so forth. We will generate a **sinusoidal time-series** (with or without noise) as a toy example, and then train a tiny RNN model with only 5 parameters on this data."
      ]
    },
    {
      "metadata": {
        "id": "oAQIXUL-oje2",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Create some artificial data"
      ]
    },
    {
      "metadata": {
        "id": "412j4v-RIfYR",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 226
        },
        "outputId": "94a3f6f0-aaaf-41c3-ed80-68372c62f279"
      },
      "cell_type": "code",
      "source": [
        "#@title Create sinusoidal data {run: \"auto\"}\n",
        "steps_per_cycle = 20 #@param { type: \"slider\", min:1, max:100, step:1 }\n",
        "number_of_cycles = 176 #@param { type: \"slider\", min:1, max:1000, step:1 }\n",
        "noise_factor = 0.1 #@param { type: \"slider\", min:0, max:1, step:0.1 }\n",
        "plot_num_cycles = 23 #@param { type: \"slider\", min:1, max:50, step:1 }\n",
        "\n",
        "seq_len = steps_per_cycle * number_of_cycles\n",
        "t = np.arange(seq_len)\n",
        "sin_t_noisy = np.sin(2 * np.pi / steps_per_cycle * t + noise_factor * np.random.uniform(-1.0, +1.0, seq_len))\n",
        "sin_t_clean = np.sin(2 * np.pi / steps_per_cycle * t)\n",
        "\n",
        "upto = plot_num_cycles * steps_per_cycle\n",
        "fig = plt.figure(figsize=(15,3))\n",
        "plt.plot(t[:upto], sin_t_noisy[:upto])\n",
        "plt.title(\"Showing first {} cycles.\".format(plot_num_cycles))\n",
        "plt.show()\n",
        "\n",
        "#both = np.column_stack((t, sin_t_noisy))\n",
        "#print(\"both.shape = {}\".format(both.shape))\n",
        "\n",
        "#print(\"both[:steps_per_cycle, :steps_per_cycle]\")\n",
        "#print(both[:steps_per_cycle,:steps_per_cycle])"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA24AAADRCAYAAAC0JeFaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBo\ndHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzsvXmUJNd13vmL3Nfaq/fGDgTAfSdF\nUSJlLUPLkjyktRzvlqnxyJJtjX1mdMZjj+TjXbJNja2xTHMkHYu2LNmUSVMSV4uLRBgkQYAgiDXQ\nQKP37tqrct9j/njxIrOrszIjIt+Lqjbedw4P0V1VWa8z47137/2++13LdV0MDAwMDAwMDAwMDAwM\nji4Sh70AAwMDAwMDAwMDAwMDg8kwiZuBgYGBgYGBgYGBgcERh0ncDAwMDAwMDAwMDAwMjjhM4mZg\nYGBgYGBgYGBgYHDEYRI3AwMDAwMDAwMDAwODIw6TuBkYGBgYGBgYGBgYGBxxpA57AQYGBgYGRw+2\nbb8Z+EXgNKLItwX8H47jPGzb9nuAX3Uc5z7Fv/N9wA86jvOXFb3eTwE/B/wyUAIuOo7zoRA//3ag\n6TjOt8Z87duBDwJzQAP4m47j/JH3tb8K/DXEHfsy8L84jnN5xn/OQWt8Dxo+CwMDAwODoweTuBkY\nGBgY3ATbti3g9xAJxye9v3s/8Anbts/q+r2O43wc+LjCl/xTwN9xHOfXIv78jwMPAzclbrZtZ4FP\nAD/iOM4Xbdv+fuC3gNO2bb8T+N+BtziOs2Pb9i8B/wL40aj/CAMDAwMDAzCJm4GBgYHBrVgBTgJf\nlX/hOM7HbNt+1HGchm3bANi2/XeAPwdkgJ/wkpgc8P8A3wUMgE8BPwv8O+APHcf5Vdu2jwM3gO9z\nHOe/2bb9Ju/rHwT+nOM432Pb9r8DLgLvBB4AXgD+pPf7/yfgV4Ea8EvAPwde5zjOBble27Z/Efg2\n4CEv2bwLeNFxnH9o2/YF4NeBPwt8L/B24OeBJNAF/gbwIPAXgB+ybfuY4zgfHHl/0sBfcRzni96f\nHwZO2ba9AKwDf95xnB3va58H/tG4N9m27b8A/F3vj18DfsJ7rV9wHOd3vO/5AeAfOo7zhgO+f/T1\nssA/A97rfSYfdhznH3tf+2vATwMWUAF+3HGcZ8aty8DAwMDgaML0uBkYGBgY7Mcm8HXgi7Ztf8C2\n7bsBHMe5MvI9Z4CnHMd5CPg3DBOK/w04C7waeBPwHcCfBr6ISKQAvhORFH679+fvQCQ4+/EjwI8B\n9wKrwPts204Cv4FInB4C7geK+3/QcZyfBR4FftZxnL835rXPOI5jO45zCfgV4E94r/dTwA95kkr5\n86NJG47j1BzH+djIX/1x4AXHcXYdx3nRcZxHAGzbziOSw0/s/+W2bd+FSDjfA9jev+FvIJi7PzPy\nre8DfnvC94/iZ4FXAa9FvP8/bNv2D9i2XQb+AfA2x3EeRCR3f2LMe2JgYGBgcIRhEjcDAwMDg5vg\nOI6LYKI+DvwMcN627Wc8uaRExXGc3/X++wlEIgciIfiw4zg9x3GawG8C38fNidt3IJI9mbi9i/GJ\n2ycdx9l2HKcHPAXcgWDfso7jfNr7nl8m2l32+yP/vQ78pG3bdzqO87DjOH8r6IvYtv06BOv3v+77\n+18E1oB5RK/gfnwf8IjjONe89/vPeK/zn4D32rY97yWpPwj85wnfP4ofBH7FcZy24zh14CPA+4EW\n4AIfsG37uOM4H3UcZ9yaDAwMDAyOMEziZmBgYGBwCxzH2XMc5+cdx3kdcAKRBPy2bdsPed9SGfn2\nPkJmCIIZ2xn52g5wzHGcl4G8Jyd8J/AxhLwwCbwD+MMxy9gb8zsW973+tSj/PmB75L9/CPFvfNy2\n7Sds2353kBfw+tk+hZCJfmn0ax7jtwR8CfiDMT++AuyOfH/LS3avIpi+9yMS2wuO45w/6Pv3veYC\n8Eu2bT9v2/bziKS76DhOF/hu7/VesG37y7ZtvzbIv9HAwMDA4OjAJG4GBgYGBjfBtu0ztm2/S/7Z\ncZw1x3F+AcF6vXrKj68ByyN/Xvb+DkQS817vNWve670fuOw4TjXg8ioIh0iJEwF/7kA4jvOS4zg/\nDhwD/iXwH6f9jMe0fRT4047jfGrk799m2/Y7vNftIZjFt3sJ6yg2EcmY/Lk5r/cPhFzyR4AfRjBw\n075f4hrw047jPOj9727HcX7MW8sTjuP8CCKx/iwQ2F3TwMDAwOBowCRuBgYGBgb7cRb4r95IAABs\n234rQqr49Sk/+/sISV7Stu0i8OeBT3pf+yKiB+4r3p+/AvwtxsskD8I5IO3Z4AP8JEIGGAm2ba/a\ntv3fbNuecxxngOi9k6/XRbBY+3/GQvTZ/ZTjOF/e9+UHgQ/btj3v/fkHgUuO4+zu+75PAd9u2/Zd\n3ut9CPiA97WPIuSjP4yQSU77folPAD/hvfeWbdt/17bt99q2/Vrbtj9q23bGcZwO8BgzvGcGBgYG\nBocDk7gZGBgYGNwEx3G+AvwV4N/Ytu3Ytv0iop/qxxzHuTjlx38ZuAw8g0gQfh+RiIBI3N4OPOL9\n+RGETPILIdbWBv4q8O9s2/4mwm1yQMRExHGcDeAzwNdt234W+G2GCdHHgV+wbfuD+37sHcDrvK89\nP/K/NwH/HpFAfc22bQchV7xlFIBn9PJXEP/2F7z1f9D72jbwR8DLcv7bpO8fwb9GOHE+AzwPPIRw\nqXwaMU/uGdu2nwH+nrcubNv+J7Zt/2SY98zAwMDA4HBgua4puhkYGBgY3J7wWL0asOA4zt60779d\nYNv2rwBPO47zK4e9FgMDAwODowHDuBkYGBgY3Fawbfvrtm3/mPfHHwOe+x8sabsf+H6EI6eBgYGB\ngQFgEjcDAwMDg9sPfxP4v2zbfgExd+0vHvJ6lMG27b8PfA74a/8jJaMGBgYGBrPDSCUNDAwMDAwM\nDAwMDAyOOAzjZmBgYGBgYGBgYGBgcMRhEjcDAwMDAwMDAwMDA4MjjtRhL0BiY6N6JDWbi4sFdnYa\nh70MA4ObYJ5Lg6MI81waHDWYZ9LgKMI8lwaTsLpatg76mmHcpiCVSh72EgwMboF5Lg2OIsxzaXDU\nYJ5Jg6MI81waRIVJ3AwMDAwMDAwMDAwMDI44TOJmYGBgYGBgYGBgYGBwxDFTj5tt268BPgH8kuM4\n/+++r30P8I+BPvApx3H+wSy/y8DAwMDAwMDAwMDA4JWKyIybbdtF4JeBzx/wLf8K+FPAtwPfZ9v2\nq6L+LgMDAwMDAwMDAwMDg1cyZpFKtoHvB67t/4Jt2/cA247jXHYcZwB8CvjuGX7XkcR2pcXH/ugl\n2p3+oa2h0ujw//3es1xaqx7aGnr9AR/5rINzaefQ1gDwpW9e5emXtw51Dc9d3OFbLx3uGr727Br/\n4XMOg8HhGbVuV1p8/vEr9PqDQ1vDYODy9PmtQ30f2t0+v/fIBdZ3m4e2hl5/wH/4nMMTL2wc2hoA\nvvniJs9d2D7cNZzb5A8eu3yoa3ju4g4f+axDt3d498bVzTr/9nefYa/eObQ1dLp9PvfoJaqNw1sD\nwAuXd6k1u4f2+13X5ZNfucBjz68f2hoAvvTEVT79tYuHuoYnXtjgtz9/joF7eGf2pbUq//6zzqHG\ndfVWl88/foV66/Cey3qry6/+/rO8eHXv0NYg767D3htHDZGlko7j9ICebdvjvnwCGI0S1oF7J73e\n4mLhyLrsrK6Wx/79R//wPJ/+ykVWl4q8/7vuj3lVAl/8gxf4yjM3cC7v8C9+5t2sLORjX8PXn73B\nl564ysZei3e9+Y7Yfz/Aja06H/mMQzqV4J/+9Lt44I7F2NfQ7w/40L/6Mq1On4/8vfdSyqe1/r5x\nz+WF6xV+/VPP0e0NeNNDJ3j3m85oXcNB+PVPP8/DT14jmU7yI9/9wKGs4ZMPn+dDH3+Kv/6jb+D7\n3n5n7L/fdV1+8d8/xsNPXqPW6vE3fuyNsa8B4GNfPMcXvnGV5y7u8r3vvBvLOtBlWAnGPZdX1qv8\n6489RTqV4MN/+3tYnMtpXcM4bFdafPj3nqHV6fPut9zBqdVS7GsA+IX/+ATOpR1edc8yf/yddx/K\nGj70u8/y6LNrrCwW+Mn3v+5Q1vCRTz3LR7/wIq2+ywd+6DVaf9dBd/jF6xX+6W9+g+94w2l+9s+/\nResaDsI3nHX+yx+eZ2kux3vfdY/2/TkO7W6f//TFF+l2+7z/u23td9c4XLxR4d/+7jN0egO+881n\nef39q7Gvodvr83//2qNc3ajxmvtXtd8b457LdrfPP/vQIzx3YRvnyh4/94G3H8oz8R9+6xs88vQN\n1vdafPBnvvNQ1vAbn3yWL3zjKk+c2+R733k3qaSx5YD45rhN/cSP6jyL1dUyGxu3slkD1+WRpwTZ\n+HtfPs+3v+o4iUT8D/aXHhfV4+1Km5//8CP83F98a+zreOSbVwF49vwWL17YYr6YifX3A3zuq6JS\n2O0N+Ie//jV+/sffylwh3nU8f3GHakNUyD798Eu85w2ntf2ucc9ltzfgF37j63R7AywLfvMzz/Hg\n6bnYn4das8tXn74OwH/8rMNDZ+Y5vlSIdQ0A/+1R8Uw8/MQV3njPUuy//5NfucDDT4oz4rHn1lhf\nr8R++W3uNvnNzzwPwPWtOo9+6xr3nJrT9vsOOi8//LFv0R+49Dt9fu0TT/EX3/ugtjUchF//5HO0\nvCr6px4+z598V/xJ08Zu01cm/Oc/eIE33rtEMhFvMLK23eDrz94A4LNfvcgfe8MpFsvZWNewsdvk\n4196CYBHnrzGD77jDm1746BnEuAL3hnxlaeuc/HyDoVcvKNtB67Lr33iKUAUFh5/+jp3nhifZOrE\nE+c2fIbpkW9c5o0PxJs0dbp9/slvPEanJxQan/zyeU4txF/c+b1HLnB1owbA5x+9qPXeOOi5/NAn\nnua5C9vkMkkee26N3/r0c3zvW89qW8c4fOulLb7gKRNevLzLw49f5sE74y2GP3thm//yhXMA7FTb\nfO6Rl3nrg8diXcNh4qBiE+hzlbyGYN0kTjNGUnk74/y1Cnu1DpYFm3stvnU+fnncje0Gl9drvO7e\nZd5w3wqX1mqHIsl6+mUhf3IRF8Bh4NHn10kmLL7nLWfYqbb5ytM3Yl/DE+c2/f9+5FB+/wZXNup8\n5+tP8q7XnuT6VoNHn1+LfR1feeYGvb7Lg3cs0OsP+K3Pn4t9DZt7TV66WgHg2Ys7sUs2XdflM1+7\nxFwhzavvWmSn2ubaVvzFqY9/+Tyd3oB3vPo4AF99Nv7n8oXLuzxxbpP7zsxzcrnAHz15jaub9VjX\ncGmtyn9/6jqnVopkUgm++swN3EOQYz36nNiPS3NZNvdafP25+CVAf/DYFVzgNfcs0esP+MzXLsW+\nht/50kv0+gPmSxnWd5tci/l5kJCy9l5/wONO/J/FY8+vc2mt5ifOT760OeUn9OAbIzLqZy/E3/Lw\nxLlNrm7WefcbTrG6kOPxF9ZptnuxrqFS7/D7j1xgrpjhzGqJ5y7uUIlZSlxrdnn0uXXuOFbi73/g\nbZQLaX7nD1+KXVb90S+9SDJh8RfeKxR1nz6EM+JTX72IC/zEDzwEwBe/cSX2NRxVaEncHMe5AMzZ\ntn2Xbdsp4AeAz+n4XYcF2S8iq7ZfOISH6uue7vetDx7jjuNC9rNdacW6hs29Jje2G5w9Jn7/4078\nidv6bpOLN6o8dOci3/VGwXJdWqvFugbXdXni3Aa5TJIHzi7w4pU91mJmkTe8pP1ND6zyA++8C4BH\nnoo3UHddly8/eZ1kwuIn/+RrOLlc4IXLu7EHyXJvzBcztDt9zl2JV6e/U21Tb/V44OwCb3uVSJqe\nOYTizvlrFYq5FH/5+x+ilE/z6HPrsff8ySLGD7/7Xt73HffguvDVZ+J9Lr95bhMX+J/fdTdvuH+F\ntZ0mF27E3xf86HOiwPTX3/86EpbF52O+N9qdPg8/dZ3Fcpafft9rWSxn+fK3rsW6PzvdPo87G5xZ\nLfKj33UfAN84F3/CUm10eOnqHscWRXvBV2J+JmF4Tv3U+15DwrJ48sX4z4j+YMCTL24xX8qQzSR5\n9mL8faiykPPWB4/x7a89Sac74LGYE+kXr+7R7Q347jef4V2vO4nrEvsa5B3+wB0LrMzned29y3R7\nA7ar7djW0O70ubZR577T87znDae5/8w8T53fYj3meObKeo2V+RzvfM1JHrpzkecv7R5ageeoYRZX\nyTfbtv0l4C8BP2Pb9pds2/5btm2/z/uWvwr8FvBl4D85jvPCrIs9KnBdl2+8sEE2neS9b7tDVGcu\n7MTeUPvY8+ukkhZvvH+FZa9nZCvmxO0Zj237jted5M4TZZ6/uBN7Q+1jIwns8cUCmVSCKxvxJm5X\nN+ps7rV47T3LvPv1pwD46jPxsl1bFXG4L83lWF3IU8qnY38ebmw3uLIhWOC5YoYTSwVanX7szf+P\nPisC5B/9YyIwfCrmpEk+f2eOlXjN3csAPB2zMUevP2Bjt8XJlSKpZIK32KtU6p3YTYTWthtYwN0n\n57j/7AJA7BfwFe/33XNqjne8WohBvvZsvPvz+lady+s1XnvPMneeKHPniRIXrldjraZf3qjR7vZ5\ni32MbDrJPafmaHX6sTILVzfrDFyXB84u8Pp7l0kmrEMxznn6/DYu4u564Mw8z1/ajb3wubXXIpVM\ncM/JOR44O8/L1yuxG8acu7xHrdnlTfevYp9d4PpWI/b34caW2J8nl4u88zVif8ZdBL7oFXLuPlnm\nrQ8ew0IUWuKETNyOeV4Fi2UR1+1U4kvcrm3VcYEzXg/wWzx54kvXKrGtodroUGl0Ob1SBODbvDP7\nuYuHa4B3VBA5cXMc53HHcd7jOM5djuPc7/33Bx3H+bj39T9yHOfbvP/9c3VLPnzc2G6wttPkNfcs\nkUknOb6Upz9wqcZ44LY6PS6v17jv9DyFXJqlebHBt2Pc4DBM3F5zzzJvvH+F/sDl+Zg31zdf3MSy\n4I0PrJJIWJxeLXFtsx6rPE5KXN54/wpvuH8Fy4JnYg7U5WUrk/ilcpbtSjvWavqNbVGVu/f0PACr\n3gW0sRtfIFBtdLi4VuXBOxZ48wOrpFMJno45cbu87iVuqyUWy1lOrxR54dJurEH6+k6Tgetywusv\nfP19KwA8f2k3tjWAYMSX5nKkUwnmCmmKuVTsstFrm3Xy2SSL5SyvvmsRyxJGPnFCytHe8qDoH7rn\n5Dz9gRurOkD275xZFQHR6ry3P/fi259yb5w9VqKQS/PA2QUu3KjGLkuTZ/br713h7V5g+FjMycJ2\npcXyXBbLsnjdvWJ/xn1WyWD49fct8yqvjynuAPn6doNsJslCKcPKfJ5yIc3adrxnxEXPmfuO42UW\ny1nuPTPPuSu7sbqeysRt1U/chIR2J0bG7Yq3P08fE2fEncdFr9XlGM8pWdg75Z1Tp73/vxHzM3FU\nYSxaIkD2dL3uHlFJX/KqInHS2es7YoOfXBYP9GExbuevV5gvZTi+mOc+L1h/+Xp8EqRmu8f5qxXu\nPjnnO2GdPVakP3C5EWNweMH7Nz9wdoF8NsVdJ8q8fK0Sq6XwVqVFPpsinxUN9ktzOdrdPo0YewVk\n4UA+jyteQWFzL77eS3nx3HVyjkw6iX3HAlc26rFWka9uiItHBsgP3rlIpzeINUi/7j3/J5dF4iaT\n6TjtndvdPjvVti9HsyyLkytFNnaasRVWev0Ba9sNTi0XsSyLdCrJsYU8VzfrsRY1vnluk8RIgC5N\nYs7HWMm+ui8gWvUMIDZi7I2WAeAdXkAo34frW/GysOeu7DFfzHB6tcib7l/B4uZeL93odPtUGl2W\nvLPy/jNif8rENi7Iz/7USpFX3S3MOOJkuwYDl7XtBieXCr5Bzcp8jq1KO1YV08W1KovlrG9q9ob7\nVnBdYh3vI+O6WxK3WnyxpTwjJOMm22AurccX18k1SMZNFh9vxHxGHFWYxC0CJMv0au+QW5oTmyvO\nwFBWHqRb31I5/jW0Oj22K20/ILrLc8N6OcZKtnNpl4Hr8qq7hu5PZ495FaIY5ZIX16qU8mn/oH3w\njkX6A5dzV+NjN2T1VmLRfy7jO/S3vMr9spewDRm3GBO3jZsvntdKqeLL8TGgVzZqZNNJfzzHKe8C\nirPv8ca2Jz9aEr+7lE9zaqXI+WsV+oN4kqYNLxA5vjgcU3JqucDAdWOrqK9tN+gPXP8zAPF51Fs9\n3wVWN/Zqbc5fq3D/mXm/wHS3l7DEeV7KgsKpZZm4xb8/L61XsaxhUCaT+rWd+NawV++wU21z14ky\nlmUxX/IYlsu7sTF/kkWRRS75fF6NWea/sdskmbBYKuc4s1rizhNlnnxpM7ZYYnOvSa/v+gUmgOX5\nPL3+ILbPYq/WZq/W8dklGCoUnnwxvv5LuQ9lwXOx5CVuMd7hUuYv92c+m+LYQp5La7XYCl3DxK3k\nr2GhlDGMmweTuIVEtzfg+Us7nFwu+JWyIdsV3+aSl9yJJXHpZdJJyoV0vGvYlqyfOHALuTQnlgpc\nuFGJrVIm5YivvmtoVStZjrgql41Wl829FnceL/kVQ2md68QkS2u0ejTbff+ZhMNJ5rf2yTVXDkEq\neXmfHOy194rELa4+t15/wPWtBqdXiyS85+GEF5ze2I4vOJWM84mRgOi+03O0u32urMdTuZTn1LHF\n4RqkSuB6TIy4HwSMzG2TQXJcvXZPvrSFi5BSSxxfzFPMpWJl3K5t1lmey/msvEzcNmPanwPX5fJ6\njRNLBTJpMbfVr6bHGJRdvCHe81Hr/Tc/sIqLkN7HgU3vrJSF33w2xdJcNnbH1Y3dJsvzOX9szHe9\n8TSuC1/+1vVYfr/83E8sDwsrK3NSqRHPc3nRY4FHn4dTywVWF3I8/fJ2bOqAjd0Wi+Wsvzdk8TVO\nxu3Kxs1nBMDZ4yVqzW5sks1rG3UsuCmZP7FUYKvSpt09vMHoRwUmcQuJF6/u0ekOfLYNRpmN+ILT\ntX2MGwhp3HalFVtV5PpIQ7HE3SfnaLb7sVXTn72wTTad9GVgMKT2r8SUuF3aJ/0BuO/0PAnLiq3f\nb39/GxyOhHer0iKZsJgvCbmJrBzGWdG/ulEjmbD8vXF8Mc/KfI5nL8RzAd/YEgyPTBxhuE/jdOa6\nvt0gmbD8zwDgvtPCHCQuuaT8944ybvK8uBaT7EWyTKdHGbeY1/BNzzXxDSOJm2VZ3H1yjvXdZix9\nNLVml716x+8XAXFnWMS3Pzf3WrQ6/ZvOyuNeUh9nT5N0FL3rxHCm4Zu82WVxyQS39249s0+tFNmt\ndWjEZPDV6vSoNLp+Ag/wtoeOkc8m+aMnr8XCzPuS7qVRxi1eib3sbxtl3CzL4vX3rdDq9HEu6y/A\n9voDtqstVkfO63I+TSppsVONJ7asNDpU6p2b7i6AO6RcMgapv+u6XN2ss7qY9xNYGCb2cfc+HkWY\nxC0kfDOOkcTtMALktTFB2fJcjm5vQDUmB79r+3poQDgyQTzyn+1Ki+tbDew7Fkglh49yIZdmeS4b\nG+N2aaSpWSKfTXH3yTIvX6/GMo9ma1/1dvS/4zr0QUglF8tZn2nKppPMlzKxBYaDgTj0T3lOiiAu\n4Nfeu0yz3eelGBIW31FyhOFZKGfJpBKxsQquK3o8jy3mb9ob952Jt89tyLjdLJWE+Bg3v9F9n1Ry\n9Gu6ce7KLscW8jcxjzDs74rjvLy6TwIFkE4lWJzLshFTgCz722RxDaBcSJPPpmKVSkoHwVGGZXUh\nH+v4El+dMHKHy88mLtZNMq2jiVsuk+JtDx1np9rm5Wv6+5r29+LCsOC3FRPjdumGvMNLN/39671+\n1GdjMBrb2mvhujd/FpZlsVDKxhZb+r3Zx25+H2RsE0efW6XRpdbs3nROwTCxN3JJk7iFxgtXdklY\nFvbZoTRvvpghmbDYibnHbWUhTzIx/Aj9QD0mueSNcYybDERiOPClxMi+Y+GWr51eLbFX78RiQ3/R\nZ9xuPuzuOzPvS4N0YxzjtjgXr9Notzdgr965qZgAwrluu9KOpXq7sduk0x3cUjGUfW5Pndd/AY+T\n5iUsi2OLedZ2mrEEhZVGl0a758vQJI4vijERL16JR8K7viNGAYwmbkvzOTLpBNdjCk6vbtYpeD0S\nEieWC1jEk7gNBi71Vo+FcvaWr8nEIY4zYvhc3rw3Vufz7FTasbDRl73A746RwNCyLE4s5VnfacQ2\nY/DiWpW5YuamZwKEDXtchk77jZxgpM8tpr0xdDG8+cyWQXIcowlubNWxrJvl1HEnblc26xRzKb9H\nXUKen3HEVPsdJSWWylkqtU4s+9Pvb9vPuMnELQbG7Zq3hlP7Ejcp+Y/TdO6owiRuIVGtdygX0mQz\nQwo3kbBYLMdXFak1u9RbPb9vRkIyf3E5S45a+ErccaxEMmFxPoYK8q6n+16Zz9/yNWlzHYd89dJ6\nlUw64Ut+JBa8xuI4ksfRGW4SsrE5LgnvdvXW5BFEUDBw3VgSyHFsFwyTe8mO6oS8gI/v25/Hlwq0\nO/3YgiG4uagC+CZCW5V2LEzw2k6Txbks6dTIeWlZnFgqcGO7ob0XttsbsLYj+g1l/ykIJnh5PhfL\nWAKZBBRzqVu+JgPFOJ6J/Q3/EisLOVziCZLX9rkhSxxfLNDru7GcVZVGh+3K0JhkFEsxFrvkPT2a\nLMjP5tpGzInbvju06BnoxDGT9fp2g9WFPOnUMBwdSiX1Pw+u67JTabEyn7/leZCS/90Yesz8z2Lf\nvbFQzuJCLEYtkoG9NZbJUMqnYzHOkaNJ9hcd5Z+vG8bNJG5hUW/1/ENtFEvlLLvVeKqW4/rbYHjY\nxZG4jbPwBUinkpxaKXJ1U78DkQx25ouZW762FJOjYqfb5/pmg7PHSn5zt4R0j4sjcRvHuMm5WXEV\nFGTgt7QvcZOJ9WYMcknJXJzel7jlsylymSS7Nf2X39ae6POTibuEvHji0OivjXFzlJCGMboDdX8U\nwMKtazi1XKTTG2gP1LerQn50bMz7cGqlSCUGVl4Gv+PuDWk9HkdQdn1TNPyPmtVAvM6S8t85t+/M\nlnfZjRh6QH2Z5Ii0XSJOefl2pcVcIX1TH8+pFfE+xMe43SqVBCFfBf13V7vbp9ro3tTXBUKuWcqn\nY0nc6q0end7gpjYDiVQyQbkane9LAAAgAElEQVSQZieGe2P9QMbNG8Idwz0uP+/yvrPKsiyW5rKx\n3J91bw2lfWtYnsuRSiYM44ZJ3EJh4LrUW92xldOlOVG1jKMy47sw7Uvc4hxLMM7CV2JlPkenq7/X\nblLi5hvGaL6A5ZDjs/sSBYg3cduqtLAsWCjf/F4szuXYqcYzhHtczwaMBIYxXMLDGTTFW742X8yw\nV9e/Pzf3WizP5W5J5OO0PZcB8jh5ngySdPc1yURgf18XDCvZuu34pcRpsZy75Wtx9bnVmwczbjKB\niSNxqzS6FHIpsiOJAsS7P/fqHYq51E3sCgwLDGsxuK5K06o7xiRui+V4Cn6u67JVad9S5MplUqzM\n5+JL3PbGSyVlkaGmeX8OE/lbz6nl+RxbMZitbY9hPkexUMrGEtfJJHV/ErsQ4xDuSUWm+WKWdrev\nXalRO2ANiYSQVN/YbsQ6f/MowiRuIdBq93BdKOZufajjnJklZ0HdwrjFOJZg2FB8a4C8GlNF/6Dq\nLcQnG5UH+rhDv1SI5/IDcfkslrM39TyCYILjMqzZP8NNIs4hv5t7LVLJxNjPY76UpdboamXFO10h\nhdz/HkC8tucVz6VQMjqjiEuGNJxTNWZvxFTUkGtYGvM8HIvpnJIOgePujVQyQTGXimWeXK3RoTTm\neZAyubgYt3Hn9fEY2Whpw78/WYFRozG9z0TVO4f2y8phyATHIVPc2G1SzKUo7Hs249qflQnF15V5\nYbZW0V3cqR58h4MoMrU7+hOWar2DBZT37VF/rE8ciVuzSzJhkcskb/maLLbpLjJJxm1coev4YoF2\nN552g6MMk7iFQK3lVU7ztz5Q8gDWfeDDUIe8X4I0F6NJykGsHwwDQ92BwF6tQyadGHvIxGXUIqUD\n86VbD30pN6g29R4y/cGAnWrHDzpGISu6cTRX75/htn8NcTDBO5UWS+XsLb0KIHT6LnpZHvke7Ddo\ngXhtz+W/UUqeRhHX7C5pcV8eE5TJJKauOTCU5/G4oExWsnVX0w+qIEvMFTPaAxHXFQYppTF3V1xK\njV5/QK3ZHRuky70Rh1RSnoXjiitxSeyHLsC3rkFKrHUHyAPXZXOv5UunR1GOOXEbl8wv+7Pc9MYR\nfnFnzGcBw89D9x6tNrsU8+lblBr+ORWTVLKYT4+9P+W+1X1e+gqFMeelVK2sx+hAexRhErcQGFYC\nxvW4xRcg7x1w2CUsi4VSJhZaf89LWMYddrKCq7uSvVdvM1/MHBCkZ7Es/cGIlN7tdycDKOXF3+lm\n3LYrbQauy8q4CnJMklG5DriVYVksZ7HQHwzJ6uy4XgUQUg/Qe/HIZ35c4iZtz+O4dPykaSLjpncd\nlbp47sexfkPzA71V7EnVdGneo3u47SSpJIjPqN7sanVdbbb79AcupTF3l7xHdLN+8vXHBemFXIpy\nIc1GDHtjq9Iim0lSyB5sFqNblibPgHHnxFxRfEa6E7e9Wodub3BLTxWInuCEZWlP3PYakxk30B9H\nbE9g5WGYuOlOnGrN7i19XaPriuMOP2gNEF8C68s1x5yXqyZxA0ziFgqN1sGVgLgqdSBkUIXsrX0C\nIPXYHe0a4KEU69b3wh+6rPHAHbgu1UbXD8b3I5VMxDL/RCaw49aRzyZJJvRffn4v0ZgLeDGmS0eu\nY66YuclBEMRnMVfKaJetSkOBgyunIjjY09hgveknbrd+FpZlsTyXi8U8qFLvkj/gjCjn02TTSe1S\nSck0j03cvEs5NqnkWGZDVpD1BiKTpJIgEhndTHDN+yxKY85rKdfUnShMYldAnF+bey3tIwG2J7Dy\n6VSSciGtveAnTZT2z8sCKOfjSaSHhla33l2WZVHKp/QzbrXpjJv2e2NKj9tiDM6SA9cVSdOY/Tlf\nypBKWtoVTIOBS6PVo3RAgUkm1zrvT4Bas0c+m7ql7QPguBfjrMc0F/aowiRuISArAeMe7MUYqyKV\nemes/AiEZK8/cGOTOBxWRV9UqN0DgwAQlaqdaltrILArNfpjGDdx+aVjS9zGVU79w1ZzUNbt9dna\na42VzoJgpHeqba3279v+SISDexUAdjUalGwe0OcnsTSXpdXp+0UgXag2OmOLKiCey5X5nP7ErX6w\nXFNWdfVLJdukU4mx1dtyIUPCsvRLfyZI7AHmY3CWrHms30HV9HIh4xfjdEGqE8axKyCq6f2B3pEA\nrU6Peqs3trdMYqmcY1uzoZNM3M6OS9w8xq2q+fMYstHj34tiDHeXZNzG9qnHNJphe2qPm1Rq6Ps8\nGi3hn7DfzREgmRCjhq5v6TXlaLR7uBws6Y7j/gQONACEUcbtle0saRK3EJBBxv5GXhAXYippaa/e\nDgYutUaX+QOCsriqyJVGh2w6edM8O4l81rPy1dhDM8lRUmJxLkd/4GpNWiq1NpY1nlUAYkrcxls6\nQ3yudTe2m7jAqTEuoyASlv7ApapxHbJoMq7XD4YXsF7G7WAJFIwEIxoLPJKNHldUkViZz9Fs97Qa\nIFQmyDX9HjfNBgw71baQ6o5hVxIJi/lSRjsbPUliD/FI42oHWGwP15Ch1tAr1zxI4i9xLIZq+vaY\neZf7sTQnDJ10ntuX1qsszWXHfh5yv8RlynGQRLCcT1NvdbUWPiexsItz8chWd6ptyoX0LUoRCV8q\nqbHAM21/nlwu0Or0tb4X/jl1UOImYwnNsWW91T1wDUvlXCzs41GHSdxCYJI5iWVZsdjGVptdXMY3\n/EM8hwzgBYbjNxfot/INkrgtx9DftVvviMp94tbAEMRBXG/1tAZEExm3mHTp172BzyfGuIzCqHmP\nvudyKuPmSz309riNm+EmsRSD3Xij1WPguhP353C2nr69UW10SSUt8tlbAyJ5hkomSAd6/QGVeufA\nwBTwe4J1VrJ9xu2AKvJcDKy4L5U8KHErpHHR+3lMchCE4fmlN3E7WB4oobvPrVLvsFfrcMexW8cR\nwLAIqJ1xm+CIDCKAd93hAHkdqNQ7WNZ4pqmcT5NKJrQysK7rsl1tHfgewGgxXGPi5iXp46SSMHTv\nvq5xhtm05NHvEdd4TnV7fTrdwYFyzUTCYmU+b3rcDnsBtxOmVU4Xyln2ah2tFarqlKplHI20ruse\naOsssSqtfDVtcl8bP0aiKBGHYcxercPChPdBHsR1jQHRxm6TVDIxVq5ZyKVIJiztjJu8UA5k3LyL\nUWej+bDJ/PDcwTb3WizNZQ9M5OMwi5nWSwT4RjY65czVhihqjGO7sukkqaSllXGTZ+C4WXYSC6Us\nvb6r1SSl3upiWZAbY4YBI6y4xkB9mlQyDmZ+WGwb/3lIxzidBiXbUxwER7+mq7gySSYJQ2nxYdvg\nxzESoDKh8GlZlt/uoAuNdo9Od3DgnQFib1joVTHJfmDZ37gfJ73B7Ne29M33m5a4ZTNJcpmkVsVK\nfYKPhMSxxTz1Vi+W+bhHFSZxC4FJwwlBGEEMXL3SPN+F6QAZlBzArLM61Gz3RH/ZRCmW3qGuQRg3\n3TbXrU6Pdrc/dhSAxHAkgL5DZmO3yepCjsQ4OZhlxWI3Lhm3cXP9YFQiqJNxk+YkB1SQcymtcmY5\nw22cMYmEPydKYzFhkqOkxEoMs9wmsfKWZVHMpbX2uE1jFCCeQle91aOYS4/dnzDi6ljXaU4yRSpZ\n0J88BjEnAb2MmywcTUzcNPerX1qvAgcnbn7CFEOPm2Ud/HnEMYd0r96ZGEcszWWp1DvaZm/Kou7i\nBAY2lUxQLma0KjV8xu2A/XnKu1dvxMC4HaQMAKHg2dPY41abQo7A8Jx4JcslTeIWApI1OYjGXYxh\nLpBvCjJVKqnxAp4wI0pCt0HJsNH94AN3SfNAct9RcgLrN2Tc9Fx+jVaXeqs3ViYpMVfIUKnrdRq9\nttkgk04ceAEuxjArarvSJpdJkj+A2bAsi/liRtvF48+xO6C/DUbnC+pNmGC846uEL5XUlLi1O33a\n3f7EoEy3+cHOFAYW4pFB1VtdCpOCoUIcUsnpPW6gl3EbGlodvIZsOqmXcQshldTGuK0Jxu2O4+MT\nN9/lUzvj1mKumCGVHB8G6mbcOt0+rU6f+eLB59RiOYuLPtnqtFEAEgvFjFa3bn9/HrA3ji8VsBgW\nSHWgPuWMAFEor3nD43Wu4SAjJzCz3MAkbqEwTfISR/XWl0oexLjF0OMWRIq1KqVYmnpoAjFumiun\n8j2etIaSZmtn35hkAsszX8rQ6Q1odfpa1jAYuKztNDi5VDyQVYijx22n2mJpLjdWmicxXxJyZh0X\nsLx8JyUr0sFN5/tQmeDU5q9jTu9ZNWT9Dg4CSrmU34+nAzLwnsi4yX4mTeel67rUm72JFeRyLFLJ\nyUGZZGd1mgft1Tueidf4sMOyLFYXcqzvNrUFyNtTnBRB/wDsy+s1spnk2MHXEuVCRmuPm+u67FQn\n93/qTtyGccT04qu+xG3yKACJhXKWtpdo6oBU5Izr9QMhLV+ez3FNJ+PWmiynBlHo0jm6ZNgPPFkq\nCa9sZ0mTuIXANMmLlCnqHOg6tM8d/2ALOVhCa+JWbUxOHmGkv0zTgbs3Yf6LRNnTpusKRmTyeJAR\nBQwPYmkOoBpDY5LJGn3QF4hsVlp0ewNfh3/QGpIJSxvT1O70qbd6Uyun88WMtnEZ8lLPjXFalUin\nEsxpnhM1aVSHhBw+rMt4wA9EpjBuLkJ6rQPTenhgZM6hLvlsb0CvP5hYQZbuvFpdJb0ze5pj3J5m\nqeSk8xqEQUmr09cmLd+qtLxZkweHPpId1dF/6bqyyFU4MI4AwZbXGvocHWtNwZpMurt0J26Thm9L\nDNlPTQXg2vQ7XHxdLzM/jXED0YZQqXe09QVPc5WEYSyh630IwvodWxRxxit5lptJ3EKg3jx4vgQM\ngwCdzbTT2C7hbpmJRyo5QeIgN7+uQ6ZS71DMjR8wLJGwLPLZFHVNgeFw+Pbk4BT0XX6THCUldM9y\nu77p9bcdMMMNxGexWM5qG6bqjwKYIIECvSMBgiRuIMZU6JwTFUQqmUomyKaT2ubJTZPFwbCqqmtv\n+APZD7HHrRGgggxCLql7jls+mzyQ7dI9kqDbG1Bv9SaelaDXoMR1XbYr7YkySRgmbjr2Rrc3oNd3\nJwbHIAoewuVT196YLiOOj3GbXgDWpVAIuj99RloT0yR73A5i3ECMBAB9zpLTWHnQb/AVhHGTZ4ju\nAfVHGSZxCwjXdSfOl4CRHjedUsnGdDmWbnfLaXJNGDa46goM9wJUb0FcwrrWIAdRTupxk4GrPqnk\n9MRNN+MmL5KDjEkkluZy7NX0NJr7owAmBCKA7wCqY4hoyysQ5DIHF3dAJBI650RNmp82ikIuRaOt\nZw1BzqnhEG59yaPF5PdhQXNfcj1Awz+IPVptdLXJRsVQ2wlyTc2BaTWAfBf0GpRUvd6caWdEMpEg\nl0lqcRqV7PJBfbgSZT841XNm+2z0hCS2pFktMpzrd/BzOewJ1rM/mx3vzB4zsmQUuhUKtWbXLzQf\nBJm4rW3rTdyCJE26jFqGBoATFApeYVSXbPV2gEncAqLT9SplEx5qWY3QKpWsd0inEhOr+gueu6Wu\nQ78SQCqZSYvqrg7GbeC61JvdidUpiWIurS959M1Jpl9+usxJgthb62bcpIxlUvII4hJ20RMky9ec\nZP0OIw5+GgLUoIybbrvxqpewTKqcgghGdO0Nv8dtIhutT5IG0Gj3yWWTB45mgFGnUc2ByJSK/lwx\nw8DVI+F1vYHsk56HXCZJJpXQdkYE6UkGvT1NNV++G/De0FDUkIF/YUqiIO82XQYlvuNqIKmkXlZ+\nYi+u5j71oIl03iu8NHWdl80upUJ6Yn+2bvO7erNLNp2cqGKSRWpdIwGmjdwCod7JZpJ+ofSVCJO4\nBUSQSkAmnaSYS+md99EQ9rmTNrjPKmhax1AqOfkSLuZTWqqWrXYfl+mHLQhWod3ta2F5/B63ieYk\nescBVBsdUsnE2CHHEroTtyBmGAALRX0yiyDFBBg+Mzr6qloBq7e6Z7lVGiIImJSwgAhGGu2eFslm\nJYA5iW6pZLPdm3pGCGl5VttZOW34toROVlwUHQcT+2csy9JqiBHErAb0vg9yzx9kLjaKYk7P3dUI\nmCjM6WbcAhj3FHNpLPRLJSc5Q5fyadKphLYilwz+81NUEtoZt0ZnaiFat2N4vdWlNCG+hdHh8Jok\nowH67ADymaRh3AymIwiNDKLir6vHbTj4evoaQF9lpuJX9Cdvcl1sl18lmxIMgd5+hb1am3w2RSZ9\ncKCey4hBwzovv7ni5ErdMBjS9zzA9KCsoLFqWQswogJ0J26ScZsmldTMuDUmz0aSKGRTuK4eyUk4\nqeThJW4gChu6nEaDNPzDsB9Rh5FSkN4VEOeErrEhgddQ0JewBGVXwCv4ddQX/AJLJTVL7IPMOEwk\nLAq5lEZJ9/Re3OEQbj1Frka7TyqZmMgywUgcoeHe6A8GNFq9qXtDt2N4rdmbek4NC9GHW+jKZVK+\nzPWVCJO4BUTQB2qxlKXZ7tHWEAw12z16/cmDr2HogKRLslltdCjm0yQT0w+7ekt934bcsEEu4KLG\nA3e70p7qYmhZFqV8Wksg4roulUZ3ai/TvJ+4aXIHm2LzLaHz8qsGTNx0Vk6DSyX1MW69vjCBCCYH\n01fUCMKAFn33PvW/33Vdmp3e1Eo6iGBk4Lpakvkgzfag19XRT5qmyTULaXp9l2Zb/d0l+xinm0Do\nkwg2vf1ZCHRviHWoPifkezs9cYunx22atLys0TRnmtOpxGI5S6XRpdvTE1NNk63CSMFPw1lVb/Vw\nmewoiff1ZEKPrLvbG9Du9qcmj7oLCvVml2zmYBMliXw2qeWcul1gEreACFo51TkXyK9QTesT8Cv6\nmqRYAY1BSrm0qOgr3mDNgPIGgEJOj7tls92j0e5NbO6WmCtktAQirU6fbm8wtW8knxV9PLoGTwd9\nHgoaA/VqQEMOKSnVcej7UslpiZsclaGBcav7fTxBGDc9wSmIiz2TSviN5OOg03G13e3jusGKOzJg\n0rEOee5MGsANw89LR3EliNU4jDDzOpPHKfdnJp0kl0lqYR6HbNf0QN0/qxQ/E02/x22aJE1vj9tu\nrU0hmyI7QS0CoshUa3bpdPWw8nJ80STIeXebe+rjmWanF0g6q7PgF8RREkRv15ynDlC+hoCKsnRK\njC7RxcLWW11KAZRUuUyKXn+gbRD4Ucf0d+gA2Lb9S8A7ABf4Gcdxvj7ytQvAZUDu9j/rOM7V6Ms8\nfMgLeFrV0h8JUGlxYoI9ehQEaeYFWJ4XgeGWhoNOVvTPHitN/d7ReTjTApcwCHUBZ/WwCtIUZHmC\nKYjEXCnDpfUarU5vqowuDIL2jViWxXxRT+U01PMgEwUNZhTVZpdU0pqaNMlAXkdjc1Cp5EI5g2Xp\nKaxUAwbpMJQa6/g8KvXO1OTRl0pq+P1DZmP6GVHOe+xGs8uxRbXrCCNTBD29XVHWoPruGvaIT38u\nRaFL/fsg74BgSg15Vqk9J+TrTUvcdA9ED1psG7XjV/1MVJuTDXMkpOnVxm5rqnNxWDTbvakz3GBY\nANaSuIU4sxdKWS6vV3Fdd2J7RFgEmZ8mUdakIAIxBPz4FJMzGBZHW50+pfwrj3+K9C+2bfvdwP2O\n43wb8AHgX435tj/uOM57vP/d1kkbDPtR5iZYv8MwadJRHfID9QDSAsvSk7gNJWnTD31dl58MyoJJ\nXvQ418lB0tOkkjA05VAdlMnqfJB+prlihj0N/StBiwmgV5pXbQi55rTLLH8EpJLJRIKFUlbLbKJ6\nQFkc6Ksi9wcD9mqdqWy0PB90jANoBGQ2YMTdUkMVeeg8G1TOfDQSt8NaA4j5oNVGV/lZFZTtglHH\nU9V3V7DksZRPk7AsLWNL+oMBtUY3WOImZd2Ki0wD16UWQOYPsLogYqoNxSMiev0Bne4g0PMg5ZRN\nHUXHhtwb09+LhVKGXt9V/lzuBTQYA1EorjXV789ur0+70w9U3JHF0Veqs2TUVPW7gf8K4DjOc8Ci\nbdtzylZ1BHHVGzJ8emVyxWfVS9w2dCRNASszqaQIDHUMO5YzRFYWpjNNMlCvaZApQjB3sIKm5FG+\nt5Ns+CV8C13ViVtAeSDAYjlHr+8qX8NeAGcwCb1SyWBBQDaTxEKfq2QqmZgq/QG8hvu28v5PPwgI\nUL3VZdyzV+swcN2pbHQmndBm3BPGiELnDLO9eodU0poujdMoUxyaBwVLHnX00ATtEQdRiOoPXA39\nZWHMSfSoA4KuIZGwWJrLsrmr/g6vNUVPVZAgXe5h1bFEs91j4LohGTe1iVvQQhsIiWAqmdDEuAUr\nyIM+g5Khw2eQxE0kj6pNrdZ2ps+klfBbHl6hzpJRdVsngMdH/rzh/V1l5O8+ZNv2XcDDwN92HGdi\nhLK4WCCVmr6BDgOrq2WubzcoFzLcd9fyxKp+3zPsqLV6rK6W1S7Ee+3TJ+anvvaJ5SLOpR2Wlook\nAwSSQfGV59cBePW9q1PXcHxVyOdSmZTS9yLhPScnjpWnvu6p4+IwsJJJpWto9cXjfPfZxQBrEF93\nFa/B9Z6HMyfnpr7ufXcs8o0XNmh0Xe5XuIYLG6KgcfJYaeoaUl4w1Ael70O316fV6bM8nw/0uoVc\nim7fVb4/u32XQi7Ys35itcRL1yqkc5lAyX9gvLgFwKnj05+Jk8fE1xMptc+lPAPPHJ++PxfnclSb\nXeWfxSVvKPzKUjHA/vRqjsmE8nXUWl0W53IcOza5rum6LpmUCAyVP5fezXvXmYWJr31XteN/v+o1\ntHsDMukkp08tTP3eY8tFOLdJKptWto7V1TID794+fXKe1SmSu5OrevaGXMOZU9PXcGq1xLde3GRu\noTC1Fy0Mat09AI4vT98bd58V2uF2T+152dmoAbC6VJj6uhmPiaooPif6W+LuWgx4b5QKaTq9gdI1\nrK6W/Xv8dIB7/LSmWKLPGgBnTk6PLVcWC8AWmXxm6jMcBi9cqwJw/53TY6rlRSHbzRUy6uPs2wCq\nGm72ZzI/B3wG2EYwc38K+J1JL7Czo2ca/KxYXS1z5douNzbrPHB2gc3N2uQfGAxIWBZX1qpsbFSV\nrmXN+939Tm/qa88X0gwGLude3vLlmyrw/HkRGM7nU1PXMPBcoK6vq30vNnfEgdttdae+btcborqx\nVVe6his3RI0i6Q6mvm7SY1UuX9tj46S6QOTamlcn6fenrmGhILb6My9tcGpR3fNw+ZoIApK40z+L\nnmgk3tlrKv0spIwnm04Eel1hftBWvj/rTWHIEeR1i16V99zLW9xzSp1Y4fq6+N1ub/oZ0fP2xrrC\nvbG6Wub8pR0AcgHei/lihvNXK6ytVabOnQuDG977MOhN3xuDrqii39ioKX0mXNdlp9LijuPlQK9b\nLmTYrrS03Ru99uTz0vXeh+vrat8HgN1Ki2Ju+p0BkPbqjBev7JJV8Eisror3f2dPFPGa9TYbg8mm\nBr2u2Btrm2rfizBrmPdY8+df3ODUFKVPGFy8sguI93n63SXWePlGRen7cOnKnrcGa+rruq5LNpNU\nHlNdXROvZbnT7y6AXFqY5qg8Kzc2qtzYCB7Xpbz9cPHKLmeXpjNTQXFNvhcBYgm5Py9c2SE55RkO\ng+df3gSgnE0Gjy3XqqxOkaHfrpiUkEalYq4hGDaJU8B1+QfHcT7iOM664zg94FPAayP+niOB61t1\nXODM6nQDhmQiwdJclo09tbQ+DN2HgsigljRJHC6v10glLU4EODSKupy5WsEsleFmgxSVkD2PQXrc\nhgOwFcsbAszKkji9Ip7da57kV90agkss0qmEYBUUS/OqAV25JPLZlDZXyaDmM3J/qu4dGTqUBXH5\n1CMjlv+mIMY9S+UsA1e9hDeMgVFJk7tlvSXGtwTZG6Bvjlq10SVhWVN7R6SkW0dfVb3Vm+pYJzF0\n2NTzTASRxvn9l5p63IKsYeimqDaWqIToZ9J1Tsl+/SBSScuyOLaQZ2O3pXRvhJHOgoglGu2e8v0p\nz53DlErK2CTIWeWfl4ql5Tc8lUQQAxq/x+0VOsstauL2OeCHAWzbfhNwzXGcqvfnedu2P2vbtnwC\n3g08PfNKDxFX1r3+ttVgVa+V+Rx7tY5yC92g83hAj7PkYOBydbPOqZXi1BluoNGZyx8HENzWWUdw\nWi6kSQeQ9857h61qG9+gvSsAJ5byWJb6xM03XwjQ4wbCyVD1Z+FffAEKGuAlbh21F7DrCs1/LkCi\nAMOEX7VBieyXkOYKk+C7SrbVXsDD/s/pz8RiWc9Mu6DzskBf4rYXom9Efp+Yo6Z2fwiHT2F2MQm5\nTIpsJklF8TnVHwxotnuUAjyTMCxEqe73a7b7Xl/l9LtLV8EvzBr8fnnFfW6+oVSQnuB0klI+rfyc\nqoY8s1cX8rS7faV9qGGcZ8X3pej1XV85ogrhXCW9IrCmWCKYOYmenuDrWw3SqUSggt+oq+QrEZES\nN8dxHgEet237EYSj5E/btv2XbNt+n+M4ewiW7au2bf93RP/bRJnkUcdVT2oShHGDYaVMNdtVa3ZJ\nJqxAB418+DcVrmFtp0G3N+BswPdB3+XnJW4BGt2lKYDKNbiuy3a17VslT8OQcVN72AYdBwCiufrY\nYoFrm3WlCUuY6i2IZF51g7dfvQ0QiIC4gF1X7aHf6Q1w3WCVdNDIuHkOjYEYN12jMiQbHYhx0zPT\nLoqrpOoK8p5XFZ8PYDcOw32sw8QoSHEHxFm1q/j3+8YkARlxf4aZBsYtKLuizxE5+Bq0MW4hgnQQ\nBZitilq2K8zdBXqcJZudkIybHMKt/P7qkkomAvUx6mPcOuSzKTIB1lDWUOgauC7Xt+scXywEkszr\nHOtzOyByj5vjOP/nvr96cuRr/xL4l1Ff+6jhimfAEFRnPlopUzl3pNbsUgxgeQ56GDf5PpwJMLML\nhhe1armJpMeDDOBOJcUQYJXJQq3ZpdsbBGIUQATzmVRCQ0AWbICpxOmVIt94ocFevRNodk2gNdTD\nXcCFbIobWw2lc2iiSCiRL/cAACAASURBVCUhXBA1DUFnuEn4NtsaGLd0KkEmHYBV0Ja4tcimk4Ec\nBBc1MY9hZFDJRIJiLqXc/TYs4zZqx6/q3uh0hXHPfDHY3lgoZji3s0d/MAikqgiCesDhvhLlop6K\nfrPTCyTNg9GCn3q1SOBkRdNoobDFtuW5HJfWatRbwd+/aRiqJIKtYdRZ8t7T80rWEGY8BIyod9q9\nwMWYIKg1BSMe5D4sFdIkE5aWxC3o8zBk3NTFMzuVNp3ugJPLwWYFygLpK9VV8pU3uS4Crm7UWJrL\nBh4iratSVmt2Aweny15gqJL1u+w1/AdN3PzLT7EEqdHukU0nA5sZFLJq5XlhGAUQGn3Zv6ISQYeo\nSsjCg0q55F69Eyp5LORSDFy1VsLVZrjk0U/cFK5BFhNyAd3f5ooZkgnLnweoCtVGN9A8OxCW4/ms\n2qIGiDNnaS4baA2LmuZEhe1fKebTGhi3iImbwnX4I0OCrqGUxUVt0iTn9AUN+vVJJYMXasTeSCkd\nB+C6bqg1zBUzZFIJ5SMBhvM/g30e/hBuhXt0OLssuFQSFDNufr9hOMZNh9Q/6PuQsCzmSxl2FUol\ne30x1y/oOSUlnVWFcd31bc+dOmDi5jNupsfNYBxqzS67tY5v7hAEq/Ne4qbwwO0PBjRavcByk1wm\nRTGXUnrYyl6/oFLJVDJBLpPUJDcJbo9czKWUVk63Q/TwSCyUslTqHWVzu/r9AfVmsNllEqdWxKF4\nVWHiVql3QlUfdfQchhkKDyMzYBQmLK128JlAIC7gxbL6IdxhggBQX9RotXvUW73ARQ1fKnmIjBsI\ntlb1UFm/4T/g/tAxhLsawsAIBOMGantoJLsSpO8SRDBvAVWF70O316fXd0Mx7KrvjU5vQH8QfA2W\nZbE8n9MilcykE8HVAfPqi8Bh+5KHiZu6Ncget6CMm/zcVBa6ev0BzXY/1Jk9X8yyW2srO6uqjS4u\nwQtM8jNTWei6vhncmARGGDcNJmO3A0ziNgUVj5JeCGE5KodTq3SWrLfE0MygjBsIueTWnjpt+sZe\nk3w2GYrlUX35gdisYS7gQi4tBn4O1LwPMtgO2uMG4lDsD1xl7GOl3gk8RFVCtbNkrz+g3uoFrtwC\nFLNe74jCy2+YuIWTQSlN3CTjFqKgsFQWF3BfkaVytzeg1QkXBOSzansOZUV8OWBRY14yj9oSt2Cf\nRzGfpj9QywSHlkoW1PfC7oXsZ/KdJRVKsWR/cRBTLRBsV6mQVso8NqQRRcDCCnguggrvrrDSPBAJ\nS73VU7qOSqMTOJGHYb/8tsI+1GqjE7ivS6xBnCc7Cp9Ln3ELeEbIoqPKeyNsAiu/V+VZFWb4NogE\nNmFZvtJFBa5vy8QtqFTSMG4GEyAvnTCJwnwxQ1qxxMHvEwiTuM3l6PQGyhInIdUMNzOjkEsrNwZp\ntnuhLj/Za9NUtMllJT1MMj9XUhuUycAqTNIkG7xVXcBhm9xhxMlQ4TNRbXSwrOA9NPLQV5u4hetx\nAyG1dV3Yrap5JqIEAYVcSmlRQyZuQRm3RMJioZTR4iqZTScD92nJgphK+Y9krYLuj7mi+t6RaggH\nQRi6w6pMHqPcXXOFjNL3ISwDC+I8aXf79PpqCitR1rDi97mpKQK7rku1EU5ir0sqGbSvC4S5Vipp\nKT2zZTwQuMdNg1SyFlIyCqMyQTWJm19gChjPJCxRWFEpp5bPlmRWp8G4ShpMhLx0gva3gSdxmMsp\nlRaEZRVgeBiouABdV7BFYS5fEElTq6Pu8ut6cpNcGMZNcaN5LWTPBow4SyqSIO35iVvwCziTTmIB\nbUVjKvwm9xBrKGqSShZz6VA9j6CW9RsmbsEr+iXfvEfNBSjPqjDPpeqixqZM3EKw0YvlHLvVjrLk\nEUSQHIb9lH0bKvtxZf9nOhXsmp3TIJUcGlEENCfx7cbVMRvS9CXMc1kupL05eIeXNA1dkdXsjTBO\npxIr82olgo22mC0Y5sz294bKYluIfn0JMX9TYeLWCtnjllN/b1QjnNmqJZuyEB3mmSgr7glutHok\nLCvw/ZnNiFjmleoqaRK3KZCHdlj3uVJBME2qeprCOnPJNYifnf3hFpVHN7SrlG+rrGiDSUOJsFJJ\nUMfy1ELaz4P6IdwyuAsy+0UiYVlkMknaiqpUfjEhRPVWh1tbtdEJVdAYWglrMCcJkbjJxEJV1TBK\nEKC6iryxE04qCcJZUvUQ7kZIVn5Y5FLJuLVD938mE5baxK0erv9zrqherinvnyAuoxLybFVlOR4l\naSoqvjfCyndh2EetSro6VEmEY+VB3Znd7fVpd/qhzmwQLtIqk6Zmu4dFcKlkXgfjFtJdE4ZyX1VJ\nbCUk4waisCKKAGoKKw3PtyAoA5uwLLKZpHGVNBiPhmTcwiZuubSYFaVoc0WRQakcLOsPiQzYYC4h\nG9JVHXbDPoFw5iSgknGTSXQY+axaCVLUgkIunaSliHELO4sHRpJoRftiMHBptHqhqrcycNLDuAX/\nPFTr9GtRqreKGVCfCQ4jxZpTO4Q7rHsfjLCfihKFridRD9o3AiIYKRfSSpMmybgFXYccE6LDnCRc\nQUFtUaMZ4bxUfW+EGQovofq5jCJvV13c8R0lQyQrIN43lcW2ZqdPLpucOpheQkdvtF8EjiKVVMW4\n+e63wYtMqp/LRqsbStUG3vNgetwMxiFKjxuMDHVVnSyE2OCyGVxF4uZXTUMybjJQV7XBw1r4ijWo\nvXhqTVEdCmqBD+plULKgEPa5zGaStBUddmEd62D0s1D0PHSEaU+Y5zKv4QKOIpVUrdMfMsERGDfF\nBaZCCGXAouIh3N2Q7n0AJa93V1WPW9iGf4m5YkapDX7YOYulQpqEZbGrSBkAwzs0zB5V3YcaqcdN\ncXAatc8OUDZjsBLhzJYmIuruz3BzNyXy2STtbl+ZmVOz3QsZR2gw1pJFjQiKEVXrCDvXD4ZJt6rz\nstnuU8iGex5ymaRxlTQYD5mwhGXciooTliiHnR7GLbwuHdRdwLNIXlR9FvVWOMt1GDKVqqQ/MhgK\n+1zm0kllPW5R+i5V97g1IlTS9SRu4QsKqhO3qn9GhOk5VHxOeUFAGDZ6SfEQ7ihyatX7M6ybo8Rc\nMUOnO1BWSa42OuSzKdKp4GMq5opp5YxbNhOu0KWaVYiSNKm8P2F4VoW6u/JyFqpaWVzY57KQSynr\ncRsyblFjCUUsbEg5tU5zkiiKEWXmJLUOFuHucX8NCj6LXn9Au9sPzbjlMiljTmIwHo3IjJvagChK\n/0pRoflB1MRNdUW/FeUCLqi7gIUrV5TETV2/IUQLAgAymSStTl/JiAh/uG8Yxk1xj1sUi+2jw7ip\nDU6jOJTJy1pV5bTW7JJOJcgEtPkG9Wx0FDm131OliO2KIm0HmC+ofS8q9U4o51lQz/rVmz1KYWVQ\nso9HlVSyI2d2RZDYHwHGTV3SFP7MBrWjEaKYWoHac1vIqfuhDIwy6QQJy6LRVtcHG0nenlGbQO7W\nOxTz6XCFFcmIKygwRdkX4vuT9PoDZX12txNM4jYFvlQy5MUjLypVEod6BEr9aDBuahtpZQIYyjFO\n4fvQ6YqDIqxkVM4+Uc24Relxc10hJ5sVtQiMm5SbKHseIjFuGgZwR0jc8qqlkq0IM4GkBb2iRKHW\nCN+roNoGP0ogoDqBlcFd2MLKMImdfR2DgUu12Q3NrhRzaf+cU4FaBDdi6RqsinkczuyKwLgpdpUM\n81zmMkmSCUs52xW2oFD0HB1VmK35PVUhzDBAbeLW6Q4YuOHk1JZleeNT1LE8s7hKqkpgd6otX/UQ\nFCrVIlFUVGINasci3E4widsU+OMAIjNuiiQvzS6WdXhSjyizeECHvEFWTkMEZUcggbUsi2I+pU52\n04xWpcrKA1eBXLLa6JBMWKE+i1xW2PiqCkR8diVEspBMiJ4NlRewZM1CSSWz6qqWMEykw87LApQx\nLLVmJ5TzLYwkTYocHaMEyKVcGgt1Caw0wwjT6wcjiZuCz6PW6uK64dkVlaYc7W6fdrcfml3J+c55\navZoJIm9YtXM0Bk6XLJQzKXUxRGRGbc0LmrUAdJZOYwZBqhNWOSZmw9xXoN4flTOIK01umTS4RQK\neYX3RqPdo9MdBJ67ecsaVHwWEe5wUO+ueTvBJG5TMKzoB99YoFamCOLQL+XTgR2QxBrUST38hCVk\nIHIUpJLFI5C4yZ9RybiFsTKWyHkXhIqRANVGl1KIIaogemhUym6iVuvy2eShM27KzUk8mWI2RBCg\nMmkaeLMew17AhaywwVfGuLXCB2WJhDdUVrX9fET2UYVUMspcP1CbsERNFPKaGLfD7HHbrLSwLFgI\nyW4U82n1/WVh3aEVJvN7EY17CgqLwPVWxGQhp3YkQa3ZiWTSAmoSFmkItRiacVMn84/a9mEYN4MD\nUW91yaaTJBPh3iqVjo5ApL6qZCJBPquG5YkyRBX0mZOElZukkpaS4HSWxE1ewCrkJo1WN5SVsYRk\n3JQkbs1OKCMMiVJBXQ9NFKmk/H7V4wAy6UTgIeCg/uKpt7qhqvkwOr9s9s+j1e4zcIWsKgwsy0ua\nDpFxA8E+KnN9jRiMzCnscYt6Vg3nl82+P6JK8/weGoVGFHC4/WVbey2WyrlQvURyHfVmT0lfcrUp\nzoiw8YzKsSFSKhlWwptTmrCI0SNhE5ZCNqVURlxtdn1H26BQOYdUjmAJ+z6oNEjx7/Cw5iQaWh5u\nF5jEbQoanvV7WAwbm2d/qHr9AfVmN7TcBERlTUniFsH4AI6GO5hlWR7bdXjBEAxn+6k4aOqtcHOq\nJFRJJbu9Ac12+CGqAAvFDNVGV8nlF1lm4fVsqAiGQDADYWSSMMK4KbPi74V+LlPJBMVcSknS1Ihg\n+y5RzmeoKtifIAJkGM6HC7yGQpp6S81Q2agVfX8AtgqpZOTETV1/djWC1TiMBGUKZxxmUgnSqeAh\nj2SvVdyf3d6A3WqblflwkjQQn8fAdZUEybVGJ/T8NLEGdUlspd6hmEuF+ixArXpHOtguh5QIqpzl\n1u726XQHodlPleMA5PsQ9qzMKTQnid7jplaxcjvBJG5TUGt2Q/cpgFqp5F6tg0v4qghIed7sAWqt\n2SWVTJBJh6zUKZZKNiIGRKV8hpqCJHpWqeToa8yCRrMbKXFTJZWMMnxbQkqFVLAKUdmV+WKG/sBV\nav8etrCSVXjx9AcDmu1e6P4yEDI2FQxo1GRFrCFNs91XYpqzsdsE4NhCPuQaPGdJFfvTLyhE7HFT\nyLiFZWFVSiWlyUpUxk3VwOXdaju0RBFE4VPF+7BdaeECKwsREjdFn8fAdak1e5HObJWzUPfqndCJ\nPKiVzw6LOyF7u3Lq4hl/dEqEYls6lVDyPmz7UsmwPW7qelCjOEOPfr/KnsPbBSZxmwDXdWm0upEY\nN5VuULs1sbkWSuEvnmI+Ta8/oNOdLSCqNbuUQ/YzgXoHv2pT9HaF7bUTgeHs1fSjkLi5rkujHZVx\nUyPPizJ8W0L2NuwqmBUVtZdo1QvqN3Zbs6+h1aPV6YeuWiYsi2w6qVRuEuW5nCukqTW6DAazFXf8\nQcsRkkeVzpLru02SCSt0UKZSptj05yyGuzvK+TSWparHLdozoXLWYjWi9XvOl2LNvoZub0Cl0WUx\n4v2pouC3sSeKCavz4YoJMMp2zbaORku4QobtqRJrUJOw9PoDas1u6P42UMw0eVLJ5ZBntspZblHM\npCSE1H/2e2OnGk2dMJT5K+xxC1uMV1hou91gErcJ6PQG9AfhLGMlpBuUikN/x6Ozo1UM1SQLon8m\n/AGj2sGv2hAzR8L0EoG6quVsPW5qzGLa3T6DgRu6QgVDeUG7O9tzKWVtkRg3L4Daq80+cLkZscdN\nSpY2vYBqFuxE7BMA8XmouPx8diWk7AYE0+QyuzQu6gUMQ+dXFZLN9Z0mKwv50GdEuahuJECj3SOd\nSgQefC2RSFiU82k1iVtE6WpBpSwuojlJTqFjnN/PFDI4BZE0tbuzM8GbHsOzHEUqKYfDz/h5RDWK\nAShk1TwT8rmej5BE5xU6jW75PW6HJ5WUn0eUuCqfUWOuJWPLsEUNHeMAwt7hqs2DbieYxG0CojaY\nSxTzaSUyC8m4RZJKKjBJ6fVFP1NYLbaESge/WjO8SQuMBIazJrBHgHGTF1ckxk2ZVFJKoCIwbt78\nnl2FUsmwe3TIuKlI3Lw+gciJmzqXtEj9ZYpmuQ0Zt2hSSZidcWu0etSa3dAySbEGdTPtGq1e5HtD\n1QDsWXvcVDgIDpn58MZamXRCyQBuWZyJwriVFLU8bHrM/mqE59Jn3Ga8N6IaxYA6qWRUR0lQa3S2\nXWkzX8yE7rNTadLiO3xGOC/z2ZSS3ujtSptSPh1qHAEImb+Fms8iapytKqa7HWEStwmIqr2VUOUi\nuONLJaOYk8ghotEf7voMMixQ5+AnNPrdUEPIJfz3YcaKfrUZXd4wTNxmlbxEG74N6sxJZgkCFqQB\ngwLGrdHukUklQju1qUzcZIN32OotCMmJEtOBiOM6YBhUz8p2DRm3GZLHGdcQtb8Nhu9DRQHrV2/1\nIjGPIBI30e8323Mxq6vkYTJuIPrcVASnsp8pao84zF5sk8ljJHOSvJpEehbGTZV89igkbgPXZbva\nCi2lBrUGKVF73EC8F53ebO6WYvh2O1LBMWFZ5LJqio5RDcZkHDhrTHc7wiRuExDFwXAU0kVw1stn\nNyKdDWokgrPIA0Ecdioc/BqtnhgoGyVpKqi5gKU7WZhZWf4a/CBgtmr6kHGL0Hup3Jwk/AUsJb8q\netya7V5oG2EYBlAqetx8uUkEKVYukxTS1xn3Rn2GgkJZ0RBun/WLJJVU0+O2ttMAYHVxBsZtxjW4\nrkuzPVviBkNjj6iQz0TYdchEQQmrUBcDhrMh5htK5LIpJYzblmTcIgSoqiT2m3stkgkrUp966Ugw\nbmqlklHMSXKKWJ5qvUOv74bubwO1Ji3y84gklfSNWqLvj2a7R7vbj7QvQBQdlTBu7Z43k9ZIJYPC\nJG4TMGviNrRVnjFxq0XXhat4uGcJCkG8f/2BO3OfgPw3RLl4yoo2eb3Zjfw+FFUxbjMwwaqcDGcJ\nAuaL6nrcokrSMukk86WMIqmk1+AdUSoJsyfS/h6N5CqpinGbfQ2zsl0+4xYhcVPl6Njpit5o2RcU\neh2KEulaUxhrhWWjxdxSS42rZCO826pEXlH/55BxC8+wSBnbrGf25l6L5blc6L5LUOdQ7RfbItxf\nyqSS3pk/H0E9ZFmWSOZnTBa2KtICPzrjpiJhkQXcKH3Jsmg7C/PnO0pGeB9Ancy/0eqRy6ZCz6RN\nJRPks0ll8z9vJ5jEbQKiNk1KqKrW7VSFDjmsHhvUJG6zMm6qJA7DWXLRJaOz6qGrzW6ki290DbP3\nuEV/LofmJIfHuOWzSTKpxMyMm89sRNyfq/N5titt+oPZCgpSKhmlmp5TUDmFYXEoSh/qnCKmqTYD\n46bKVXJ9J7pUUlkCG1H6I+HPcpu55zDaeAjfWGvGIN11XaqNbqQzAsRZJZLgGY1BdhUwbjMkTe1u\nn0q9E2kUAKibCVv1C5/hPw8hR7dmlgju+lLJaCxPQUG//NBRMvznodLZcibGTc5Rm2GPbs/Qmw1e\nn52Cwkqz3Y18h6uaz3u7wSRuEzDrBayqUrZba0cKCuFoJW6zHnbSyTCSMYgCPXSvP6Dd6Udm3EqK\nEvlZEjffnERBj1vCsiLtDcuymC9l2K3Pxrj5rq8R9+fqQk70O1RmW8dOtU0+m5opkZ41GJlNKqmG\n7ZKMW6QeN0VJ08ZuEwvx2YZFIZsimbAUGKTIUQAREzdFYwmiGjmBeI5mnY/U6vTp9QehjUkkVMjB\nQDBuCcuK1Fel4v6UjF+U/jZQF0fUZlBJWJZFIZeeuc+uUove4wZqbPClo2Qkxk3DqIyoPW4wmx3/\n9gxuyCAY8V5/diVV1NFGMJzPO2sbzu0Gk7hNwOw9blJmEf3AbbbFjKiom0uFBX3Dr6RH73GD2W18\nZ7l4VPSXzZrAppKi1+NIMG6zMjzNLsV8eHmDxHwpS6XemWl22KyuryveTKXNGeWSO5VoDd6gzlZZ\nBnWRHFcVsV31Vo9MOhlJGaAqaVrbabJQzoa24QcRnJYLaWW9flELfrLINMuZLS3sI/cl51LUZwyI\nZjEmgeGsqJmlcXtN5kuZSDJFFYnbdjV6ogDinLdQ0eMWfYQLiD06azK/V++QsKxIBmMwdFOc5bmU\nidvyfIQeN0/+rEQq6Y8DiOYqCbMVw33vhBl63ACaMySPA9el1e5HPivLBTGneNZC9O0Gk7hNgIpx\nAPD/s/fmsZZk93nYV9vd79u33qane2Z4ZzjDIUWJy4SUKYmyJcpMbCkyrFiwIsFyFvgP2UYAy//E\nCJxEhhFYgBEDgcHAAqIltmzYBmxFCwklNLWQI0pcZsi5M5zumd5ev/Xut/aq/HHqVN3ufvfeqrO9\nbk59gCBOL/edrlvnnN/y/b6Pj+LQ51CUBMQoGdJDqsYwYA5kZqrcQ8UciRMVP+B5DryzfgAZNOdN\n3Hhm3KjsL2+iQMQX2J/DWquKOOZLFnhVX1NlyQG7QInjBZi6Afflx0s5SX3cGC7AVo2YPvPK4E8d\n9g6PlgRzPB03P4jQH7lMNEmKlUaFv/PIydRIC10c7wTvWdWsWSSo4jgnRom4CvXHKwo6x+NwFPxI\nR93hKHzyJ9FTTlVmymzgV5X0Ua0YTEUNgJwtRCCMPWkaTFysNC3mgl+9aiIG3/11yjHjVqsSgRQh\nVMmJj1ql+AwqIGZv0GfI2pigcR3Pd+G4AWKw3+Gi1MKfNJSJ2wKIsAMA+CgOvFWRqmVA0/iCAFrN\nqDGoGAISZtwYqnUVS4dl6lwdN3pYswTHFC0B3n48HTfTINL5vBWqqRugwfg+AJklAM+cG+8MKqXT\n8QiU9LirloI6bnaQvOPFvxNd19CqW9zzn1MnYK6kA6S4MuLZn46PGFkHkWkNDQuuF8Lj2B82Z8Gv\nJsBomMceAhBjCUCLMsziJAKokuOpjyCMmRSZATEdN95zCiCxBLcBN8d8NkAo0GEUw/PZqXHDic+k\nKEkhIpY4HTowDZ3pWeiJQIoQqqTtMbOYRHTcaLGQtSBP5+x4VNOnvOyE96iXW5m4LYAIOwCAr+OW\nebixXTyaphGvKK7KDN3gfBVkUTNuLAeupiXBKUdlhrcDCxDxCC+I+ALD9L1k7IBWDC6qpB9E8IOI\n6zlQVbEBx5wbq/8LRUqV5Oi4nXImbiJmFQC+eSYg6TRxdNyiOCaJG8ca2g0r8S9jCwzTCjJjIAKI\n8ZPLOm7nJ+aUddz4FJF57q4hJzUvTWA59gYtrKwx7s+GAJoibyIPkESah7pKhGI8ZtoqkJ2zrMl8\nGBFKG2uyAgjaG46PdsOCxtj1I/ZG/InCeOox708R9wY9L1njOhFMKh72EJCdLe81SwDmk6TT6fwy\ngI8DiAH8QrfbfXXm934YwP8KIATwW91u9x/yLvQ8kPllcdoBcLxUtCPBevEAVLZVxAbn67jx+tll\nM27stFGe7gpN3FjFMIBZSwAfGwxecED2XrIedlWLT8aXt6ABZIUIro6bw7cOEYd+j4N2A4idcdvm\noAg2aybuHU8QxTETjclOKC88gWEmUOIxPU+bs8AEPCjFv8koJsErTiIiOB1z0vPoOcUz00Qpp6wd\nNxEzbj1O5TxdJzRFHoVNMR03E0EYwfMjJk88IhQTMyfRwIPCHBsrxf8+bxwBZMVKHn+/qRNwxVON\nmpkaqrMiCCPYLnsSK6bjxhnXpTT/84slmiVVMj86nc6nADzX7XZfAfA3APzTh/7IPwXwXwL4BIC/\n0Ol03s+1ynNCGMWomDozRVAoVZKx4wbw+22IStx4O25j24eha8zraNUtOB57RT+jSnLMuAmk3jBX\nyhLTZ96fz9rpAmY6bhxebrzVuopFbAm4ErdEeIC1Iy4icQvCCI4XcnW76OwIayc2Nd/mWAPdV6zn\nhMNhTE8xmzyygnd/iHgneIWU0o4bR8JCZyZZk/m6gBmalLHCEag3OentvOcUwE9d5fFwo+Blztic\n9xYwI4PPuIYojhOaP8caqoTBFHHM+vGel3UBdGpaSGcpBACZlQ1PR5yXKtkuqZKF8GkA/w4Aut3u\ntwGsdzqdFQDodDrXAZx2u93b3W43AvBbyZ9/4vBTn34W/+Bvfpx5kLZWSYxMORI3HrW4bB0m9wCp\nhkzYoihEqUqObB8tDooDbxWZt5IOZEEMr9JovWoyqaQB5KA+744bb5AubB2cQRntKrDKW4sQJ5lw\n+KdR0C4ya0BE9wbPjFuDcw0iOm4iCiu8qpKmoaNi6Vx7YyxAnATgKzpSqiTrTBNvkA7MvJc8glJ1\nIijFSlPkpXTP/l3W2SoeD7dH1nCOhRXebrTrhYhjvju8QQVSBFCZW4zvhIiuvOOFqFYMdqEYWmDi\niOsmnDRiEWrhTyJY3949AF+d+e+j5NeGyf8/mvm9QwDPLPvA9fUGTEa1I1nY3m5zf0a7UYHjhcyf\nFYFsqsuX1pgvn3azgiCMsL7RZFIwCiLCZ97dYeBHAAg08jNjTeN6phMnwPZanfkzttYbAIBKvcL0\nGbFO/h0X91aY13BhuwUA0EyD+TPcIEKzZjL/fd734c4poYlsbzSZ1+Al7zV0nf2dEPB9rLaqOOpN\nmf++n9gZXL28ju314lTFkUe6v5rB/j7Y4RAAsMXxfWwk8361ZpXpM26dTAEQY13m/bnRBABUamz7\ns3KrDwDY3mR/Dpf2yBkX6+zfR5jE91curmGVsRPbrFnwg4h5DXESjF2+sMr0GRf3yPcJnvcyebev\nX91gOmf2ktlTg+Os1AwSU1zYYT8jNlbruHFviNZKnWluMX0fLq0xzz1uJ3vDqllM/46bRxMAwN52\ni/k57GySu8uqsN09RwktfnO9wb4/d+n+ZLs3DnvkvV7niCM21uhZWcP2RoPpM+izYD2r2itkDUEU\ns99dYcwVR+xuD+BU3QAAIABJREFUk3fKsNg/QzMPAQAXdtn2p51sriDmiy2fNLCXHR7EopQ9Vzrf\nSzbU44bt7TaOjkbMf79eNTAYe8yfMUioWOPhFPaYrUFqJN/A7bt9puRvbHuoWAbzv8FOKq+9gc38\nGUEYYWL7uLLdZP4MA2ST3743QN0oXmU6Tt5Rz2H/PhGRYObu/SHzZ0ymHjbX6sx/X0sCibv3+kyB\nxP1D8nOjIGR/JxIK02mf/Z04od+Hzf591CwdEyfA/YMBDL34/jpJZiZd28VRULz6aU/I/u5xPIfb\n9wYAyPvN/l6Sl+Le/hANhr1B17DaqjKvIQpI5Xb/YIijreIB0eHxGADgez7zGsKka3dwPGL+jH7i\nEzUdO/AYK8EVi/g9sq7h6IQEVb7L9hl+Ir5weDJmXsNxf4pW3ULvdML0953k3jjuTdnPiD45Ixzb\nZf4MK2E2vHunlwoaFUF/6EADMBramIwYhZBCcm/cuz/E7krxYsCdfbI/tThi3xs+2Rv3j9jeif0D\n/ntDj8lzuMN4f945JGeEEYNjDeSsvHOvDz1k6zbd3SfFNi1iO7PjOIZp6DjliKkmtod6zWL++56T\n7M/TCfNnHCRndsh4ZvtJR/2IYw2PKxYloqxUyXsgnTWKiwD25/zepeTX3pNo1ixMHJ+ZD217ISqW\nzhRUUmQzE4wUBy/k46ULmHGjLfUWB9Ujpecx0k1EqINRqgqral0Ux1xDzQD/DI2IGbe6CH68SMom\n4zsxnvqomDqqjDRiIVTJ1MONZ8aNrJ91j9L3mdVvcnYNrPSfTFWSgypJqcyc6rNVRn8mikbV4BMn\n4bQDEOGPNJx4QqTfuYS16BnB8U5Q5T9Whc2U2s5ISQMyGjTr/uQV9gJmRx5Y96cAUatkVpHO/hcF\npc7yCIyJUrYE2Ontmqah3eDzhCVxHTvLTYQBdzp+wmpbkuzNUpwkH34XwE8CQKfT+TCAe91udwQA\n3W73HQArnU7n6U6nYwL4bPLn35No1S3EMTsP2HEDrqQJyDYYq/CA4wVcG9w0dFRMneugEzFc3eSU\nMxaRKPAqGdpugCiOuQIiOozMKlDCq+YIAJapw9A1rndiOOGTGwdmgjLWxC2Zu2SFECEKGgQwSksD\nWVDGGiTTeaZVHrW2Kt/co83pSwSIGXafuj5XcQcgZ7YfRAhCNiGlseND1zTmeSI6szlgtIgIwggT\nJ8AKx94QIcBgC5ir4p17pIkbDxqcYjGjNHHjl+JnnnETUFih/p89RlErEV6svPOGwKxdB5+YE+s7\nGUYRvCDisk6pCTABn3LOZxu6jmbNfM/ZATAlbt1u9w8BfLXT6fwhiILk3+p0Oj/b6XR+PPkj/z2A\n3wDwnwD8y263+6aQ1T6BSC0BGJMFxwu5NhfAFxxGieEm7xrqVZOvgjzlHzJPVT4ZN/nE8VG1+Crp\nWceNLSCiBxRX4mYJ6rhxBCPEX9DgOvT7YxfNmslkOk1Bu1SsBz+vf5pl6tA1jatqSTsBXAJGnEEZ\nfZ9XmzyJG5+IkcNp3wLMdlfY3oc4jjGc8BkdA/xV/bEdoFk3mYWcahWiuMqauNGiCuuMHzCjWifE\nZPj8BGumbsDFTgCyjgSrsJYQVUle8SCqKsmRRFcsA82ayWwjI8KLVYQv7ViAoFSrbsJ2Q6bijpsq\nhfMrfPJ0xNOiI6dS93tNVZL5W+t2u7/40C99feb3vgjgFdbP/m7CA8kCg8+S7fF5jgB8iRuvSSNF\nq2GlnlcsoAEET8KSJdGsynn8FzBdAytVUgTlhb4PrB1YEWqO9O/zJCz9sYcNhnmPWfAk834QwvX5\nZPi1pCvCVb0VcPnxJgr0fV5tVZjneHjpmiI6boauo1E1mQMB2w3g+iGzITvFrF9Vm0H/YGL7XN0V\nTdOw0mQ3ZU8VJTnOqYqpwzQ0rgB56gaoVw1mBV4gS9xYmBpRHMNxQ+6zkteeQYSqJO8ZIcIOACDW\nKz1mquT5K3wCM2c2j9pp8l1ObL9wgeRx8tQzDQ0Vi68YftQfIggjrqL6k4T3xr/yHMFDzwujSEi3\ni2eOhv4dVq8PirVWFdMkqGFBn/rxcFRweSWubQGVU9MgrX3WwFBIx40m8pxUSd5nwdOFdb0Qthtw\nvQ8A3/4cC+h0AeSdZg2QgRlpaQEeTazfx3DiwTJ1rgCVNzBMgxHOILnVsJhnJmhAuc5oyE6RVrMZ\nnkUUx5g4PldQCJAkfDjxmOazB2NaaONLHlt1i8tTjyRNfM+BpyvvJMb0vNTZLFlg7bj5MA12D1Rg\nptPEmLCkVEmOjhtA5mhZYwkRbBEhM/sC5pJ5OsG2gMTNNMi4A5ctghOgUWO3eAKA3fU6ojjG8YBR\n+OcJRJm4SUZarWMYbBbRzgZmAnWujhvfYUsNxPuM3PQ0IOI0UgXYuisijDspWo3KuVIlaVDIGgTY\nAi4/sg6D2ci0P6GJPPtzAGaSeYb9mdGP+Naw2qpg4gTwGBNpXs8uYDZpYlvDaOphhcNjEeCnYmVC\nFHxnFY9vV3pOcb6XPEms7QaIY3ZhEorVZhVhFDOdl0MBDAmAVNNZ2QkA6cLyFph4AmQRc9HAbNGR\nncrcblS49ifvXHLK1BDQcQOAAUMskRUd2fcG7W71WBVCIWbmkOe9FEEhJmwRkzuB5aGLAsDeJqEk\n7B+zqdc+iSgTN8lo8lRFBAxWA3xUSVrV4k3c1trkAmdVg6IJH1fixkE3cVxi3MlTIaOgalAsCUt6\n4HMERNRv7LBnM/39aWLIztvZqCdGpiyUzb6ARB6YSeYZkthswJzvOdC5MNZ5IiEG3KkQRPG9Eccx\nhlOfi4YFkNlLXWOnxjleCEPXuOkyrbqFMIqZzsuswMTZceNIpMcCOrAAn0DJUMDMI0D+DY4Xwg/Y\nRFpsN+Snt3PMPdLvj1+sJtkbHFRJ3rlL3kDdFtQRT5UlGebcpi6/quT2KtnbrPcnQPZHq25xnVV8\niZuYuK5ZZxdIieMYUyfgjqkubBKPw/3Tx9NSTAbKxE0yeOh5Iqois3+fiSopkJcOsKtB9UcuNI2P\nelOxDFimznQBpwe+gI5bO1EaZbmE6XvEU8m+kJiG3jthO+hsN0CtanDJWwN8XYWeAOoswNcRFzE3\nAswUNRj3xtj2UeOUn69zdLtoYM3bXaHzfjarXYdHZol4ugoAX0AkghkAZAIOLDOg9F3mLyiwJ26U\nKrnK3XFj/y6oKmeDkyqZzbgV/y4y+Xm+AFnTNDRqJlMc4QchXC/k6u5QNGrsiZuojvgaB3uHV8UQ\nIHHEeruKoz574jaaeFzCPUAmNMNG4RWTuPEUoR2PsG14CysXaMftpOy4lRAEnsAwq1CdX8dNOFVy\nxNZV6I1drDQrXH52ADmwWS4/UXNdAJ+yJO248QTJ6+0qahWD+aCbOmIoo6lqHFPHjTy7851xE9PZ\nWKMdN0altInDp2wJZPQllsQtpYwKCAx5Kvq8tiUUXImbAGYAwDdzKOq9XEnonkOG9zIVJxFAlQTY\nzsrUN4zzzK5aBgxd42LN8CaPQJI0MSSPIwGCVhQ8c8m0I26ZfHd4WgRmYO+kVjachejttTpOhy5T\nJziKYoxsn1t0jofNlTYFOO9xWoRmKYbzetlRbK/VYega9hkL0U8iysRNMngCQ1HdrschccvoDcUP\n2ziO0Rt5afLHg2bdYrr8RFTqKGiAyzK7MRbQ5dE0DRc2Gzg4nSKMil88UwG+REBGAWYZbhYhVgPw\nXX7CErc2Xzd6bPvcdBNd11C1DCZqnsjAsMERGNpuyH1WAnz7U1jHjSORFuERBXBSJQV4LAJ8vnq0\nIMRbbKMiKWyJG51x4y8okKIje+LGe04BZH96Ppu/oJ0UVng74jwMBREqowCws1ZHDOB4ULzrNnZ8\nxDGZbebB40CVzAor7MVw3rvLNHTsrNexfzJlmkt+ElEmbpLBVxWhhpWcHTcOc12RqpIA22E7cQIE\nYcQdpAPkkJg6AaKo2AYXoUZFkQYjDFXkse1DQyYFzIq9jSaCsLgSE5G3FtNx4+nyZIkb33MgkuM6\nW8dNUECUBsgMnQ0/COH5EVqctDiABJcs34UI6fdsDSah0BTcn3Eck46biACZQ8SoN3JRqxj8hssi\nZtwEiJMAwGBS/MweTPhneIDZJJqh40bPbAFzya26xUixF7eGRs1CEEaFRYxEdsR5OsGOoIJfJnTG\nMOMmiC2ywzEnPhLgcQgQ9VsATAq4jgDrFIBvf9L9JILFdGGzCdsNuNSZnySUiZtk1CqEZsESGIry\ny3ocfNxWmhY0jY3eIEqIAiBVyxjFpXxTioVQqiRbl6dZt2BwVgwvblElpmL0AscNEUPMrF+dhyo5\n9qCBn4qlaRqadZOJyiys49aiiVvxvTFO55nOj6YoQiFtdg1A8dkuz48Qx/wUKICvy9MbuULOKZ4Z\nNxEqo8Dse8nWcePdm0B2VrIEp6LUb4GMqcFa8BPVcQOKz9qJpkoCrGqnoRAqM32vWITOSMeN/5xK\nEzeGObdh8n1wz2fz2FQIiut4YplUVEvA3ZXNub036JJl4iYZmqYRigNDYCiqnV2z+BM33q6foetY\naVaYOm6pEIWQxI1NLIYOmYuYVUirVCz0vKkn6KBLlJgKzrmlwZCABDY18GTsuLWbFSGGm62axTfj\nxpmwrHJ1o8UE6UBGUyxKNxEl/Q7MBIYFg1MR5tsUrBQkPwgxtn0hiRuPjxt9J3gLCjQoK0qVDMII\nEyfAioBEvpWyE3iokvzrYC340fdYRKGL/juK2rhkIkpi5uyA4s8hjmNClRTwHExDx0rDKnxeRlEs\nxIsVIHNVAHDE0nGbium41avss5fiqJLssQx9h0SMn+xtvLcESsrETQFYJVNtQQOklqlD1zRGqqSY\nygxAKA69kVc4MKRdOl5aHDAr7Vyw4ybwkGEduI/jGGM74JZ1BtgrVKJ8iQB2qmQcx+iPXSHvA0C+\nU5Zq+sj2UTF1VC2+y69qEWpdn4HmIcLIlaJWNRFGceH5FZFUSdbAUJS5L8CeuGUebgISt8fADsAy\ndTRrZuHEbSiICgbwBYaOwCIT6zshkmLP3nGjVMnzK6yI7IgDpFvVHxeLJWg8JeIO5+q4TcQIaxG2\nCFts+XhRJfnvrotbpBDNqpT9pKFM3BSgmVT0i0qmUslW3sNO0zTUKgZjx01cJXutVU2rsUUgwsON\ngr3jJiEIKFhFtt0AURwLGTLPlJjOr+NWY6Td2G4Az4+EBMgAKaywVNMnCW1VBNZaFSbqD6VKingn\naFA2LZgsSKFKMrwTgJgCUyul5xULRtLEbUUAVZLDV2+SJm78z2K1VS1M4RWZyLcZvwtAnDgJwJ64\niaRrpkUNZqqkyDOCrSMuorACEMVT1w9Tn9k8SO9wIUm0hWbNZLIEoFRJXnESgNC6z5UqWeenSvLO\n4gLAxgrx1mMZN3gSUSZuCtCsmYjjLBHLC5FJU61qFP75ZA1iWupAlngVpTj0BVayWYUHhIqTMFap\nRoKoeUCmxHSvoBLTVCD1J6NKFnsve8nMjQjqLDCTzBd8J0SY2lKstaqYOEFheWlRksoA0GCkrmYK\ngmJUJQH2jpuIc4o+y/PsuOm6hmrFYJxxC1CxdFgm/7NYbVYKv5eTtJggYraMfAYTVVKgMAj3vSFg\nf7IWHUV23Pj3p5iOWyuljeZfh8g5dYAUP4/6TuGCvCiqJMA+eymcKslQWEnHT0SMXXDoODyJKBM3\nBcgMPAtW61L6j4DuRsUsVJ2ioBuBV1USyKiORTsLoiS2AXa6SdZx4w8CKpaBqmUUDkZEUaAoNldq\nsN2gEDVOZAWZVYhClBUARVpNL7A//SCC64VCkmggq74WrRhOBL4TrBL0o6mPetXg9mcC2OceRVmn\nAKSoUa+axRO3lBlQ414DwG6NMHH47SEoqOJpkcAs7X4KOCMMndA1mewABNK6eTpupqEJSaLTpKng\n3TW2fWiaqLlkNgqvSFsEIEtii3wfIouvAKFLBmFUOJ5JqZIC4pl2whYpGls6XgBd4/fU4xJaEyhO\nko0DlYlbCUFgtQTI6D8COm4Vg9kOQAO453iAGePMgsFpb+yiYupCLmDmyqlDZPhFyI0DJBAokigA\nGbVSVJenxnAJy5hxKyrAIFIMA5ide8z/fYhOolO7jILzRKIUBAF2qe+x7Ql7DqyBocgZN4CNgkTV\nF0VQoAByZrPOuIn6PlIFvwLKkiK7nwAJDlmokpRhIrTbVfjeECOBT9bA5glLJfB1Tv80YJauWTBR\noPeGoI4bC21UZPEVmGUQFWTOTH0YuiakuMIaWzpeKMRTzzJ11CrFi9BA9g6JYIvQcSAWhsKTiDJx\nUwDWA1dkt6tWMRCExYUHXC9EVcAGB2YOuoIVqv7Yw1q7KmQNrZRuUlycpC7o8gPYzFRFJwuUXlDk\nsBNKb2Dsrog0QwfYKtmiTI4p1pqMHTdBCoIA+3zZ1AmEBUOs4iSZqqSgIDlJ3ArRiAUKGAFsKp9B\nGMHxQmFnBKVCFbm70nkmUbS4hoWxHRSmpNF1iPFxS+izRdVOBXleArOqkmx3l5A1sHbcBCfzTQYW\nU6YMLei9ZGRSDaceWg2L2wR8dg2FEzc3FFaEbjcsjGwWcRJC6RahDA2QeKLoOBBlzjxpKBM3Bci6\nPMXpP7WKISRZqDJaAtDKjAjQ6u2wQHUmjmOMp76QwWqAPYmeOGJkhCkaNROuFxZKpIUnbtSYvcBh\nl0pL1/m7CpZpwDS0wj5uIpNHgM1PTuRsGZDRZopWb1MfN4E0qCJJkx9E8IJI2HNgVa0T3nFrWAjC\nuNB5KZIiSD+nqMqn6IICS1Ejpa0K7H5GcVw4YREppMQz4ya641bYg1SQBD7AXmwTvTdYRh5EzhuS\nNTDORk99IfcnwC505niBsCIX6YgXK3IBYindACnaFWWV/fK/+hr+8W/8mbA1qEKZuCkAa2VGZNJU\nY6SlidzgLFQs1w8RxbEQ/zQgq1qyXMAiEzd6YBW5hEX5hlGwKNeNBSqUkTUUn+NJjTsFHfosnaZs\nXkJQZyP1qyqWuIm0A2ARi8mSaMGJwjmqSgJsQZloOlidIWmSVdwp8n1k3p/iAkOg+N5wBM6Is3Q2\n/CCEL7CowUIPDCPSURDe9WO26xC8PwvEVKL3Rtb1K1bost0AK01Ra2Cb2RcZW7brFsIoLtyFnTqB\nsL0BJAJ8XlgogbxzNGGaIz5vlImbArAM0gKE6iHqoKPVz/PsuLFcPPQwEFVJr1dJB7PIQSf68gPY\n5qooj1z4PFGBKlWmUCZqDQYzVVJ4x61I4iZ4DawB0djxyfyKANpN1oEtQBkVTFtlpWuKn6sq7h9m\neyFMQxci0gIALapuWYTCK5pGXCv+faS0OEHnZaZcV3xGXBQViyVxEz1TVaua0FCQtuqKTZhYbSpS\ncRJhVEl6fxa5u8QWHdOuX6E7XJxVBpAVDouyJMIoFjqDCqAQXTKMSAIruuNWhKEQRTEmtjg2l0qU\niZsCsJo+224orIJcY5BLDSNCgxLW9UsuniKHTEZvELO5NE1Do2YyXX6i1jD7WYXmqhyxNCiWd2I0\n9VGtGEJU0gASUBSlSmY0xfPr8oies6szBMgACRrOs7siPIFlTdwEiuYAMyyJIombGwgrMAFZZ/08\nRXNYun6OJzZIbzPO8dheKKzrZxo6qhWj0L0hOonWk7urkCCHYHqgaeioWkZh2qojOJlPGSsFvg+a\nNAnruNGCfKE10ORRUOLGIBbjCJ4HZimsDMYeYoiz9AFmZvZzdv7Gjo8Y4r4LlSgTNwVoMbT1gzBC\nEApMmihV0i9AUxTsvaJrGmrVYoc+nXURGRA161ahYGgieKYKYOPoZ7Q4sd2NIh0Wkd5lAKFTuV5Y\nyIdGpJEqwEuVPD+BlDiOMbaDtDDEC5bnIDqJtkwdhl6sIw6I77jR4K5Qx03gPBOQ3RtFAiLRiRtL\nIk1nZkXO0AAMVEk3EJYoAKQDWuTeEO0bBqB40VGg7ybrGoBZwRqx4iRFutHUFkFU4ZOlID+YiGWs\nsBQdZai+AsX2p0jPS4o0xs3JIBLdgVWJMnFTABbJVtGc8LS7wiD9LmrIHKBKaefnvQIkF7CTX61N\ndKIAsHH0Jw6p6Bu6mG2b+nbl7HjFcYzR1BNaoUqTxwJ0TfocRNADZ9dQRBBjIni2i2Xo30nEbVqC\nBt2bDFRm0R03TdNIYaWweBD586I7bkWG/m1XXIcHyDpu4wIUJBoQiapks1B4RdrYABx2Ol4gLFGg\n6xgXCNJFFzUA8n0wddwE3l2rzQoGE6/QLNEoSViEJU1MNEXCThCnDF38Du+NHABiPGmBGSoz0911\nfh030ecUMEvjzdlxEzz2oRJl4qYAVBmyGD1PbIWqxiD9PpyQ9a4K8ssCgHrVYgoCxM6XkWHavBRB\n0XST2c8qSpUUGQQUTRZIohALPehYBDFs1xf6PtSqBjQUnOMRnLAYOqFiFdkb/dTwWSztplgxQXxw\n2irYEQeIUm2zZgqTlm43iiULURTD9UOxVEmGhOVkSALDzRUxJuD031OMKhmiaokrrGS01WJzyZ4f\nie2A1i24PhEcyQMZ90azZsILovxrkFB03FipwQ+iQt3ow76NiqkLiyVYKJujqTi/SYDEVIQdUGR/\nkjNb1P5ksU+hfpNrgrpdWeJWoMCU3l0CE7eCReC04yao8KkSZeKmAKSKbBa6eBzBNEU6DDssYPA7\nmJDNtdoUt7kaNRO2m58aJ9LwmaKoJUA2zyTu0GfquNlih3kzA+x8SZNoYRIgq+gXCU4nAn3DgBkK\nb4HkcSI4cQNIcFUkEOkLvoANnZjcF+kqiO64ASRAnjpBIfrscOIJM2QHind5UiqY0EShUmgNAHCa\nJG4bggKiqkWKjsXESQKhLA3qoVZE/MB2xdLByDqKndkykqZGwdmutAAscH/SYLs3zOc5GccxDns2\nttfrQrxYKZr1/JTNMIowdQKhbBFN04gfa4Hz8mSQFFZWBRVWKsULwJSuKSqJzqiS+c+pvgSqZOZL\nmzOeEazSrRJl4qYIrYL0H1swTZEGd4MCPlGiNziQXWJ5qyIivXgomgWFQUQbdwIZPz7vGoIwguuH\nwuaZgOzdytuFFT1YDQAbK+S9pAHnMoQRMRgWKSMMJCIpBWfcNAieHSm4BtpxE5W4AdQY/vwEGABy\nVsbIHyCHUYSx7QtTagOKC2KItiMAZkyfC3XcXLQbFiqWmHuDijkVnaER+xyKd9xSKpbQvVHsnZDR\njS46Gy3atgSYObNH+c7ske3D8ULsrNWFrQEg/6a8Z8TEDhIhCrFBelFa98nQgQZxnSZd11CvGoVm\nggdjWpAXlLjVGaiSj0XHraRKlliCZs0ih0dOXrhoHvJai2zSfoGO2zBJ8lZaAhO3ghTBqaThaiA/\nP34qoWpZ1E9OtHcZUFyMQsYwL6WMnORM3ERLbFMUTtwcInwgal4iW0N+HxopiVvBQES0jxtQPGFJ\n38tz7LjRrrXQWVzaccsZEMVxjNOhgw1BNCyKetUoplrnip0tqyf7rEgCe9izAQDbApOFoibcMrrR\nRe/PbNRA3Pex0Sbv12nOjhv9LnbWxSZurTo5L8NoOW00DdIFUiWB4nHd6dDBWrsqjNINFNcNSAvy\nwqiSxe0AekMXGoBVgbFlUZXskipZYimaNRNRnN+kkL5UoqrI7UYFmpYFe3lAk7w1CR23vBVcWwZV\nsqBxphxxkmKVU9GKkgDLQUcvP5Edt2JBgIxgCKC2BAUEa1xfeNevkZwRrp/v+8gGvMV9H62aCc+P\n4Af51iCj49Ys2GGh9O9VgR0309BRrxq5q8gyhJyIEJGWW258ZPvwg0gYTZKiUbVy31uiLWSAbNSg\nSOJ21BefuBWdORRtGUI+i5GuKbCwUrTjdtibAgB21hvC1gAUY86kHqiCpd9pXJfnDo2iGL2RK2y+\njaJRK6YbQFlXojpu1YqBiqkX7ri1mxWhCezjMPqhCmXipgjNgvx4GoyImtvQdQ0rzUraJs+DtKUu\nsKLPXjEUL7OdNyCScQHXqaddXqqHYA83gMw0VSw9f8fNPv+Om4xEASDvVxznT2KnTiD0naRroJ+d\nB3TGTeScQPGihi+cMlo0QB7SC7gp9gIuQm+n9ByR7wRV2MzbcesJFj6gaNRMuH6Yy9hWtBoyRatu\nFUvcBjRxE/csaCc4d7FNQje6MGPFpWqr4hLpojNusjpujQLFz7GEuwso1oXtj12EUZwmvqLQqBbT\nDRhMPOiaJnS2q92wUpXGZYjjGP2RK/TeAmYL0ec3+qEKZeKmCEUHm0V33ABCqeqP88v4DiceDF0T\nLsAA5O+4SREnqbNRJUULYjRqJiY5nwPtPoikSgKkSpV7mDetUIl7J1ebFRi6lnvGTQY1DyhGG6Vz\ndjK6fkD+vdEfu9A1Tej30SxK4U28y0RSRlsFZ4lGifqtSHESgJzZo6mf67zMOm5i34l2gYSFFj/E\nUyXz7w0Zs35ARuGNct5dRzKokgXfy8fBRoZ2SkWel2utKjQAp6OciVvS/RQ941akGC6TKknWsHxv\npIqvgoRJKFJLgJwJy2DiYqUpzhYBIJ3MIuwEL4iEzrcBMzP7BVhttYoBy3zy0iCmE6XT6VgAfgXA\nVQAhgJ/rdrs3HvozPoA/mPmlT3e73fyybd9loB2C/HMb4tu4a80K3r0/gu3mU+UbJEptQud4ClcM\nA2iaWHWwIoctIOcCBhIj08KD7mLXUKuauQ24M7qJwARW17DeruafcXPldNyKBafigyHyecVmDvsj\nF6utijDZdWCmqFFgb4hOYFk7biKLXGQdFQThCJ4fobrk/KH0HJGzXQAJTu8dTxBG0VL/RmmB4cze\nWFYkSM23BXZ4AJLMxzFZQ57i1VHfRrthnWsneOqK9ZsEWGbExXfcTEPHaquSu9h22LNh6JrwTlPm\n5ZafKim6u0LPyzzsnVRRUnRHfIapsWxvxHGMwdjDhc2m0DW0GxbeDSK4Xrj0rEzNtwUnbvWiBty2\n90TSJAGjtgfyAAAgAElEQVT2jttfA9DvdrufBPC/APilM/7MoNvt/sDM/71nkzaAYW5DRuKWbJR+\nDmXJOI7RH3upqIkosMy4NaqmYBnhokPmPgxdQ8USW5kpYqaazrgJrhjWKgbDMK/YNWyu1DAYe7mo\nWDJk+IFifnITCSqjs5+X552QtT9ZlPNEd4GZaeXCE7dELS3H0L2MWVyA7DWisLn8nUitAERTsQp4\nRaVUScEdtyJJUxTFOB44QrttRdcAkHtD9BlRdDbadomn3rKkvyjW2zX0Rm6uDuhhz8bWak34Gop0\nH2XQ/B9YQ453QrTHIkWRYrjjhfCCSKgoCJDNvefxcpNhvg3MehUvv8PjOMZ46j+RNEmAPXH7NIB/\nm/zvzwP4hJjlfPeiKMVhOPFRrxqwTHGVMjqMmkegxHYDBGEk1MMNmA1O89NNRAdDLJLKjZrY5BFI\nhCCCfEIQY0mzXfWKAdfPx48fTT2Yhi60+wkQaleMfNSbadp5FNztKlBQkCmQAuTruE0csj9FKkoC\nxQKRICQmx49Nx03CjFvedcjwcQOKFZlEm/tS1AsUFOhzEH1GtAoYop+OHIRRLDxxK1rwE+03CRT3\ncZu6vvD9CZDiQBjFGC1RqZ46Aca2L1yYBMi+j3ziJGSdIg24yRryxxLp/pTUEc9zd8mweAJmTLhz\n7I2eBA83oJgdgO0GCKNYeBFaFVh39B6AIwDodrtRp9OJO51Opdvtzu7iWqfT+XUQOuW/6Xa7/2TR\nB66vN2AKTFJEYnu7zf0Zl/YIzzvW9VyfN3F8rLVrQn42xeULqwCAKMcabh+MAAC7W02haxjQjkbO\n5+B4IS5sil3DRtLZ8cIo9xrajYrQNQDA+modQA+1Zm3pXArtRV2+uJauQ8R6VhNp5+ZKfemlNnUD\nrLUq2NlZ4f65s7hyYQV/9Pp9RNrydyLWSK3p0t6K0O9je7MFAKhUraWfe5fOzwh+L/eSz9Itc+nn\nTvaHyd9pCV3Dxd0x+R+GsfRze4mq3MZq/YE/y7seq0aCCj+K8+1Pn+yOa09tCA2Ud7cJncioLH8n\nkHQTLgp+L3e38q9hNPVhGhqeuboplJ63k9CqrBx7o3KPvJei98buFtmfRo69sZ9Q0p6+uCr0rNyI\nYmga4IXL38swJHOwou/welJIDeJ8/ybHi7DWrgq/uy7ttvHV7hFic/E58Z07fQDA1Yti9wUAXOqR\n7zlPTOUG5Iy4fnVDaDH80i65C/LEMzSp6VzfSs8pEc9kJ9kbVnX53rifJI8XdtqC7y6yhjx3l5cE\nM9eurAldQxzH0DUgjJY/13vH5J7b3hB7TqnC0sSt0+n8PICff+iXP/bQf591S/wPAH4VQAzgi51O\n54vdbvdP5v2cXiIZ+7hhe7uNo6MR9+cEHtm0h8eTpZ8XxTGGEw+bKzUhP5vCTGgNt+8NcPTU2sI/\n+87tHgCgomtC1+AmtKOT3nT5c4hi2G4AyxC7BoBQ4/pDN9fnjm0fa62q8DWYya65dbeP0F3MOT9O\n9odnezg6Ggl7L2nL/c7d/tJKYH/sYm+jIfw51JIH8fatU+ytLq7CHfcmAMh7JHIdoU+qdPePR0s/\n915S1IiDUOgasjNivPRzb946BQDUTF3wGshzODxZvob9E/JdGBrSPyvivaS+TKd9O9dnHfdtVEwd\n46GNSU6J8jzQki703f0Brmws7t6cJgIMziTfmcKyhp0ltg/3TydYa1VxcjIW9vMBIEoYAfcPl++N\ngyPyswMvEPwcyDtx9/4QT28vPivfeofsjUZFF3pWAqS70Rs6Sz+PdgZNHUKfQxTH0AD0Bsv3RhzH\nmNg+dtfr4u/PRNDh7Xd7WFvQ0buRnFMNyxC+Bj9RzDw6WR5TnfQd1CoG+oLjzPTMznFe3j+eoFE1\nMRk5mIwcYe9llNjH7B+McLS3+PNu3SWJtKWJfS/1JLa8sz/A1a3F3dW7B6S4o4WR+FiiYmKU4wx+\n9+4AAGAJ3p8isSihXJq4dbvdzwH43OyvdTqdXwHpun09ESrRHuq2odvt/h8zf/4LAD4AYG7i9t2O\nIlTJqZO0cQXzsSmvOc+MW2bSeH4zbpR2I4Pq0azlk/r2gxB+IJ4OBsxSPZavQ5YMfjrbtYRe4Poh\nPD+SwgnfTL3clgfdMozIgWI0RVl0zSJr6KXm23LmuvLQwWTYZADEpqJeze/bNZx4iU+lWCpzuwhV\nks64CX4WeSlIQRhhOPbwviuLC3IsKKJ26shSlSxA4T2SpGIIJBYROdYwkXRG6JqGejWfGrHrh4ji\nWMrdRYUllnm5jdPZbBl3OKUpLv8+xpKEKIoIpJwMHWytin8ni8ygivZwo0jPqRzKkv2RnLsLIKJI\neWb2U+rse0yc5HcB/JXkf//nAH5/9jc7BL/e6XS0TqdjgszAvc6+zCcfRS4eGbLrANJ5mMFk+SxR\n6uEmeMYtE4HIkbg5cuZGgPyJmyxFSaCYD83U8WGZOiqWWDoxfbbLTCtlmlVmXm55ZtzkJPONQomb\nnCC9yAUsw8MNyIoJ4xzvpCyhGIB4ZuVJmOI4xmjqCbcCIGsonrgJn+3KuYbhxEMM8YU2oNjeyHzc\nRD+H/Op9Msy3s3UQe4ZlFhGyzin6mbnmDV05nnoAsJFQ7PtL5pJpQtMSnMAC2Vm1LFkgZ4SPVl38\n3sgrpOQHIRwvxKrgOVygmKiVrII8jVXziJMMJh4sU5fyXtYqZq5zKhNae2+Jk/xLAEan0/kSgL8F\n4O8DQKfT+cVOp/NKt9vtArgN4CsglgC/1e12vyJiwU8q6lUDuqblungy822xm3ylaUHD8sMWkLfB\nDZ2IW+Q5ZGR4uFE06yY8f7kwSObhJid5BPIl8xM7EN7ZAGaVmBZ/H+kFLGGYl6rg5em4TZ0AVcuA\naYhVKEu7Xc7yap0sS4IiF3BWtZQlsZ2nqCGnqwDQADlYGiATEaUYKxIKCq2cgSFAlMwqli5cOS9v\n4pb6fkpIYItI0GfiJLIEa5av4XjgwNA14ap1AAnUwyheWtWXZd8C5E/cppLUb4H8neD0OUi4N2oV\nAxVLT+OlebDdUAqDCSD3hobl52XKFJHwHIp13GhB/vw6boOJh9WmeIYEQMTWHC9cem/ILESrANOO\nTqT9f+6MX/9HM//773Gs67sOmqahWTfP1XPE0HW0mxX0lxx0gDz1ISC5eAqYucq4eGa93NZa86vD\nMrsKzQIBERGrER+I5KXn0cBRRuJWq5ho1sxcqpITR45KGu0QFFKVFPxeFqFKUmVY0YUV09BRrRi5\nutFS90bdSlUrF/kCDSUmLPQzhzmqyI4rXv0WmFFTXBIQyTyvi7yX8jpu+bufvZGLtVZVqP8oxWyx\nbdH3nXXcZNDzLLj+GEEYLSxgyey4tXO+lzITFk3TsNqsLN2f1M5DRpCuaxoaNXN5AmvLK3IVKaxk\n54TYWCKvHUCc6Dc8vSdHEKRWMRBGMYIwWihCIyvGVoUnzzL8CUarbuVqJY8kGcoChFfcH7tLKxJS\nE7dqzoqh1I5bvm6XTKpkXs+sKIqJuaaENWSmlXkryHIqVO1GZWnlFKAmoxKq2IWk+JNKtuBnYZk6\nTEPLlTyObB+6pkkJiFq1fAUm2R03IB9FEJCTuLXqFnRNy/Ve2m4g3LuMrgHI03GTQ68HilX0bVkz\nbjnP6ygiBsOizX0p0u9jSWFDNlVy9mfMXYNLzykZhS4TuqYt9Tik31dLwhoAsu+HE2+hn5zMMwIg\nfmTLLJZkzvo1quSdzHN3DSYeahVjqUl2UdSrBgxdy9GBJfoNsr4LagmwzMstS9yezI5bmbgpxGqz\nkvovLUJaRZbwUq21qvD8aGmgfjp00KyZwmeqABIk226w1LxT5uWX18stu/zEfxeZWMziQ3/qBogh\niW6Sc+ZQZscNIO/6xPYX+smFUQTbDaQkCjTQzDXjJok+qyXCA3k5+q26Ka2rkKfjRos7Mi6/Vs6i\nxnAiL2HRdQ3thpX+OxfB9kIpBaZ61YRpaEvPiKHEYh9NSPMa/ALiZ/2of+Sy92GQBPEy2AlAdmYP\nlgh8yaRK5hXlkFl01DQNrYaVo+Mm7/4ESOcojOKFCT09I1YldVc2V2qw3XDh/qBxhoxZv1rVgIZ8\nImeDsSulGK9p5Kxc1pjIRnDk7E967iyLb2V2YVWgTNwUglYZlvGAhxKrp2s5koU4jnEydIQbuVLU\nqyZiLBfEUEKVXBIIUIEUGRcwHfDuLRvwltjZyJuwyE7c2s0KYiwO1Mc2SWClUF50DdWKkVucxNA1\nVEzxx2febvR46kmjeTTrFhwvXFpgOk78srYEG8oCMxTBJcHISJL5NsVqs7I0cQvCCH4QCacHAoSK\ntb1WTwU35mE0SSrIEp6Drmuo5dwbjhtAA4RX9IFMGGQR6L0mWrSHIs/9CcwUd6oyqHH5jKdtiYwV\ngKiuLvs+xraPqmXAknBWAhkjaNEeHUjuuG3kUNhMqZIS7k+qNLqsIx5GEUZTX0riBpB4dWlsK2nG\njiJlEC15FqMpEXurSmhMqECZuCkErYYuo96MaBVZwstNuc2LLAHGtg/Pj5b6erEio94sqRhKktgG\nsuB/kLNCJOPArVYMNGvm8sTNpnMC8ma7llIlJa4BmNkbC76PlA4m6dBv5Lj86DpadUvKcHWe+c8g\njDBxAmnVwrzd6JOBg0bVlFJNz2tLkAZlkpLYlWYFrhfCXbA/0gBZAlUSAHbXG5g4wcIgWWbHDSDv\nZR71W9sLUa0YcjrBORIFepbKokpSMaDlZ7YKGfzF3we942UwdwByhy5jEE3sQNqdAWQJwKKYSjZV\nciO1s5n/TsiysaHIc2+Mpn6iPCtnb7QbpODnB/PfB5kjOEDGIFoWz5DCp5w7XAXKxE0hUprFEjn+\n4dSHhkwCWSTWUqrH/DWcJOp+sjputAq5bI5GZsctlaAfLFYypIfxlqRnsd6uLhXlkEk3yWbccnbc\npM24LVelGkkO0vPQFIMwwsnAlSI1TtfgB9HCyy+dG5HYcZv9OWchjmMcDxwp3TYgv6KjTDVFYKai\nv6CgILuzsbNO3rWDBebBQ9kJbKOS0hAXwfHkiLQA5J3wg2hhEi07caOfu8wLtS/JLwvI7oFlVEmq\n0iurAJsWVxZSBH1pyQqQ7ftFHbeh5GQhVUVe0HGTOeMGkBhpWSIvy8ONIo8lgOzELS+te2T7T6wV\nAFAmbkpBL9Vl1JvR1EOzbgmXlgayiuGii4cmM7IO/K018rnL6D8TiRx9+m87WSJBT3+fHs6isd6u\nwXaDhQmDzAOXBlnLDn2Zss5AvkNf5uwnQKg/UzeA688PDE9HLqI4lpa45RFJkT1Y3cwRGI5tH64f\nSjsjaBV72RkhPWFJCl3DBeclVe+rSaBKAlnidtib/yyGUw9VS7zowOwagjBaaCUTRhEGYzlGx0A+\nSpqqjtsyquRgLM+rKjW/znl3rbclJW7JnhvPObODkMzSyxgzoEgLKwv2p3yqJO24LaBKOnJHDZp1\nC64Xwltwd9GGgQyvRwDYTuK6e8eTBWtQ0/08Hsw/K10/hOdHT+x8G1AmbkqxkqOtH8dEFUtaBTnH\nxZMmbpK6TBc2mwAWb3AgOwhlDJqvt6vQtOUdt5Ohg5VmZaG0LO86gMXUG/pdifbsAsgskaEvFz+Y\n2D6MZNZFBvLsDZmznwBwZaeFOAbuHI3n/pmjHjX3lVvFzkUZlSgUAywOhrL5NjkJ7N5GAwBw/3R+\nlwkgz0nT5AVEqzmKbfQck5XM766TZ3Gw4FmMpr7UQGSHrmFB8rh/PIUXRLi6K0fqmz7fRck8PUdl\niZPUKgaqlrHUC7U/cbHWkuNVtZt2YBcXNU5HRIhC1nxZe4ni6VSiFQBFWlhZcm/IUuAFgI3VHFRJ\niXYAQFZYXhRHZAVgOXvj+oVVAMCNe8Mca5Bzh+9uLN8b9P5slYlbiTzIqJKL58umboAdSUHAWo41\nHEumWFzcIkHAvZPFidvJ0EW7YUkZIDUNHWut6sKOWxTHOB062JTUbQOyKvKiA7c3lldF1jUNq63K\n0kBkbPtoSprrArJkYbiIKinZNPNq4i3z7v3R3D9zmASNtAsiGnnmJajksqwEdpt2eBYEyDKFSQCS\niK00LNw/WZK4TTy06xZ0XdJ7mQaG87+Pm/skULl2YUXKGnaWfB/UG0lWsQ/IkoXDBXTNm/fJc3ha\n0nPIErf5Z3YmTiLnWWiallrqzEMUke9D1izR1modGrIi0plriGOcDt30PJGB1hJ6u+wuEzArTrI4\nYWk3LSlzl0AmhLO44yZPIRvI1/Xrp4qOcvbG9Ytk37+9IHGj56iss2pnjcSWi9gJKWOlpEqWyIM8\n4iS0wry32ZCzhmYFGrAwUJdNldxYqaFqGbh3PD8IiNOkSd7Fs7laQ2/kzh2uHk08BGEsdQ3rOfjx\n9Ltak3Tgrrer6I8Xz6+MbV/qBbyMdgMAw4nceSbaKViUuNFqP70gRGNzZfkFLJsqSbsri4J0SkWR\nlbgBpOt2NLDhB/PpP8OpL02sBsjXcbt5fwhd0/DUTkvKGjZWqjB0bW4wYruJN5JEM9k8dE26b2SZ\n6+btuLXqljSGBEDOy+HUn39vTD3EMbAm6b20TB0bK9WFhZVRsj6ZRUfacZvn3ZUKWkmcccslTjL1\npHV4APJ9rDYrC4ttY9tHrWIsNEznQTZnN38NQ8ndrpVmBVurNdzcH871CR5MPNSrcmymAJIYtxvW\nwnngJ93DDSgTN6XIQwdLE7cNOYGhaehoNay0+nIWToYOKqYujYqlaxr2Nhu4fzqd69s1nPrwg0hq\n0rS1UkMcz09iT5KDWGbVMrUEWHDo98ceDF2Tljitt4gXzmjOOxHFsTTja4piHTc5F8+FrQYsU8e7\nBwsSN8lUSXoBL+oEy6ZK7qzVoAE4OF3ecZNV3AGAvc0m4ng+7cUPQthuIDVhWUmq6fPO7DCKcOtg\njEvbTWnBiKHr2Fqrz02ahgoCkTxUyXfuj2DoGi5vy0lgt5fMRsdxjN7IlTbfRrG25J1IhUkkddwA\n8n30Ru7cmabTdDZbfsdtXrFNtiAHAFimgXrVnFtYcbwAnh9J7UYD5Nw+HTlzi58TR27hcz1Hxy2b\ncZP3Xl6/uIKx7c8tKgwmcpNogBSZTgbOwsIKUCZuJXLCNHQ0a/MPGSALlnYlUbEAcvEsVJUcONhc\nrUmVSr242UQQRjiaM0QqWxFr9rPnBcmy1TWBmUHzJTNua62qtO+DzoP05rwTU4f4p8m8eAgNc9ls\nF5mzk0U3MXQdV3ZauHs0mavqeNi3UbUMaYFAro6bZKqkZRpLK/onkqmSAHAhYR3Mo0vKVpQElvtE\n0Xfl2gU5XSaK3fU6xrZ/pmCMbLlzgBRWqhVjbhc2CCPcPiQJrKyZqlbdQq1izE3cbDeE64fKErd5\n9PZsJll+B3Tes5A9pw5kVLNlVEmZHTdgsdeibPNtio12DUEYz38WdiD1OeSZcetPyDywrIIfAFy/\nmMy53X2ULhmEEcYSfeQodtcbCKN4blyXddxKqmSJnFhpVhZ23A4kd9wAwnF2vPBMCXjbDTBxAqkH\nPjAz5zZHoETFxbOZKhDNSdwUdBWWiZNEERGrkRmMLDMCl2keSqFrGjHwXEJ5aTXkzSoAZM4tjGLc\nPX5UoCSOYxz1bWyvyStqrOUQzVFB9aAV/XkKm8cSPdwo6Bm4P0eUQ7aHG0A8swxdm3tmv0PpgZLm\nuigWURVlK2sCZLZrN+n6nUWDundMElhZNEm6BmJG7py5hp5EEadZrLUXC3zR91LmOugM/Lwu7Klk\nNWQgO3/miZPIFuSgWG1WMJ76CKNHi22yVQwpNhYU3PwgguuHUjuP6YzbEqrkSqMibR4YyObczhIo\nyXzk5HfcgPl7Y2SXHbcSBbHarCw0rbzfm6JWkVfRB7IL5SzVuBMFnS6AdNwAYH9ONZ0mUzKpHo9D\nx61eNVGvGujNmXEbTcnsmczq7VqbfPY8yuiYDplLvoDbDWuxj9vUkz5QTOfc3jljzm1k+3C8UJp6\nIDArmrNgXmIqzxSegnb8zxJAIB5uttRuGzDbcTu7uEMpLytNec9B0zSsLKjop8Ike3ITt0XKkpkx\nvdz9ubNehxdEZ1rJZPNtkhPYtTpcPzyTUk3Prw3pHbfkvJyjukoTOpkB6jLBGhU0/9aSGbdxIsgh\nw492FivNCmKc3flT0Y0GZmbMzoglpgo6j42aiVrFWEKV9KQnTVd3WzB0De/cfzRxU/VdLFPhLTtu\nJQpj0ZxbFMU4OLWxt9GQSlPMLp5Hg0MVnS4AuLi12BKAJk0yg8NlJtwq6JoAqZbN63apqCKnqlhL\nO26SL+BGBVP37KIGmWcKpQbpQCascOvg0Y5bNt8mL3EDyHvZG7lz5z9Hto9G1ZQ26A7MzjSdkSjY\nPjw/kr4vtlbrMA1triWAio4bkFGxzuryvLM/gmnouLTdlLqGVJjjjLMq8zeUHBBtzBetuX1I9stT\nkqwAKBYJlFCBJ1lWABTLqJK0ILomSXYdmBUQWjJqIPEer1jEGmGe96ZKqiRwdiFatvk2RRpLnFFw\nox1JmaMGAEnS5wmk2Ik/qSwrAArLNLDePlutOy1oKJhxA+bvjXEpTlKiKNLE7YzD7nRIBipl0iSB\nzMfjrIqhCtEBgJhwm4aG/TnVdCUzbivLO24VS5cqygEQuuTECeB6j9LS+iMv/TMyfz75WXM6boou\nnvYCeemRouCUHvpnGXgeSbYCoNhYqSKK47lULNmeXcCs/Pujz+FEsocbha5r2F1vYP9kembSlFZO\nJQcCK80K/ICYCT+Mw/4Uu+t1qUk0kBWwTs54L+ldIn1vLKDn0aLPliTRHopFAiXHffnFPmA5VVJF\nx40+h3kzhydDB6ahSz8nWnVrOVVS8r2R2Sw9+n08DlTJSepnJzeO2GhXMXWDM0dgVCWwdB2DifcI\ndZUWOmTPoO4u6UaPbCL21qjK/T5kokzcFGORfO395BDelZy4ZVTJRw86FTN2ABGC2Nto4N7x2UHZ\nycBB1TKkJk3VioFW3ZpLSzsZEDsCmd1PIDvIFlWppHbcloiTqJB1BrLA86wKrmzzbYpaxUS9ap5Z\nTacBqyyPRYrNBV5uURxjPPWlP4edjfkdN9kebrPY22jA8cIzqYoqZruA+QIlthvAdsPU0kMm6LM+\ny8NMFQVpET2vN0oSBclB+qKOW2bVIXd/ri9grJBfl6sCDJBzarVZmRucUv9R2XdXq2GlHYyHkSYs\nkgufa6mP2qPfh6q9QffnWfPyqmb9Mrrm/ARWNlUSIIWNOH60A6pi9AQAGjULrbq1kCrZkuhJqwJl\n4qYYKwva+qmi5Ibci2cRR1+2HcEsLmw24frhmQfNydDBhoKLZ3OlhpPBozK+jqdGpAXIEvWzDhoV\nCmWWSRLYedQfZR23Bd1old4rG+3qmd1Hahgve29sLOgET50AURxLfw7UEuCs7ooKDzeK7QXqeUMF\nM25AFuw8/E6cpjNV8p9DJVEyPYvWfTwgSZPs/bnIEuB06GKjLf+8XpS4HfRsGLomda4LIOdls3Z2\ncQcgnZ+1VkX6s9hZr+P4DNlzPyAzgLKfA0DOYy+IzmSLjG0fFVOXZpNBsaigoKrj1m5YqFj6mUyN\nsSLKaCZQcn40xdl1PLw/VI3hACSGPh44ZwrWqGCsyEaZuCnG6oLglMpeyw4MaYWqfwa1YP9kitVW\nBXUFbWQ65/YwXTJVtlQQGO5u1BGE0SOVMppEy55nAjIRhntn0Ebp4adibmNu4uYopkpOzm/IHCDP\n+izq6v7JFBVLx4bk93IRhVeVBw21BDgrSFdFpwYWB+ojRR03Sgl92LqECgrJFsPI1lHDydB5YPYx\nimPsn0ywt9GQqhYHkOJRxdQfoecFYYThRK7yLcXmag2Grp0panXUt7G1Vpf+HAAk6pb2I4FhHBMV\nYJleWRQXt4jP4cNz4odJV1aW1+QsMhPuR+OZie1Lp0kCWUHhLCGl4cSDrsntfgJExGhrtX5mYYUy\nVmSvYT1Xx03+e7k2Ryn7eOhA0+THMgCws5ZYAjz0fQRhBNsNpH8XslEmbopBjRLPqmSnVMl1yTNu\nrQpMQ3/kwCfdLwcXFHTbgPkCJZQnvqWgMnNlh5jF3j18UIzizhH578s7csxkZ0EVNu8dn9Vxky8t\nDRCaheMRQ+OHoWpWgQbgZ9HiVHbczqKORlGM+6dTEiBLrqQvMuFWqYg1zxJA1YwbAGxTCtIZFMHB\nxEetYkiv6M+b9+slAZIKqiRAErcwenD28XTowPOj1F5FJjRNw856HQcPWQL0xy5iqHkOpqHj0lYT\ndw7HDyRNUyfA2Pal0yQpLm01EYTxI+/E2PYRRrGSzsZTyd10+6G7az+5Ty9syhXMAebTFKMoxnDq\nSe8yAYnHoGWcWWQ67BP1W9lnNkD258QJMHUevENTkRbZM24L5uzoe6qi2zVv9OJ06GC9XZU+Dwxk\nrLWH9+d3g6IkUCZuynFxq4FqxcB37g4e+b2DUzXdLtPQ8fSFNm4fjh8I1A9Op4gB7Ck48IHZTtOD\nCcv9pNu1peASvrydXH5HD15+9DK8si0/cVsk1NIfu6hWDOnvxCKltElKlZS7Bhogn+WhpkqAAcg6\nKLPP4njowA+iNMmWCdrJOj2jept23JRUss+2BDgeOGjWTGlG6LNY2HGbekreh3kUQZVUSSBLlGfZ\nAbTzpCJIB8izcL0H5fhp0K7qOTy114YXRNifKXSpmm+juJTcCw8XHQeKCm0AcGWHKHg+nLjRNdHC\nqExcnpM8vnswgudHuH5RrsooMOvv92BBYeoEGE196WJSFJvpnNujyTyggio5v+OWvRPyCzxnedMG\nYYTeyFWSOALZ3fXomU3OThWzfjJRJm6KYeg6nrm4gv2T6QMiDH4Q4mTgYE9yt43iuUuriGPgxn7m\nt6Fyvg0gnUVd0x6hCN7YJ0mtTDNXCpq43Tl6cA204yZb5hsg78TuRgP7p48KtfRGbirXLxOLjMAH\nE7DiS1cAABf6SURBVB9Vy4Blyu1sXNhsoloxzjTvpF0GFR23jOoxEyDTKraCYKhRNdGommcaT9P3\nVLaAETDjhzNz+VEPNxU0SYAEQxoelcGPohijqa+GOkspgg99HylVUmHHDXgwMMy6K2rO7ExqO3sW\nqtTiKOi98O5B5rVIRXS2FQXpNCm6+9C9Qem0Kt4Jejc9kridqHsnaNfv1sGDvpfdW30AQOepdelr\nAMh7+bC/32FfDYOJIlN+ffCsunUwhqFr0pOW7bU6DF3D3TMslu4ejbG1WkOtIr/YRuOV2Zng/shF\nHKvp+AGzd9eDZzbdr5cVFORlokzczgHPXV4DgAe6boc9GzHUBGQA8OzlVQDA23eyNdxPq7dq1mCZ\nOnbW69g/njyQsNy8N4QG4NoFuWauALlg61UTdx/quN05JAedilk/IBFq8cIHEqfh1CP0HwXBCKWM\nvn7z9IFfD8II908nSt4JXddwba+N+yfTB+gmcRyje6uPZs1UQs87q+NGOxsXFTwHTdNw7eIKDnv2\nI1LbmdGx/KLGWUH6aEo83FR8DwBhB6yvVB/puO2fThHFsZK9kVIEH6ro046biu4KkEntz9JG76Xv\npaqO26MUpJ4i42uKq8m7/879LFlQ3nGjidtDQfKdQ1rwkx8Y1qsmdtbquH04fuC9vHc8RdUylIiT\n7G02YBo6bj2UPL5xqwcAeF5h4gY8eFbROfUdyWJvFGd1xG03wLv3R7h2YQXVitzCp2nouLzdwu3D\n8QOCNcOph+HUT99Z2VhtVaDhwfvzRJEnLsU8ejstcpSJW4nCeC5Jmt6aTZoUd7uevZSs4e75rQEg\nSeLECdJKWRTFuHl/hAtbTSVJk6ZpuLzdxP3TKbxklmcwIQfdFQXzbRQXzxAouX1ATW3lr+MD1zfR\nrJn4w9fvPzA7cu94giCMpRvrUly7uIIYwLv3s67b3aMJeiMXL13fVCI8sJ6qc2UXT1bFVnP5PXOR\nFC1u3HuQUv3O/SHW21UlQ+a7Z9BNVFoBUGyv1tEfufCD7L2kz+X6RfnFHYBUcF0vfMDGpTd0Ua+a\nyoo7Z1MlJ9A0dQW/s7qwdKZGRaIAEPq6rmlpEQOYsepQ1HHbWKmiVjEeoUqmRuSK7o4rOy2MbT8N\nkuks7oVN+bO4AGGLXNpu4u7RJE0WwijCm7f72F2vK+vCnuUxeKBIM4DiLEuAN2/3EcUxnr+6pmQN\nV/daCMLoAfGee0mXSUUxASAJ5EqzcnbipuiMSC0BHkrc7hyOoQHKklhZKBO3c8D1iyvQNQ1v3emn\nv6Y6aWo3KtjbaODtu4NUpWz/ZArL1JVtLuBRgZJ7xxO4XojrCrptFJd3WojjrKty5xyqMjQhmJ3b\noPSTp3bkJ02WqeNj79/FcOI90HVLAxEFySOA9HufpfB+88YJAODl65tK1nCWIfn+yQSGrikLDK9f\nTDrid7Pn0B+76I89XFWURG+vPVrFpjQ9VZVTgHSaYjwo1nIzodOqStzOmpk4HbnKaJJAFvQ8QJU8\nmWJ7rQ7LVHOV7z4GVMmKZeDiVgO3Dkfp3XXUt6FBjZIiQAp+F7dIwW+2u3H7cIxmzVT2LK48NGN2\nNLARhJGyAhNAklTCzCDvxK2DMRwvxPNX1XTbgLNtQ2gSJ9teieKsGbdvv0s6jy8o6jzSu2G2qEG7\nwiqTlbV2Fb2xm3aCTxQqEVPsrtdxPKP8Gscx7hyNsbNel979lI0ycTsH1Comntpt4Z39UdrlUeXh\nNotnL63C8ULcOSJUi/unU+yuq5FTpsgUFcnhQgN2VQEZMCNQklx+aTtdYceNUhFnBUpuKU6aPvGB\nCwCAL33zfrYG2vVTkDwCWcIyO+f2jbdPoAF48fqGkjU0ayYsU087bnEc497xFDvrdSWKWED2/s92\n3N5RSJMESID8sCXAyXl03JIE8ngmKLtxb5hSg1Rg9yFDcmK+HSgL0AFSXFlrVdKKPqVSq6JJAiQg\nMw39gc7G6ciFaWhoKfRGurrbhudH6RzoYd/GWrsqfQ53Fhe3mgijON0frhfisGfj8nZLmbnvw4mb\nShEKCsrGoAwRSpPsPKWmywQAu3M6boauKTur2nWibjk74/bGrR5MQ8MzCcNJNp46Y/4zTdwUzOtT\nbLSr8IMoNWE/oQJGCpsCO+uJJUDys/tjDxMnUBrXyUKZuJ0Tnru8hjCK8XZCVbzfm0JP1JHUrYEc\nJt3bfRz0bLh+qESJahZXkqSEBuqqKVBAphxJBUlSKwCFB93eRgOa9qBK2a2DEepVQ4m6JkASgotb\nTXztrSM4XpCuQQNweUfNs1hvV7HWquBmksBPnQBv3Rng6QsrShQEAVJNX29nvnb9sQfbDZRWsVt1\nC3sbDdzYH6bm8Ol82wU1iRtAKEizlgDHCq0AKLZXH6ymu36IO0cTXN1rKUukH6Zi9RQrSlJsrdZx\nOnQJHUqxMAkA6GdYApyOiMy3CmoeBZ1zu3FvANcP0Ru6yubbKC4/xBa5czxGDCil2NP7k54N+4pn\nHoHs33vrkKzhW++onW8DSEJg6NoDJtwHp3bi+6fmjCBebrX0jBzbPm4fjPHMxVXpliUUKY14NnE7\nGkPT1J4TD3u5nSi0eKJI2QFJcUelUrhslInbOeGlpHvwzRuElnZwOk1k4dV9JS9eI2t4/eZpSo9T\nSW8ASPt+pVnB6++cIo5j3Lg3RMXSlVaHLm03Yegavv1uD3Ec483bfdQqhjJuPEC6G5e3W7ixP4Tj\nEfPn+ydTXNlpKwuINE3Dh9+3hSCM8e13yLO4dTjGzkZDiRoVxbULK+iPPRz1bbx28wRRHOMDirpt\nFBvtKoYTD0EY4VvvkL3xjMJiAv15thumwRgNzq7uqVsH7TTRpCk131Z4AaeWAMnPfvf+CFEc4/oF\nNVVsYLbj9nDipq7jBhAWQBSTgh8Vt1I9aL+zVoftEt+0IIwwHHvpXKgqvJDcU6/fPCXnNtQW+wDg\nYnJH0X2ZBoYKE7fNlRq2Vmv41rs9BGGk1AqAIk3cDsbw/BBv3u7j8nZLmWgPQISttlZrROQtjjF1\nfIxtX+kdDhAq4NQNMHV8fOudU8TI3lUVqFgGLmw1cPtgjCiKE7bIBDvrDaXd6PWH7IVOBg5adUsp\nRZGK0tAzW6U3r2yUids5oXNlDZap45s3TjBxfIymvlJREIBUqS5sNvDGrR6+9tYRAOCla2oDZE3T\n8OLTGxhOPLz6xiHuHE3w3OU1ZVUygKhzvXRtA7cPx/iDb97H8cDB9zy3rZQyCgAffJYkTa/dOCX0\nVagbcqd4+ZktAMDX3z7B8cCB7Qa4qoiqSfHBZ8kavvLtA3zl24cAgO97fkfpGmjFsD928bXvHAMA\nPvTcltI1XE/oNW/eIkn0zf1EmESBBD5FOtt1SmYFbh+OlHm4UdC5Jeond0PxfBtAlNIqpp7OdlFB\nDlXm2xQfTt7Br755hFffOISha/jAM2pmPynoPXVzf4jB2EMM9Qnsxa0mNldqeO3GKf70TXJ30XND\nFZ69tIqKpeNP3jhEHMfnQrHXNA0vP7MJ2w3w1u0+3rjVQ7VipAqkKlCvmri83cKbt/v4k+4h/CBK\nC9MqcXWvjbHt48b+MA3WdxXNJFPQhPnb7/bwanJ3ffh920rXcHW3DdcPcdCb4qhvY+IEysU4KIX8\ndOTA8QIcDxylxT4gE6WhwmJ3zmF/ygJzdNzpdD7V6XQOO53OZ+f8/k93Op1XO53Olzudzt9gX+J3\nJyqWgeefWsfd4wn+8DUyU6Q6cQOAl65twvMjvP5OD3sbDaUUKIoXr5GK1K/+7psAgD/3wYvK1/DR\nF3YBAL/+ebKGj7+4q3wN35MEZX/21nEmTKJIiILi+oUVtOoWvvH2cVpJVllBBoDv62zDNDR86Rv7\n+MbbJ7i01VTeVaAXz/7JFK/dPMXOel35/nz5+iYMXcMX/vQu/vTNIwwmntLqLZBdfof9Kf7szWP0\nx166V1RhpVnBaquC126eYjjx8I23SSJ9TWHipmsaruwQqe17x5M0GFBNlXz+6jrqVRN/9Np93DoY\n4/1Pb6ClwIx9Fh95gRRRfu9P7uAbiXCQSoo/kCUsUzfAH712H626lSolq0KtYuJ737eNw76Nt+8O\niWKdpl6xjhbbfuML38Hp0MUrL+4pLXwCwKc+dBFhFOM3Pv8WAPUFYAD45MtkRvuLX7uHg4Qep0pt\nleITL+0BAH7n1dv4xo0TXNhsKGUPAUjn6b70zX383qt3AGSxhSrQmOEr3zrAf/r6PoIwUr6GS1tN\ntOoWvvz6AXojF99+t0dGTxTOZ8sC0+7udDrPAPi7AP5gzu83AfyPAH4YwA8A+DudTkf9Tn7MQelf\n//fn34Kha/hkIg6hEi/OHLDnUSUDgPc/TX7u2PbRbljKNzhAuimWqcPxQrQblvIAGSAzZuvtKr7x\n9nEqEHJVkRAFha5r+MD1DfTHHn7t90gSS30HVaFRs/DyM1s46BGFtI++oLbbBgDvS/7Nn/sP34Lr\nhfjQs1vKBAcoNldr+Nj7d3HveIJ/8VtvQAPwF1+5qnQNdC7ij18/wG9/5RYA4Ie/77LSNWiahs++\n8jRcP8Q/+Vdfwxu3+njx2obymaYfe+Uq4hj4F//Pt/GFr97FSrOinJ5nGjo+9OxmOvR/Hnvj2oUV\ndK6s4fWbp/jN3/8OahUDP/jhS8rX8XLSaQyjGC8/o8Yq5GG8kgTqv/Z7b+I7dwa4stNSNs9E8cLV\nNVQsPaWC/eD3qP8uXnlxD1XLwMQJUDF15XcGQOKIrdUavvztA/z7L90EoMYLdhaXtlt49vIqvnNn\nAD+I8NEXdpXfG594aQ8bK1X83qt38MVv3MPmCrlHVOKp3TZefmYTb9zq49996QYsU8cPKD4jKpaB\nH/noFUzdAP/o176KwcTDp7/3itJZXFlgLcvsA/gJAIM5v/8xAK92u91Bt9u1QRK8TzD+rO9aUIpL\nDODHPn71XFq4nafW0rm6l66ppdxQrLWqaUflkx+4oHTOj6JeNdNA4CPP75zLGjRNw4ee3cLECXBz\nf4iPvrCjVCCFglZwBxMPn/n4U3jfFfWX8MdnLhrVHR6A0K7+s5f2MEr8Bc+jmAAAn/k4SdSmboCP\nvX9XqUAKQGwq/twHL+L24Rg37g3x0vUN5WsASEV/a7WGWwdjVEwdf/1HOsrX8KFnt/Ds5VW8fXeI\nIIzwMz/SUebhNosPv48ka6ah4XueU0vDovjRjz0FAHC8EP/FJ64pnWeieP7qemqD8CHFNEmK91/d\nwGqrgncPRrAsHT/7meeVr8EyDbz/Kil+Pnd5VTlDAgAaNTNlqXSeWldmTzELXdPw/R+8CM+PcNCz\n8ZmPPaW8sAIAn5phDH1EMcUfIAnLT37qGQRhBD+I8KMfe+pc4pm//P3XAAC2G+ITL+0pExebxQ99\n+DIaVRNHfQebK1XlhU9ZYLp1ut3uFAA6nbmX5x6Ao5n/PgSwsJ20vt6AqXB4sgi2t+V0Pba323ju\nyhrCMMbP/aWXlA6PzuIj79/Fa2+f4BMfvqxUhGIWP/SRK/jNL7yJH/+h92H7nMwR/8oPd3DvZIqf\n+PT7pH3ny/DnP/40fv/P7uJjL+7hF//rjyw8cGWt8Qc/WsO//dJNfN/zO/jvfuJl5RVDAPj0WgO/\n/vm3cGGriZc66hM3APiF/+rDeOeX/z84XohXPnQZxjlcftvbbXzi5Yv48uv7+JnPvngu7+Xf/env\nhR/F+PJr+/ipP//80jXIWuPPfvZF/G+/9lX89I++gBefUx8QAcDf/MsfwN//Z1/C93/oMn7kE9fP\nZQ0/sFrHb3zhTXzwuW1cvaKeGQAAP7TZwn/88i2EYYSf+tEXziVQB4CPvbiHr791hE995Ck0avMp\nozL3zWdeuYbf/MKb+Hs/8xF85P170n7OInz6o0/ha985xk+e4931V//C8/j6d07wY5+8dm5r+Es/\n8Cx++8vv4oVrm/hvf/JDMM6hC/uZ72/g33zxBrbX6vjgC4vfB1nP6bOfauEPv3WAw9MpfvzT70NV\ncRcYIP+2T37wHv74tfv4qz+y/N6QhZ/4wWfxq7/9Bv6bH38Zly+qL0LLgEYlfeeh0+n8PICff+iX\n/0G32/2dTqfzKwD+dbfb/Q8P/Z2/BuAj3W737yT//T8DuNXtdv/5vJ9zdDRavJBzwvZ2G0dHo+V/\nkBFhFCGOcS4VEQrHC+D5EVYUih48jDiO4QeRcprJ44g7R2Nc2GwsnFOQ/V7GcXwuCdss+mMXlqmj\nuSAgk42p48MPY6WCIA/D9UMMJ57yOaJZRHGM/shd6sMj+7087hOJ7/N8Nw97U2ysqFUAfhieH8Iw\nNOWzTLOgxtPn+RxcP4Tnh2gvqObLfiejKMbUDZTPGs4ijmOcDJ1zmVF/3DC2fdSrxrnujf7YhWno\nC98J2e9lEEYIo/hckjYKPwgxmHjn+l5GcYzjgaOcWs+L7e323EtuaXul2+1+DsDnCv7MeyBdN4pL\nAP644Ge8J3CehwtFrWKidn5xKQBCEyyTNgLVQhxn4byTNgDnQr96GIuq+KpQtYxzTdoAQkNSaZ46\nD6o8DRdhR7HE+Fl4HM7K80zYKKqWca6BKUDmgs8zaQOoh9j5743HAef9XQCPx91lGjrOm8Rmmca5\nv5e6pj1xSdsyyOLFfRnA5zqdzhqAAGS+7W9L+lklSpQoUaJEiRIlSpQo8V0NVlXJv9jpdP5fAD8K\n4Jc6nc7vJr/+i51O55VEkOQXAfwOgM8D+J+63e48IZMSJUqUKFGiRIkSJUqUKLEAS2fcVOG9OuNW\nogQLyveyxOOI8r0s8bihfCdLPI4o38sSi7Boxu38SeolSpQoUaJEiRIlSpQoUWIhysStRIkSJUqU\nKFGiRIkSJR5zPDZUyRIlSpQoUaJEiRIlSpQocTbKjluJEiVKlChRokSJEiVKPOYoE7cSJUqUKFGi\nRIkSJUqUeMxRJm4lSpQoUaJEiRIlSpQo8ZijTNxKlChRokSJEiVKlChR4jFHmbiVKFGiRIkSJUqU\nKFGixGOOMnErUaJEiRIlSpQoUaJEiccc5nkv4HFGp9P5ZQAfBxAD+IVut/vqOS+pxHsInU7nJQD/\nHsAvd7vd/73T6VwB8H8BMADsA/jr3W7X7XQ6Pw3gbwOIAPzzbrf7f57bokt816PT6fxjAN8Pcn/8\nEoBXUb6XJc4JnU6nAeBXAOwCqAH4hwC+jvKdLPEYoNPp1AG8BvJefgHle1mCE2XHbQ46nf+/vbsJ\ntaoKwzj+vzPlilaTvgicxBvRSJG+sHvFwLSPO9BqEBilOKlJhaM+UEdiFKI004lY0CyKoqJRUSk5\nqJE8EZSQd6AQiYWI1W2w94WjcUA63rs3nv9vtNY6e/AOHtY5L3vvdWoKuDPJ/cA24EDHJWmMVNUk\ncJBmo5+3B3gnyVrgJ+D59ro3gIeBaeClqrppkcvVmKiqdcA97b74CLAfc6luPQ6cSDIFPAW8jZlU\nf7wG/NaOzaVGZuM23HrgA4AkJ4Ebq2p5tyVpjFwENgGzA2vTwIft+COajf5e4Lsk55JcAL4GHlzE\nOjVevgSebMe/A5OYS3UoyftJ9rXTO4BfMZPqgaq6C7gb+LhdmsZcakQ2bsPdApwdmJ9t16QFl+Sv\ndhMfNJnkYjs+A9zKf3M6vy5dc0n+TvJnO90GfIK5VA9U1TfAezSPnJlJ9cFbwMsDc3Opkdm4Xb2J\nrguQBgzLoznVgquqGZrG7cUrPjKX6kSSB4AngKNcnjczqUVXVVuBb5P8POQSc6n/xcZtuFkuv8N2\nG83LpFJX/mhfdAa4nSajV+Z0fl1aEFW1AXgV2JjkHOZSHaqq1e3BTST5nubQnPNmUh17FJipqmPA\nduB13Ct1Ddi4Dfc5sAWgqlYBs0nOd1uSxtwXwOZ2vBn4FDgOrKmqG6pqGc2z8V91VJ+uc1W1AngT\neCzJ/Av35lJdegh4BaCqbgaWYSbVsSRPJ1mT5D7gEM2pkuZSI5uYm5vruobeqqq9NF8K/wAvJPmh\n45I0JqpqNc3z8SuBS8Bp4BmaY6+XAKeA55JcqqotwE6av604mOTdLmrW9a+qdgC7gB8Hlp+l+WFi\nLrXo2jsYh2kOJlkK7AZOAEcwk+qBqtoF/AJ8hrnUiGzcJEmSJKnnfFRSkiRJknrOxk2SJEmSes7G\nTZIkSZJ6zsZNkiRJknrOxk2SJEmSes7GTZIkSZJ6zsZNkiRJknrOxk2SJEmSeu5fAsmbLKb3MXMA\nAAAASUVORK5CYII=\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fdec9add850>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "Eaypg9gKos21",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**TASK**: Adjust the parameters above to generate data with different properties.\n",
        "\n",
        "Now we pack the data into train and test batches. Note that while RNNs can in theory learn the dependencies across all inputs received so far (using an algorithm called **backpropagation through time**, or BPTT; see the Aside box below), in practice they are trained using an algorithm called **truncated BPTT** where we truncate the inputs to only the last $T$ symbols (this is the `truncated_seq_len` variable below).\n",
        "\n",
        "**QUESTIONs**: \n",
        "* What issues can you think may arise by truncating the training data in this way? \n",
        "* Despite these issues, why do you think it might be necessary to do this?"
      ]
    },
    {
      "metadata": {
        "id": "RtwMLLCNorw4",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 71
        },
        "cellView": "both",
        "outputId": "b21cacaf-acbb-4d67-dc56-a4ac6c2d1339"
      },
      "cell_type": "code",
      "source": [
        "#@title Pack truncated sequence data {run: \"auto\"}\n",
        "\n",
        "def pack_truncated_data(data, num_prev = 100):  \n",
        "    X, Y = [], []\n",
        "    for i in range(len(data) - num_prev):\n",
        "      X.append(data[i : i + num_prev])\n",
        "      Y.append(data[i + num_prev])\n",
        "    # NOTE: Keras expects input data in the shape (batch_size, truncated_seq_len, input_dim)\n",
        "    # We have only one real-valued number per time-step, so we therefore expand \n",
        "    # the last dimension from (batch_size, truncated_seq_len) to \n",
        "    # (batch_size, truncated_seq_len, 1).\n",
        "    X, Y = np.array(X)[:,:,np.newaxis], np.array(Y)[:,np.newaxis]\n",
        "    return X, Y\n",
        "\n",
        "# We only consider this many previous data points\n",
        "truncated_seq_len = 2 #@param { type: \"slider\", min:1, max:10, step:1 }\n",
        "test_split = 0.25  # Fraction of total data to keep out as test data\n",
        "\n",
        "# We use only the sin(t) values, and discard the time values\n",
        "data = sin_t_noisy\n",
        "data_len = data.shape[0]\n",
        "num_train = int(data_len * (1 - test_split))\n",
        "\n",
        "train_data = data[:num_train]\n",
        "test_data = data[num_train:]\n",
        "\n",
        "X_train, y_train = pack_truncated_data(train_data, num_prev=truncated_seq_len)\n",
        "X_test, y_test = pack_truncated_data(test_data, num_prev=truncated_seq_len)  \n",
        "\n",
        "print(\"Generated training/test data with shapes\\nX_train: {}, y_train: {}\\nX_test: {}, y_test: {}. \".format(\n",
        "    X_train.shape, y_train.shape, X_test.shape, y_test.shape))\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Generated training/test data with shapes\n",
            "X_train: (2638, 2, 1), y_train: (2638, 1)\n",
            "X_test: (878, 2, 1), y_test: (878, 1). \n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "vgVxLBp8CaN6",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**NOTE**: We reshape the training data into (batch_size, truncated_seq_len, 1) and (batch_size, 1) arrays."
      ]
    },
    {
      "metadata": {
        "id": "xiUrsPAI36M-",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Intermediate Aside: (Truncated) Backpropagation-through-Time and Vanishing and Exploding Gradients"
      ]
    },
    {
      "metadata": {
        "id": "SCJlN3O11pr1",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "RNNs model sequential data, and are designed to capture how ***outputs*** at the current time step are influenced by the ***inputs*** that came before them. This is referred to as **long-range dependencies**. At a high level, this allows the model to remember what it has seen so far in order to better contextualize what it is seeing at the moment (think about how knowing the context of the sentence or conversation can sometimes help one to better figure out the intended meaning of a misheard word or ambiguous statement). It is what makes these models so powerful, but it is also what makes them so hard to train!\n",
        "\n",
        "The most well-known algorithm for training RNNs is called **back-propagation through time (BPTT**; there are other algorithms). BPTT conceptually amounts to unrolling the computations of the RNN over time, computing the errors, and backpropagating the gradients through the unrolled graph structure.  Ideally we want to unroll the graph up to the maximum sequence length, however in practice, since sequence lengths vary and memory is limited, we only end up unrolling sequences up to some length $T$. This is called **truncated BPTT**, and is the most used variant of BPTT.\n",
        "\n",
        "At a high level, there are two main issues when using (truncated) BPTT to train RNNs:\n",
        "\n",
        "* Having shared (\"tied\") recurrent weights ($W_{hh}$) mean that **the gradient on these weights at some time step $t$ depends on all time steps up to time-step $T$**, the length of the full (truncated) sequence. This also leads to the **vanishing/exploding gradients** problem, which we'll explain below.\n",
        "\n",
        "* **Memory usage grows linearly with the total number of steps $T$ that we unroll for**, because we need to save/cache the activations at each time-step (look at the Python code above to convince yourself of this).  This matters computationally, since memory is a limited resource. It also matters statistically, because it puts a limit on the types of dependency that the model can successfully learn, by preventing it from correcting errors that stem from an input more than $T$ steps ago.\n",
        "\n",
        "**NOTE**: Think about that last statement and make sure you understand those 2 points.\n",
        "\n",
        "BPTT is very similar to the standard back-propagation algorithm. Key to understanding the BPTT algorithm is to realize that gradients on the non-recurrent weights (weights of a per time-step classifier that tries to predict the part-of-speech tag for each word for example) and recurrent weights (that transform $h_{t-1}$ into $h_t$) are computed differently:\n",
        "\n",
        "* The gradients of **non-recurrent weights** ($W_{hy}$) depend only on the error at that time-step, $E_t$.\n",
        "* The gradients of **recurrent weights** ($W_{hh}$) depend on all previous time-steps up to maximum length $T$.\n",
        "\n",
        "The first point is fairly intuitive: predictions at time-step $t$ is related to the loss of that particular prediction. \n",
        "\n",
        "The second point will be explained in more detail in the lectures (see also [this great blog post](http://www.wildml.com/2015/10/recurrent-neural-networks-tutorial-part-3-backpropagation-through-time-and-vanishing-gradients/)), but briefly, this can be summarized in these equations:\n",
        "\n",
        "1. The **current** state is a function of the **previous** state and the current input: $h_t = \\sigma(W_{hh}h_{t-1} + W_{xh}x_t)$\n",
        "2. The gradient of the loss $E_t$ at time $t$ on $W_{hh}$ is a function of the current hidden state and model predictions $\\hat{y}_t$ at time t: \n",
        "$\\frac{\\partial E_t}{\\partial W_{hh}} = \\frac{\\partial E_t}{\\partial \\hat{y}_t}\\frac{\\partial\\hat{y}_t}{\\partial h_t}\\frac{\\partial h_t}{\\partial W_{hh}}$\n",
        "3. Substituting (1) into (2) results in a **sum over all previous time-steps**:\n",
        "$\\frac{\\partial E_t}{\\partial W_{hh}} = \\sum\\limits_{k=0}^{t} \\underbrace{\\frac{\\partial E_t}{\\partial \\hat{y}_t}\\frac{\\partial\\hat{y}_t}{\\partial h_t}\\frac{\\partial h_t}{\\partial h_k}\\frac{\\partial h_k}{\\partial W_{hh}}}_\\text{product of gradient terms}$\n",
        "\n",
        "The problem is that $\\frac{\\partial h_t}{\\partial h_k} = \\Pi_j \\frac{\\partial h_j}{\\partial h_{j-1}}$ for j from $k + 1$ to $t$.  Because of this **repeated multiplicative interaction**, as the sequence length $t$ gets longer, the gradients themselves can get diminishingly small (**vanish**) or grow too large and result in numeric overflow (**explode**). This has been shown to be related to the norms of the recurrent weight matrices being less than or equal to 1. Intuitively, it works very similar to how multiplying a small number $v<1.0$ with itself repeatedly can quickly go to zero, or conversely, a large number $v>1.0$ could quickly go to infinity; only this is for matrices.\n"
      ]
    },
    {
      "metadata": {
        "id": "-o3qjqeZpLjN",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Build a tiny RNN in Keras\n",
        "\n",
        "Building an RNN in Keras is quite simple. We simply chain the layers together as follows:"
      ]
    },
    {
      "metadata": {
        "id": "mEobTD6spOWx",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def define_model(truncated_seq_len):  \n",
        "    \n",
        "    input_dimension = 1\n",
        "    hidden_dimension = 1\n",
        "    output_dimension = 1\n",
        "    \n",
        "    model = tf.keras.models.Sequential()\n",
        "    model.add(tf.keras.layers.SimpleRNN(\n",
        "        # We need to specify the input_shape *without* leading batch_size (it is inferred)\n",
        "        input_shape=(truncated_seq_len, input_dimension),\n",
        "        units=hidden_dimension, \n",
        "        return_sequences=False,\n",
        "        name='hidden_layer'))\n",
        "    model.add(tf.keras.layers.Dense(\n",
        "        output_dimension, \n",
        "        name='output_layer'))\n",
        "\n",
        "    model.compile(loss=\"mean_squared_error\", \n",
        "                  optimizer=tf.train.AdamOptimizer(learning_rate=1e-3))\n",
        "    \n",
        "    return model\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "JHRscZaQze26",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**NOTE**: We're building an RNN for **regression**. We therefore use a linear layer (which outputs real-valued numbers) at the output with the \"*mean_squared_error*\" loss function."
      ]
    },
    {
      "metadata": {
        "id": "ONvBy4AopTrF",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 215
        },
        "outputId": "4cff0a47-5688-457a-c071-a5458decf8f6"
      },
      "cell_type": "code",
      "source": [
        "model = define_model(truncated_seq_len = X_train.shape[1])\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "hidden_layer (SimpleRNN)     (None, 1)                 3         \n",
            "_________________________________________________________________\n",
            "output_layer (Dense)         (None, 1)                 2         \n",
            "=================================================================\n",
            "Total params: 5\n",
            "Trainable params: 5\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "ddb6_04dfZvn",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**NOTE**: You need to re-run the above cell every time after training to reset the model weights!"
      ]
    },
    {
      "metadata": {
        "id": "hD94X5iQc8Jg",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Train the tiny RNN\n",
        "Now let's train the model. This may take a few minutes (it takes much longer if you increase `truncated_seq_len`). Set `verbose=1` **before** you run the cell to see the intermediate output as the model is training. Set it to 0 if you don't want any output."
      ]
    },
    {
      "metadata": {
        "id": "xvahCyhk7Wvr",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "''' SOLUTION TO ONE OF TASKS [DELETE]\n",
        "patience = 5\n",
        "train_history = model.fit(X_train, y_train, batch_size=600, epochs=1000, \n",
        "                          verbose=1, validation_split=0.05,\n",
        "                          callbacks=[tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=patience, verbose=1)])\n",
        "'''"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "xumvRz2lrPus",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 36035
        },
        "outputId": "39973883-59f5-4233-8f02-83f902e0dc35"
      },
      "cell_type": "code",
      "source": [
        "train_history = model.fit(X_train, y_train, batch_size=600, epochs=1000, \n",
        "                          verbose=1, validation_split=0.05)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 2506 samples, validate on 132 samples\n",
            "Epoch 1/1000\n",
            "2506/2506 [==============================] - 0s 26us/step - loss: 0.4968 - val_loss: 0.4914\n",
            "Epoch 2/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4965 - val_loss: 0.4911\n",
            "Epoch 3/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4961 - val_loss: 0.4908\n",
            "Epoch 4/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4958 - val_loss: 0.4904\n",
            "Epoch 5/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4953 - val_loss: 0.4898\n",
            "Epoch 6/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.4949 - val_loss: 0.4893\n",
            "Epoch 7/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4944 - val_loss: 0.4887\n",
            "Epoch 8/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4939 - val_loss: 0.4883\n",
            "Epoch 9/1000\n",
            "2506/2506 [==============================] - 0s 13us/step - loss: 0.4933 - val_loss: 0.4877\n",
            "Epoch 10/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4927 - val_loss: 0.4872\n",
            "Epoch 11/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4920 - val_loss: 0.4865\n",
            "Epoch 12/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4913 - val_loss: 0.4856\n",
            "Epoch 13/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4905 - val_loss: 0.4847\n",
            "Epoch 14/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4896 - val_loss: 0.4838\n",
            "Epoch 15/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4887 - val_loss: 0.4828\n",
            "Epoch 16/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4876 - val_loss: 0.4819\n",
            "Epoch 17/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4865 - val_loss: 0.4808\n",
            "Epoch 18/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4853 - val_loss: 0.4797\n",
            "Epoch 19/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.4840 - val_loss: 0.4784\n",
            "Epoch 20/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.4826 - val_loss: 0.4771\n",
            "Epoch 21/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.4811 - val_loss: 0.4756\n",
            "Epoch 22/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.4794 - val_loss: 0.4741\n",
            "Epoch 23/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4777 - val_loss: 0.4725\n",
            "Epoch 24/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.4758 - val_loss: 0.4707\n",
            "Epoch 25/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4738 - val_loss: 0.4689\n",
            "Epoch 26/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4717 - val_loss: 0.4669\n",
            "Epoch 27/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4695 - val_loss: 0.4646\n",
            "Epoch 28/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4671 - val_loss: 0.4622\n",
            "Epoch 29/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4646 - val_loss: 0.4596\n",
            "Epoch 30/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4619 - val_loss: 0.4568\n",
            "Epoch 31/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.4590 - val_loss: 0.4540\n",
            "Epoch 32/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4560 - val_loss: 0.4510\n",
            "Epoch 33/1000\n",
            "2506/2506 [==============================] - 0s 13us/step - loss: 0.4529 - val_loss: 0.4479\n",
            "Epoch 34/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4496 - val_loss: 0.4446\n",
            "Epoch 35/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4461 - val_loss: 0.4412\n",
            "Epoch 36/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4425 - val_loss: 0.4376\n",
            "Epoch 37/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4388 - val_loss: 0.4337\n",
            "Epoch 38/1000\n",
            "2506/2506 [==============================] - 0s 13us/step - loss: 0.4349 - val_loss: 0.4297\n",
            "Epoch 39/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4308 - val_loss: 0.4256\n",
            "Epoch 40/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.4266 - val_loss: 0.4214\n",
            "Epoch 41/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4223 - val_loss: 0.4170\n",
            "Epoch 42/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4178 - val_loss: 0.4127\n",
            "Epoch 43/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.4133 - val_loss: 0.4082\n",
            "Epoch 44/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.4086 - val_loss: 0.4036\n",
            "Epoch 45/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.4038 - val_loss: 0.3988\n",
            "Epoch 46/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.3988 - val_loss: 0.3939\n",
            "Epoch 47/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.3938 - val_loss: 0.3889\n",
            "Epoch 48/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.3888 - val_loss: 0.3839\n",
            "Epoch 49/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.3836 - val_loss: 0.3789\n",
            "Epoch 50/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3785 - val_loss: 0.3738\n",
            "Epoch 51/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3732 - val_loss: 0.3686\n",
            "Epoch 52/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.3679 - val_loss: 0.3634\n",
            "Epoch 53/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.3625 - val_loss: 0.3581\n",
            "Epoch 54/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.3571 - val_loss: 0.3526\n",
            "Epoch 55/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3516 - val_loss: 0.3472\n",
            "Epoch 56/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3461 - val_loss: 0.3417\n",
            "Epoch 57/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.3406 - val_loss: 0.3362\n",
            "Epoch 58/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3350 - val_loss: 0.3306\n",
            "Epoch 59/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.3295 - val_loss: 0.3249\n",
            "Epoch 60/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3238 - val_loss: 0.3192\n",
            "Epoch 61/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.3182 - val_loss: 0.3136\n",
            "Epoch 62/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.3126 - val_loss: 0.3079\n",
            "Epoch 63/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.3069 - val_loss: 0.3022\n",
            "Epoch 64/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.3013 - val_loss: 0.2964\n",
            "Epoch 65/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.2956 - val_loss: 0.2908\n",
            "Epoch 66/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2900 - val_loss: 0.2851\n",
            "Epoch 67/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2844 - val_loss: 0.2794\n",
            "Epoch 68/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2787 - val_loss: 0.2737\n",
            "Epoch 69/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.2729 - val_loss: 0.2679\n",
            "Epoch 70/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.2673 - val_loss: 0.2621\n",
            "Epoch 71/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2615 - val_loss: 0.2563\n",
            "Epoch 72/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2557 - val_loss: 0.2505\n",
            "Epoch 73/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2500 - val_loss: 0.2446\n",
            "Epoch 74/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2442 - val_loss: 0.2388\n",
            "Epoch 75/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.2385 - val_loss: 0.2329\n",
            "Epoch 76/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2328 - val_loss: 0.2271\n",
            "Epoch 77/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.2270 - val_loss: 0.2213\n",
            "Epoch 78/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2212 - val_loss: 0.2155\n",
            "Epoch 79/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.2155 - val_loss: 0.2098\n",
            "Epoch 80/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2099 - val_loss: 0.2041\n",
            "Epoch 81/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.2042 - val_loss: 0.1984\n",
            "Epoch 82/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.1986 - val_loss: 0.1929\n",
            "Epoch 83/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1931 - val_loss: 0.1873\n",
            "Epoch 84/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.1876 - val_loss: 0.1819\n",
            "Epoch 85/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1822 - val_loss: 0.1764\n",
            "Epoch 86/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1768 - val_loss: 0.1710\n",
            "Epoch 87/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.1714 - val_loss: 0.1656\n",
            "Epoch 88/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.1661 - val_loss: 0.1605\n",
            "Epoch 89/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1610 - val_loss: 0.1554\n",
            "Epoch 90/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1559 - val_loss: 0.1505\n",
            "Epoch 91/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1509 - val_loss: 0.1456\n",
            "Epoch 92/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.1461 - val_loss: 0.1409\n",
            "Epoch 93/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.1414 - val_loss: 0.1363\n",
            "Epoch 94/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1368 - val_loss: 0.1317\n",
            "Epoch 95/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1323 - val_loss: 0.1273\n",
            "Epoch 96/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1279 - val_loss: 0.1231\n",
            "Epoch 97/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1237 - val_loss: 0.1190\n",
            "Epoch 98/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.1196 - val_loss: 0.1150\n",
            "Epoch 99/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.1156 - val_loss: 0.1111\n",
            "Epoch 100/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.1117 - val_loss: 0.1074\n",
            "Epoch 101/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.1080 - val_loss: 0.1039\n",
            "Epoch 102/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.1045 - val_loss: 0.1006\n",
            "Epoch 103/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.1011 - val_loss: 0.0973\n",
            "Epoch 104/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0978 - val_loss: 0.0941\n",
            "Epoch 105/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0946 - val_loss: 0.0911\n",
            "Epoch 106/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0916 - val_loss: 0.0882\n",
            "Epoch 107/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0887 - val_loss: 0.0854\n",
            "Epoch 108/1000\n",
            "2506/2506 [==============================] - 0s 14us/step - loss: 0.0859 - val_loss: 0.0828\n",
            "Epoch 109/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0833 - val_loss: 0.0803\n",
            "Epoch 110/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0808 - val_loss: 0.0779\n",
            "Epoch 111/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0783 - val_loss: 0.0757\n",
            "Epoch 112/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0761 - val_loss: 0.0736\n",
            "Epoch 113/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0739 - val_loss: 0.0716\n",
            "Epoch 114/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0719 - val_loss: 0.0697\n",
            "Epoch 115/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0699 - val_loss: 0.0679\n",
            "Epoch 116/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0681 - val_loss: 0.0663\n",
            "Epoch 117/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0663 - val_loss: 0.0646\n",
            "Epoch 118/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0647 - val_loss: 0.0631\n",
            "Epoch 119/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0631 - val_loss: 0.0616\n",
            "Epoch 120/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0617 - val_loss: 0.0603\n",
            "Epoch 121/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0603 - val_loss: 0.0591\n",
            "Epoch 122/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0590 - val_loss: 0.0579\n",
            "Epoch 123/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0578 - val_loss: 0.0567\n",
            "Epoch 124/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0566 - val_loss: 0.0557\n",
            "Epoch 125/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0556 - val_loss: 0.0547\n",
            "Epoch 126/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0545 - val_loss: 0.0538\n",
            "Epoch 127/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0536 - val_loss: 0.0530\n",
            "Epoch 128/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0527 - val_loss: 0.0522\n",
            "Epoch 129/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0519 - val_loss: 0.0515\n",
            "Epoch 130/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0511 - val_loss: 0.0507\n",
            "Epoch 131/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0504 - val_loss: 0.0500\n",
            "Epoch 132/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0497 - val_loss: 0.0494\n",
            "Epoch 133/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0490 - val_loss: 0.0488\n",
            "Epoch 134/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0484 - val_loss: 0.0482\n",
            "Epoch 135/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0478 - val_loss: 0.0477\n",
            "Epoch 136/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0472 - val_loss: 0.0472\n",
            "Epoch 137/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0467 - val_loss: 0.0467\n",
            "Epoch 138/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0462 - val_loss: 0.0463\n",
            "Epoch 139/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0457 - val_loss: 0.0459\n",
            "Epoch 140/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0453 - val_loss: 0.0455\n",
            "Epoch 141/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0448 - val_loss: 0.0451\n",
            "Epoch 142/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0444 - val_loss: 0.0448\n",
            "Epoch 143/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0440 - val_loss: 0.0444\n",
            "Epoch 144/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0437 - val_loss: 0.0441\n",
            "Epoch 145/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0433 - val_loss: 0.0438\n",
            "Epoch 146/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0430 - val_loss: 0.0435\n",
            "Epoch 147/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0427 - val_loss: 0.0431\n",
            "Epoch 148/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0424 - val_loss: 0.0428\n",
            "Epoch 149/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0421 - val_loss: 0.0426\n",
            "Epoch 150/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0418 - val_loss: 0.0423\n",
            "Epoch 151/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0415 - val_loss: 0.0421\n",
            "Epoch 152/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0412 - val_loss: 0.0419\n",
            "Epoch 153/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0410 - val_loss: 0.0416\n",
            "Epoch 154/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0407 - val_loss: 0.0414\n",
            "Epoch 155/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0405 - val_loss: 0.0411\n",
            "Epoch 156/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0402 - val_loss: 0.0409\n",
            "Epoch 157/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0400 - val_loss: 0.0407\n",
            "Epoch 158/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0398 - val_loss: 0.0405\n",
            "Epoch 159/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0395 - val_loss: 0.0403\n",
            "Epoch 160/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0393 - val_loss: 0.0401\n",
            "Epoch 161/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0391 - val_loss: 0.0399\n",
            "Epoch 162/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0389 - val_loss: 0.0397\n",
            "Epoch 163/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0387 - val_loss: 0.0395\n",
            "Epoch 164/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0385 - val_loss: 0.0393\n",
            "Epoch 165/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0382 - val_loss: 0.0391\n",
            "Epoch 166/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0380 - val_loss: 0.0389\n",
            "Epoch 167/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0378 - val_loss: 0.0387\n",
            "Epoch 168/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0376 - val_loss: 0.0385\n",
            "Epoch 169/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0374 - val_loss: 0.0383\n",
            "Epoch 170/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0372 - val_loss: 0.0381\n",
            "Epoch 171/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0370 - val_loss: 0.0379\n",
            "Epoch 172/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0368 - val_loss: 0.0377\n",
            "Epoch 173/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0366 - val_loss: 0.0375\n",
            "Epoch 174/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0364 - val_loss: 0.0373\n",
            "Epoch 175/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0362 - val_loss: 0.0372\n",
            "Epoch 176/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0360 - val_loss: 0.0370\n",
            "Epoch 177/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0358 - val_loss: 0.0368\n",
            "Epoch 178/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0356 - val_loss: 0.0366\n",
            "Epoch 179/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0354 - val_loss: 0.0364\n",
            "Epoch 180/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0352 - val_loss: 0.0362\n",
            "Epoch 181/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0351 - val_loss: 0.0360\n",
            "Epoch 182/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0349 - val_loss: 0.0359\n",
            "Epoch 183/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0347 - val_loss: 0.0357\n",
            "Epoch 184/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0345 - val_loss: 0.0355\n",
            "Epoch 185/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0343 - val_loss: 0.0353\n",
            "Epoch 186/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0341 - val_loss: 0.0351\n",
            "Epoch 187/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0339 - val_loss: 0.0349\n",
            "Epoch 188/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0337 - val_loss: 0.0347\n",
            "Epoch 189/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0335 - val_loss: 0.0345\n",
            "Epoch 190/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0333 - val_loss: 0.0343\n",
            "Epoch 191/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0331 - val_loss: 0.0342\n",
            "Epoch 192/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0329 - val_loss: 0.0340\n",
            "Epoch 193/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0327 - val_loss: 0.0338\n",
            "Epoch 194/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0325 - val_loss: 0.0336\n",
            "Epoch 195/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0323 - val_loss: 0.0334\n",
            "Epoch 196/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0321 - val_loss: 0.0332\n",
            "Epoch 197/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0320 - val_loss: 0.0330\n",
            "Epoch 198/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0318 - val_loss: 0.0328\n",
            "Epoch 199/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0316 - val_loss: 0.0326\n",
            "Epoch 200/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0314 - val_loss: 0.0324\n",
            "Epoch 201/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0312 - val_loss: 0.0322\n",
            "Epoch 202/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0310 - val_loss: 0.0321\n",
            "Epoch 203/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0308 - val_loss: 0.0319\n",
            "Epoch 204/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0306 - val_loss: 0.0317\n",
            "Epoch 205/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0304 - val_loss: 0.0315\n",
            "Epoch 206/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0302 - val_loss: 0.0314\n",
            "Epoch 207/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0301 - val_loss: 0.0312\n",
            "Epoch 208/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0299 - val_loss: 0.0310\n",
            "Epoch 209/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0297 - val_loss: 0.0308\n",
            "Epoch 210/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0295 - val_loss: 0.0306\n",
            "Epoch 211/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0293 - val_loss: 0.0304\n",
            "Epoch 212/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0291 - val_loss: 0.0302\n",
            "Epoch 213/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0289 - val_loss: 0.0300\n",
            "Epoch 214/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0287 - val_loss: 0.0298\n",
            "Epoch 215/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0286 - val_loss: 0.0296\n",
            "Epoch 216/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0284 - val_loss: 0.0295\n",
            "Epoch 217/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0282 - val_loss: 0.0293\n",
            "Epoch 218/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0280 - val_loss: 0.0291\n",
            "Epoch 219/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0278 - val_loss: 0.0289\n",
            "Epoch 220/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0276 - val_loss: 0.0288\n",
            "Epoch 221/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0275 - val_loss: 0.0286\n",
            "Epoch 222/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0273 - val_loss: 0.0285\n",
            "Epoch 223/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0271 - val_loss: 0.0283\n",
            "Epoch 224/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0269 - val_loss: 0.0281\n",
            "Epoch 225/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0268 - val_loss: 0.0279\n",
            "Epoch 226/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0266 - val_loss: 0.0277\n",
            "Epoch 227/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0264 - val_loss: 0.0275\n",
            "Epoch 228/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0262 - val_loss: 0.0274\n",
            "Epoch 229/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0260 - val_loss: 0.0272\n",
            "Epoch 230/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0259 - val_loss: 0.0270\n",
            "Epoch 231/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0257 - val_loss: 0.0268\n",
            "Epoch 232/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0255 - val_loss: 0.0267\n",
            "Epoch 233/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0253 - val_loss: 0.0265\n",
            "Epoch 234/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0252 - val_loss: 0.0263\n",
            "Epoch 235/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0250 - val_loss: 0.0262\n",
            "Epoch 236/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0248 - val_loss: 0.0260\n",
            "Epoch 237/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0247 - val_loss: 0.0258\n",
            "Epoch 238/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0245 - val_loss: 0.0257\n",
            "Epoch 239/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0243 - val_loss: 0.0255\n",
            "Epoch 240/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0242 - val_loss: 0.0253\n",
            "Epoch 241/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0240 - val_loss: 0.0252\n",
            "Epoch 242/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0238 - val_loss: 0.0250\n",
            "Epoch 243/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0237 - val_loss: 0.0249\n",
            "Epoch 244/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0235 - val_loss: 0.0247\n",
            "Epoch 245/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0233 - val_loss: 0.0246\n",
            "Epoch 246/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0232 - val_loss: 0.0244\n",
            "Epoch 247/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0230 - val_loss: 0.0242\n",
            "Epoch 248/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0229 - val_loss: 0.0241\n",
            "Epoch 249/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0227 - val_loss: 0.0239\n",
            "Epoch 250/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0226 - val_loss: 0.0238\n",
            "Epoch 251/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0224 - val_loss: 0.0236\n",
            "Epoch 252/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0222 - val_loss: 0.0235\n",
            "Epoch 253/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0221 - val_loss: 0.0233\n",
            "Epoch 254/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0219 - val_loss: 0.0232\n",
            "Epoch 255/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0218 - val_loss: 0.0230\n",
            "Epoch 256/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0216 - val_loss: 0.0229\n",
            "Epoch 257/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0215 - val_loss: 0.0227\n",
            "Epoch 258/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0213 - val_loss: 0.0226\n",
            "Epoch 259/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0212 - val_loss: 0.0224\n",
            "Epoch 260/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0210 - val_loss: 0.0223\n",
            "Epoch 261/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0209 - val_loss: 0.0222\n",
            "Epoch 262/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0208 - val_loss: 0.0220\n",
            "Epoch 263/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0206 - val_loss: 0.0219\n",
            "Epoch 264/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0205 - val_loss: 0.0217\n",
            "Epoch 265/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0203 - val_loss: 0.0216\n",
            "Epoch 266/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0202 - val_loss: 0.0215\n",
            "Epoch 267/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0201 - val_loss: 0.0213\n",
            "Epoch 268/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0199 - val_loss: 0.0212\n",
            "Epoch 269/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0198 - val_loss: 0.0211\n",
            "Epoch 270/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0197 - val_loss: 0.0210\n",
            "Epoch 271/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0195 - val_loss: 0.0208\n",
            "Epoch 272/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0194 - val_loss: 0.0207\n",
            "Epoch 273/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0193 - val_loss: 0.0206\n",
            "Epoch 274/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0191 - val_loss: 0.0205\n",
            "Epoch 275/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0190 - val_loss: 0.0203\n",
            "Epoch 276/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0189 - val_loss: 0.0202\n",
            "Epoch 277/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0188 - val_loss: 0.0201\n",
            "Epoch 278/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0186 - val_loss: 0.0200\n",
            "Epoch 279/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0185 - val_loss: 0.0198\n",
            "Epoch 280/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0184 - val_loss: 0.0197\n",
            "Epoch 281/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0183 - val_loss: 0.0196\n",
            "Epoch 282/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0182 - val_loss: 0.0195\n",
            "Epoch 283/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0180 - val_loss: 0.0194\n",
            "Epoch 284/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0179 - val_loss: 0.0193\n",
            "Epoch 285/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0178 - val_loss: 0.0192\n",
            "Epoch 286/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0177 - val_loss: 0.0191\n",
            "Epoch 287/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0176 - val_loss: 0.0190\n",
            "Epoch 288/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0175 - val_loss: 0.0188\n",
            "Epoch 289/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0174 - val_loss: 0.0187\n",
            "Epoch 290/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0173 - val_loss: 0.0186\n",
            "Epoch 291/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0172 - val_loss: 0.0185\n",
            "Epoch 292/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0171 - val_loss: 0.0185\n",
            "Epoch 293/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0170 - val_loss: 0.0184\n",
            "Epoch 294/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0169 - val_loss: 0.0183\n",
            "Epoch 295/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0168 - val_loss: 0.0182\n",
            "Epoch 296/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0167 - val_loss: 0.0181\n",
            "Epoch 297/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0166 - val_loss: 0.0180\n",
            "Epoch 298/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0165 - val_loss: 0.0179\n",
            "Epoch 299/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0164 - val_loss: 0.0178\n",
            "Epoch 300/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0163 - val_loss: 0.0177\n",
            "Epoch 301/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0162 - val_loss: 0.0176\n",
            "Epoch 302/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0161 - val_loss: 0.0175\n",
            "Epoch 303/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0160 - val_loss: 0.0174\n",
            "Epoch 304/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0159 - val_loss: 0.0173\n",
            "Epoch 305/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0158 - val_loss: 0.0173\n",
            "Epoch 306/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0157 - val_loss: 0.0172\n",
            "Epoch 307/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0156 - val_loss: 0.0171\n",
            "Epoch 308/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0156 - val_loss: 0.0170\n",
            "Epoch 309/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0155 - val_loss: 0.0169\n",
            "Epoch 310/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0154 - val_loss: 0.0169\n",
            "Epoch 311/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0153 - val_loss: 0.0168\n",
            "Epoch 312/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0152 - val_loss: 0.0167\n",
            "Epoch 313/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0152 - val_loss: 0.0166\n",
            "Epoch 314/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0151 - val_loss: 0.0166\n",
            "Epoch 315/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0150 - val_loss: 0.0165\n",
            "Epoch 316/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0149 - val_loss: 0.0164\n",
            "Epoch 317/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0149 - val_loss: 0.0164\n",
            "Epoch 318/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0148 - val_loss: 0.0163\n",
            "Epoch 319/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0147 - val_loss: 0.0162\n",
            "Epoch 320/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0146 - val_loss: 0.0161\n",
            "Epoch 321/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0146 - val_loss: 0.0161\n",
            "Epoch 322/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0145 - val_loss: 0.0160\n",
            "Epoch 323/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0144 - val_loss: 0.0159\n",
            "Epoch 324/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0144 - val_loss: 0.0159\n",
            "Epoch 325/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0143 - val_loss: 0.0158\n",
            "Epoch 326/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0143 - val_loss: 0.0158\n",
            "Epoch 327/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0142 - val_loss: 0.0157\n",
            "Epoch 328/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0141 - val_loss: 0.0157\n",
            "Epoch 329/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0141 - val_loss: 0.0156\n",
            "Epoch 330/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0140 - val_loss: 0.0156\n",
            "Epoch 331/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0140 - val_loss: 0.0155\n",
            "Epoch 332/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0139 - val_loss: 0.0154\n",
            "Epoch 333/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0138 - val_loss: 0.0154\n",
            "Epoch 334/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0138 - val_loss: 0.0153\n",
            "Epoch 335/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0137 - val_loss: 0.0153\n",
            "Epoch 336/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0137 - val_loss: 0.0152\n",
            "Epoch 337/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0136 - val_loss: 0.0152\n",
            "Epoch 338/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0136 - val_loss: 0.0151\n",
            "Epoch 339/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0135 - val_loss: 0.0151\n",
            "Epoch 340/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0135 - val_loss: 0.0150\n",
            "Epoch 341/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0134 - val_loss: 0.0150\n",
            "Epoch 342/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0134 - val_loss: 0.0149\n",
            "Epoch 343/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0133 - val_loss: 0.0149\n",
            "Epoch 344/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0133 - val_loss: 0.0149\n",
            "Epoch 345/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0132 - val_loss: 0.0148\n",
            "Epoch 346/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0132 - val_loss: 0.0148\n",
            "Epoch 347/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0132 - val_loss: 0.0147\n",
            "Epoch 348/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0131 - val_loss: 0.0147\n",
            "Epoch 349/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0131 - val_loss: 0.0147\n",
            "Epoch 350/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0130 - val_loss: 0.0146\n",
            "Epoch 351/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0130 - val_loss: 0.0146\n",
            "Epoch 352/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0130 - val_loss: 0.0146\n",
            "Epoch 353/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0129 - val_loss: 0.0145\n",
            "Epoch 354/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0129 - val_loss: 0.0145\n",
            "Epoch 355/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0129 - val_loss: 0.0144\n",
            "Epoch 356/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0128 - val_loss: 0.0144\n",
            "Epoch 357/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0128 - val_loss: 0.0144\n",
            "Epoch 358/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0127 - val_loss: 0.0143\n",
            "Epoch 359/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0127 - val_loss: 0.0143\n",
            "Epoch 360/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0127 - val_loss: 0.0143\n",
            "Epoch 361/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0126 - val_loss: 0.0143\n",
            "Epoch 362/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0126 - val_loss: 0.0142\n",
            "Epoch 363/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0126 - val_loss: 0.0142\n",
            "Epoch 364/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0126 - val_loss: 0.0142\n",
            "Epoch 365/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0125 - val_loss: 0.0142\n",
            "Epoch 366/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0125 - val_loss: 0.0141\n",
            "Epoch 367/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0125 - val_loss: 0.0141\n",
            "Epoch 368/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0125 - val_loss: 0.0141\n",
            "Epoch 369/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0124 - val_loss: 0.0141\n",
            "Epoch 370/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0124 - val_loss: 0.0140\n",
            "Epoch 371/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0124 - val_loss: 0.0140\n",
            "Epoch 372/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0124 - val_loss: 0.0140\n",
            "Epoch 373/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0123 - val_loss: 0.0140\n",
            "Epoch 374/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0123 - val_loss: 0.0140\n",
            "Epoch 375/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0123 - val_loss: 0.0139\n",
            "Epoch 376/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0123 - val_loss: 0.0139\n",
            "Epoch 377/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0122 - val_loss: 0.0139\n",
            "Epoch 378/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0122 - val_loss: 0.0139\n",
            "Epoch 379/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0122 - val_loss: 0.0138\n",
            "Epoch 380/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0122 - val_loss: 0.0138\n",
            "Epoch 381/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0122 - val_loss: 0.0138\n",
            "Epoch 382/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0121 - val_loss: 0.0138\n",
            "Epoch 383/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0121 - val_loss: 0.0138\n",
            "Epoch 384/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0121 - val_loss: 0.0138\n",
            "Epoch 385/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0121 - val_loss: 0.0137\n",
            "Epoch 386/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0121 - val_loss: 0.0137\n",
            "Epoch 387/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0121 - val_loss: 0.0137\n",
            "Epoch 388/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0120 - val_loss: 0.0137\n",
            "Epoch 389/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0120 - val_loss: 0.0137\n",
            "Epoch 390/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0120 - val_loss: 0.0137\n",
            "Epoch 391/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0120 - val_loss: 0.0137\n",
            "Epoch 392/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0120 - val_loss: 0.0137\n",
            "Epoch 393/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0120 - val_loss: 0.0136\n",
            "Epoch 394/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0120 - val_loss: 0.0136\n",
            "Epoch 395/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 396/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 397/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 398/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 399/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 400/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 401/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0119 - val_loss: 0.0136\n",
            "Epoch 402/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0119 - val_loss: 0.0135\n",
            "Epoch 403/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 404/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 405/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 406/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 407/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 408/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 409/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 410/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 411/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 412/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0118 - val_loss: 0.0135\n",
            "Epoch 413/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0118 - val_loss: 0.0134\n",
            "Epoch 414/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 415/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 416/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 417/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 418/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 419/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 420/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 421/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 422/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 423/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 424/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 425/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 426/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 427/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0117 - val_loss: 0.0134\n",
            "Epoch 428/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0116 - val_loss: 0.0134\n",
            "Epoch 429/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0134\n",
            "Epoch 430/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0134\n",
            "Epoch 431/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0134\n",
            "Epoch 432/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0134\n",
            "Epoch 433/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 434/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 435/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 436/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 437/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 438/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 439/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 440/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 441/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 442/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 443/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 444/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 445/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 446/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 447/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 448/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 449/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 450/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 451/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 452/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 453/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0116 - val_loss: 0.0133\n",
            "Epoch 454/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 455/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 456/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 457/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 458/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 459/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 460/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 461/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 462/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 463/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 464/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 465/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 466/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 467/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 468/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0133\n",
            "Epoch 469/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 470/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 471/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 472/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 473/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 474/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 475/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 476/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 477/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 478/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 479/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 480/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 481/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 482/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 483/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 484/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 485/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 486/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 487/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 488/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 489/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 490/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 491/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 492/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 493/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 494/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 495/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 496/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 497/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 498/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 499/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 500/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 501/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 502/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 503/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 504/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 505/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 506/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0115 - val_loss: 0.0132\n",
            "Epoch 507/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 508/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 509/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 510/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 511/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 512/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 513/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 514/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 515/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 516/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 517/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 518/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 519/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 520/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 521/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 522/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 523/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 524/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 525/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 526/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 527/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 528/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 529/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 530/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 531/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 532/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 533/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 534/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 535/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 536/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 537/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 538/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 539/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 540/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 541/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 542/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 543/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0114 - val_loss: 0.0132\n",
            "Epoch 544/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 545/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 546/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 547/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 548/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 549/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 550/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 551/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 552/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 553/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 554/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 555/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 556/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 557/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 558/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 559/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 560/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 561/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 562/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 563/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 564/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 565/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 566/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 567/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 568/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 569/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 570/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 571/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 572/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 573/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 574/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 575/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 576/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 577/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 578/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0114 - val_loss: 0.0131\n",
            "Epoch 579/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 580/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 581/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 582/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 583/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 584/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 585/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 586/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 587/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 588/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 589/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 590/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 591/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 592/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 593/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 594/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 595/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 596/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 597/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 598/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 599/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 600/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 601/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 602/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 603/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 604/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 605/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 606/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 607/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 608/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 609/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 610/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 611/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 612/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 613/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 614/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0131\n",
            "Epoch 615/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 616/1000\n",
            "2506/2506 [==============================] - 0s 21us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 617/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 618/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 619/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 620/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 621/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 622/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 623/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 624/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 625/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 626/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 627/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 628/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 629/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 630/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 631/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 632/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 633/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 634/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 635/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 636/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 637/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 638/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 639/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 640/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 641/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 642/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 643/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 644/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 645/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 646/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 647/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0113 - val_loss: 0.0130\n",
            "Epoch 648/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 649/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 650/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 651/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 652/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 653/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 654/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 655/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 656/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 657/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 658/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 659/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 660/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 661/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 662/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 663/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 664/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 665/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 666/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 667/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 668/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 669/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 670/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 671/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 672/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 673/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 674/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 675/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0130\n",
            "Epoch 676/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 677/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 678/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 679/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 680/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 681/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 682/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 683/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 684/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 685/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 686/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 687/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 688/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 689/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 690/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 691/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 692/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 693/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 694/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 695/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 696/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 697/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 698/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 699/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 700/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 701/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 702/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 703/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 704/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 705/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0112 - val_loss: 0.0129\n",
            "Epoch 706/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 707/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 708/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 709/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 710/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 711/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 712/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 713/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 714/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 715/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 716/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 717/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 718/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 719/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 720/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 721/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 722/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 723/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 724/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 725/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 726/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 727/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 728/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 729/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 730/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 731/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 732/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 733/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0129\n",
            "Epoch 734/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 735/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 736/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 737/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 738/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 739/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 740/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 741/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 742/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 743/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 744/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 745/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 746/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 747/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 748/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 749/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 750/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 751/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 752/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 753/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 754/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 755/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 756/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 757/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 758/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 759/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0111 - val_loss: 0.0128\n",
            "Epoch 760/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 761/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 762/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 763/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 764/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 765/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 766/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 767/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 768/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 769/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 770/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 771/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 772/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 773/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 774/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 775/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 776/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 777/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 778/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 779/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 780/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 781/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 782/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 783/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 784/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 785/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 786/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 787/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0110 - val_loss: 0.0128\n",
            "Epoch 788/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 789/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 790/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 791/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 792/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 793/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 794/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 795/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 796/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 797/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 798/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 799/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 800/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 801/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 802/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 803/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 804/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 805/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 806/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 807/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 808/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 809/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 810/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 811/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0110 - val_loss: 0.0127\n",
            "Epoch 812/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 813/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 814/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 815/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 816/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 817/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 818/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 819/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 820/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 821/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 822/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 823/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 824/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 825/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 826/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 827/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 828/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 829/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 830/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 831/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 832/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 833/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 834/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 835/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0127\n",
            "Epoch 836/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 837/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 838/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 839/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 840/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 841/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 842/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 843/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 844/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 845/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 846/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 847/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 848/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 849/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 850/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 851/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 852/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 853/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 854/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 855/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 856/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 857/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 858/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 859/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 860/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 861/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 862/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0109 - val_loss: 0.0126\n",
            "Epoch 863/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 864/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 865/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 866/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 867/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 868/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 869/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 870/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 871/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 872/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 873/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 874/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 875/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 876/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 877/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 878/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 879/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 880/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 881/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0126\n",
            "Epoch 882/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 883/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 884/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 885/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 886/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 887/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 888/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 889/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 890/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 891/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 892/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 893/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 894/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 895/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 896/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 897/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 898/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 899/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 900/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 901/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 902/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 903/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 904/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 905/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 906/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 907/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 908/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 909/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 910/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 911/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 912/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 913/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 914/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0108 - val_loss: 0.0125\n",
            "Epoch 915/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 916/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 917/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 918/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 919/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 920/1000\n",
            "2506/2506 [==============================] - 0s 20us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 921/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 922/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 923/1000\n",
            "2506/2506 [==============================] - 0s 23us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 924/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 925/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 926/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 927/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 928/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 929/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 930/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 931/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 932/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0125\n",
            "Epoch 933/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 934/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 935/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 936/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 937/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 938/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 939/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 940/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 941/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 942/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 943/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 944/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 945/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 946/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 947/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 948/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 949/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 950/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 951/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 952/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 953/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 954/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 955/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 956/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 957/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 958/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 959/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 960/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0107 - val_loss: 0.0124\n",
            "Epoch 961/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 962/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 963/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 964/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 965/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 966/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 967/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 968/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 969/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 970/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 971/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 972/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 973/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 974/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 975/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 976/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 977/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 978/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 979/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 980/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 981/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 982/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0106 - val_loss: 0.0124\n",
            "Epoch 983/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 984/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 985/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 986/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 987/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 988/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 989/1000\n",
            "2506/2506 [==============================] - 0s 17us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 990/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 991/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 992/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 993/1000\n",
            "2506/2506 [==============================] - 0s 18us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 994/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 995/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 996/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 997/1000\n",
            "2506/2506 [==============================] - 0s 15us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 998/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 999/1000\n",
            "2506/2506 [==============================] - 0s 19us/step - loss: 0.0106 - val_loss: 0.0123\n",
            "Epoch 1000/1000\n",
            "2506/2506 [==============================] - 0s 16us/step - loss: 0.0106 - val_loss: 0.0123\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "CebcKzSW_g6P",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Let's visualize the training and validation losses."
      ]
    },
    {
      "metadata": {
        "id": "U9G1OHXoroBs",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 369
        },
        "outputId": "d0ba7ee2-92b9-4eda-87c0-11b40ce83449"
      },
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(15,5))\n",
        "\n",
        "for label in [\"loss\",\"val_loss\"]:\n",
        "    plt.plot(train_history.history[label], label=label)\n",
        "\n",
        "plt.ylabel(\"loss\")\n",
        "plt.xlabel(\"epoch\")\n",
        "plt.title(\"The final validation loss: {}\".format(train_history.history[\"val_loss\"][-1]))\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA3YAAAFMCAYAAACH0y5vAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSUVORK5CYII=\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f9c93e9b410>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "Rs_-XnYhKl_N",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Finally, let's look at the parameters for the trained model."
      ]
    },
    {
      "metadata": {
        "id": "q4gtwBT7Kgh0",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 73
        },
        "outputId": "c212be9b-bd62-41cf-9163-33f69e306d95"
      },
      "cell_type": "code",
      "source": [
        "for layer in model.layers:\n",
        "  print(\"{}, {}\".format(layer.name, layer.get_weights()))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "hidden_layer, [array([[0.7633411]], dtype=float32), array([[-0.56165993]], dtype=float32), array([0.04831408], dtype=float32)]\n",
            "output_layer, [array([[2.4384913]], dtype=float32), array([-0.06346191], dtype=float32)]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "bqFBu_dCsUqi",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**QUESTION**: \n",
        "* Relate the above weights to the terms in the equation for the vanilla RNN we saw earlier, namely:\n",
        "  * input-to-hidden $W_{xh}$,\n",
        "  * hidden-to-hidden $W_{hh}$,\n",
        "  * hidden-to-output weights $W_{hy}$\n",
        "  * recurrent and out biases."
      ]
    },
    {
      "metadata": {
        "id": "0FHaN-VXfxEl",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Make predictions using the trained model"
      ]
    },
    {
      "metadata": {
        "id": "IQl_msx-4o3E",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 231
        },
        "outputId": "7c2ac8d5-231f-47af-d227-092510c058ff"
      },
      "cell_type": "code",
      "source": [
        "y_pred = model.predict(X_test[:100])\n",
        "plt.figure(figsize=(19,3))\n",
        "\n",
        "plt.plot(y_test[:100], label=\"true\")\n",
        "plt.plot(y_pred, label=\"predicted\")\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABE0AAADCCAYAAAC8Njw/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIYQQYhD/HyUXQgA7s75JAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7f9c91608110>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "MQj5P-FKOgqG",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**YOUR TASKS**: \n",
        "* [**ALL**] Change the learning rate up to 5 orders of magnitude larger and smaller and retrain. What happens when it is too large? What happens when it is too small?\n",
        "* [**ALL**] Change the `SimpleRNN` to `GRU`.\n",
        "  * You can read more about LSTMs in [this blog post](https://colah.github.io/posts/2015-08-Understanding-LSTMs/) by Chris Olah. \n",
        "  * What is the effect on the number of parameters? Can you explain why? Now do the same for `LSTM`.\n",
        "* [**INTERMEDIATE**] Note that the loss does not decrease much after around epoch 400. Add \"Early Stopping with patience\" to the `model.fit()` function to stop it from training beyond this point. **Hint**: Look at tf.keras.callbacks.\n",
        "  * *Early stopping* is a technique where we stop training the model once it's performance on validation data stops improving, and is a very common approach to prevent overfitting. Early stopping *with patience* means as soon as the model starts doing worse on validation we wait for at least `patience` more evaluations before stopping training, and if it improves within that time, we reset the counter. The patience parameter is a way to ensure we don't stop by accident, as the validation loss can fluctuate randomly from epoch to epoch.\n"
      ]
    },
    {
      "metadata": {
        "id": "_totIpxmZ8_v",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "##Generating Shakespeare\n",
        "\n",
        "Now let's build an RNN language model to generate Shakespearian English! A language model is trained to assign high probabilities to sequences of words or sentences that are well formed, and low probabilities to sequences which are not realistic. When the model is trained, one can use it to *generate* data that is similar to the training data.\n",
        "\n",
        "Our data is now sequences of discrete symbols (characters). But neural networks operate in continuous spaces, and so we need to take the discrete language data, and **embed** it in a continuous space. To do this, we'll simply break up the data into sequences of characters, and represent each character using a learned vector. This is a standard trick for processing text using neural networks. "
      ]
    },
    {
      "metadata": {
        "id": "9dFQtnDLZa3c",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "### Download and Preprocess the Data"
      ]
    },
    {
      "metadata": {
        "id": "Tmmjc-EigwCw",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "We first download the data and examine what it looks like:"
      ]
    },
    {
      "metadata": {
        "id": "hybIopOLPD4f",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 791
        },
        "outputId": "eded9c5c-2ece-4774-b207-8990b931f782"
      },
      "cell_type": "code",
      "source": [
        "context = ssl._create_unverified_context()\n",
        "shakespeare_url = 'https://cs.stanford.edu/people/karpathy/char-rnn/shakespeare_input.txt'\n",
        "\n",
        "data = urllib2.urlopen(shakespeare_url, context=context)\n",
        "all_text = data.read().lower()\n",
        "\n",
        "print(\"Downloaded Shakespeare data with {} characters.\".format(len(all_text)))\n",
        "print(\"FIRST 1000 CHARACTERS: \")\n",
        "print(all_text[:1000])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloaded Shakespeare data with 4573338 characters.\n",
            "FIRST 1000 CHARACTERS: \n",
            "first citizen:\n",
            "before we proceed any further, hear me speak.\n",
            "\n",
            "all:\n",
            "speak, speak.\n",
            "\n",
            "first citizen:\n",
            "you are all resolved rather to die than to famish?\n",
            "\n",
            "all:\n",
            "resolved. resolved.\n",
            "\n",
            "first citizen:\n",
            "first, you know caius marcius is chief enemy to the people.\n",
            "\n",
            "all:\n",
            "we know't, we know't.\n",
            "\n",
            "first citizen:\n",
            "let us kill him, and we'll have corn at our own price.\n",
            "is't a verdict?\n",
            "\n",
            "all:\n",
            "no more talking on't; let it be done: away, away!\n",
            "\n",
            "second citizen:\n",
            "one word, good citizens.\n",
            "\n",
            "first citizen:\n",
            "we are accounted poor citizens, the patricians good.\n",
            "what authority surfeits on would relieve us: if they\n",
            "would yield us but the superfluity, while it were\n",
            "wholesome, we might guess they relieved us humanely;\n",
            "but they think we are too dear: the leanness that\n",
            "afflicts us, the object of our misery, is as an\n",
            "inventory to particularise their abundance; our\n",
            "sufferance is a gain to them let us revenge this with\n",
            "our pikes, ere we become rakes: for the gods know i\n",
            "speak this in hunger for bread, not in thirst for revenge.\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "jIuagNcdQLqM",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "training_text = all_text[:1000000] # Keep only the first 1 million characters"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "UK1x69AngvLJ",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "We now preprocess the text data as follows:\n",
        "\n",
        "1. Extract the vocabulary of all `vocab_size` unique characters appearing in the data.\n",
        "\n",
        "2. Assign each character a unique integer id in `0 <= id < vocab_size`. This is so we can map the characters to unique embedding vectors. This is a common way to map discrete inputs to continuous vectors that neural networks can work with. See e.g. this [blog post](https://www.tensorflow.org/tutorials/representation/word2vec), or [this one](http://ruder.io/word-embeddings-1/) for more information.\n",
        "\n",
        "3. Split the data into sequences (\"windows\") of `max_len` characters (the input to the model) followed by the next character as target. E.g. using `max_len=5` the sentence \"I saw a cat\" (11 characters) will get split into \"I saw\" and /space/, \"/space/saw/space/\" and \"a\", \"saw a\" and /space/, etc. To add some variation, we skip `step` characters between each sequence (i.e. we use a \"sliding window of `max_len` with stride `step`\")."
      ]
    },
    {
      "metadata": {
        "id": "zPvsNXKZPJKz",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 73
        },
        "outputId": "1dcaaa7f-b946-4f8d-cb1b-32674b125465"
      },
      "cell_type": "code",
      "source": [
        "max_len = 30  # We only consider this many previous data points (characters)\n",
        "step = 3 # We start a new training sequence every `step` characters\n",
        "sentences = [] # This holds our extracted sequences\n",
        "next_chars = [] # This holds the targets (the follow-up characters)\n",
        "\n",
        "chars = sorted(list(set(training_text)))  # List of unique characters in the corpus\n",
        "vocab_size = len(chars)\n",
        "print('Number of unique characters: ', vocab_size)\n",
        "print(chars)\n",
        "\n",
        "# Construct dictionaries mapping unique characters to their index in `chars` and reverse\n",
        "char2index = dict((c, chars.index(c)) for c in chars)\n",
        "index2char = dict((chars.index(c), c) for c in chars)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of unique characters:  39\n",
            "['\\n', ' ', '!', '$', '&', \"'\", ',', '-', '.', '3', ':', ';', '?', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "O2PhcQwrc2JX",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Now we encode the training data by mapping each character to its unique integer id."
      ]
    },
    {
      "metadata": {
        "id": "xY0qXZmVq8kW",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "89446005-6eba-42da-fce5-8bcfb723a567"
      },
      "cell_type": "code",
      "source": [
        "for i in range(0, len(training_text) - max_len, step):\n",
        "    sentences.append([char2index[s] for s in training_text[i: i + max_len]])\n",
        "    next_chars.append([char2index[s] for s in training_text[i + max_len]])\n",
        "\n",
        "print('Number of extracted sequences:', len(sentences))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of extracted sequences: 333324\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "PsZy6Kshc-Sp",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "This yields the following numpy arrays:"
      ]
    },
    {
      "metadata": {
        "id": "vhgSaP5ntDtq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "2af89a0d-e7e3-4744-8d8d-aa9ec1ab9293"
      },
      "cell_type": "code",
      "source": [
        "X, Y = np.array(sentences, dtype=np.int64), np.array(next_chars, dtype=np.int64)\n",
        "X.shape, Y.shape"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "((333324, 30), (333324, 1))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 22
        }
      ]
    },
    {
      "metadata": {
        "id": "sBR2LsXjmqTU",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Let's take a look at the first example."
      ]
    },
    {
      "metadata": {
        "id": "2H1jobcrmwK7",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 89
        },
        "outputId": "3826ee10-ca7f-466f-c0f4-5ab37ef7aab7"
      },
      "cell_type": "code",
      "source": [
        "print(\"X[0].shape = {}, Y[0].shape = {}\".format(X[0].shape, Y[0].shape))\n",
        "print(\"X[0]: \", X[0])\n",
        "print(\"Y[0]: \", Y[0])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "X[0].shape = (30,), Y[0].shape = (1,)\n",
            "X[0]:  [18 21 30 31 32  1 15 21 32 21 38 17 26 10  0 14 17 18 27 30 17  1 35 17\n",
            "  1 28 30 27 15 17]\n",
            "Y[0]:  [17]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "tEQWEZlSZgpF",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Build an RNN language model"
      ]
    },
    {
      "metadata": {
        "id": "rCysOLWadPHF",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "A **language model** estimates a probability distribution over sequences $\\mathbb{x}_{1:N} = (x_1, x_2, ..., x_N)$ by breaking up the full joint probability into a sequence of conditional probabilities using the **[chain-rule of probability](https://en.wikipedia.org/wiki/Chain_rule_(probability))**:\n",
        "\n",
        "\\begin{align}\n",
        " p(\\mathbb{x}_{1:N}) &= p(x_1) \\cdot p(x_2 | x_1) \\cdot p(x_3 | x_2, x_1) \\ldots \\\\\n",
        " &= \\Pi_1^N p(x_i | \\mathbb{x}_{1:i-1})\n",
        "\\end{align}\n",
        "\n",
        "In other words, to model the probability of the phrase \"*i saw a cat*\" at the character level, the model learns to estimate the probabilities for p(i), p(/space/| i), p(c | i, /space/), and so forth, and multiplies them together. \n",
        "\n",
        "There are many different ways in which to estimate these individual probabilities. But one particularly effective way is to use an RNN! To do this, we'll therefore be modeling the  $p(x_i | \\mathbb{x}_{1:i-1})$ terms using an RNN conditioned on $\\mathbb{x}_{1:i-1}$.\n",
        "\n",
        "* We model these probabilities at the character-level, so we'll use an `Embedding` layer as the first layer of our model to map the discrete character id's to real-valued embedding vectors. \n",
        "* Next, the RNN-core will map these sequences of character embeddings to a probability distribution over all characters $p(x_i | \\mathbb{x}_{1:i-1}) \\in \\mathbb{R}^\\textrm{vocab_size}$ at every step of the sequence. To do this, the RNN will map the embeddings to a sequence of *hidden states*. We will then use a `Dense` layer to map from the RNN hidden state to an output distribution over the total number of characters using a [`softmax`](https://en.wikipedia.org/wiki/Softmax_function) activation.\n",
        "\n",
        "We can do this with a few lines of code:"
      ]
    },
    {
      "metadata": {
        "id": "k8AxhQuePOCN",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 251
        },
        "outputId": "35b12bba-5085-487e-8f08-83d149261ca1"
      },
      "cell_type": "code",
      "source": [
        "embedding_dim = 32   # Map each character to a unique vector of this dimension\n",
        "vocab_size = len(chars)\n",
        "\n",
        "model = tf.keras.models.Sequential()\n",
        "model.add(tf.keras.layers.Embedding(\n",
        "  vocab_size, embedding_dim, \n",
        "  input_length=max_len, \n",
        "  embeddings_initializer=tf.keras.initializers.TruncatedNormal))\n",
        "model.add(tf.keras.layers.LSTM(\n",
        "  128, \n",
        "  input_shape=(max_len, embedding_dim),  # NB: Ensure this matches the embedding_dim!\n",
        "  dropout=0.1,  # input-to-hidden drop-probability\n",
        "  recurrent_dropout=0.2))  # hidden-to-hidden drop-probability\n",
        "model.add(tf.keras.layers.Dense(vocab_size, activation='softmax'))\n",
        "\n",
        "model.summary()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "embedding_2 (Embedding)      (None, 30, 32)            1248      \n",
            "_________________________________________________________________\n",
            "lstm_2 (LSTM)                (None, 128)               82432     \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 39)                5031      \n",
            "=================================================================\n",
            "Total params: 88,711\n",
            "Trainable params: 88,711\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "m4fFjFtZokf2",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Select the optimizer and loss"
      ]
    },
    {
      "metadata": {
        "id": "RHOWm55whZUt",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Once we have a model that can map sequence of characters to a probability distribution over the next character in the sequence, we can train it using **maximum likelihood** on the training set to find the model parameters which maximizes the probability of the training data. Again, this is very simple to do by choosing an optimizer and selecting the `sparse_categorical_crossentropy` loss function:"
      ]
    },
    {
      "metadata": {
        "id": "hKJNZXs7PSW-",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "optimizer = tf.train.AdamOptimizer(learning_rate=0.001)\n",
        "loss='sparse_categorical_crossentropy'\n",
        "\n",
        "model.compile(loss=loss, optimizer=optimizer)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "dEr9hEqFiGx6",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Helper functions"
      ]
    },
    {
      "metadata": {
        "id": "3zXa95pBPUw3",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def sample_with_temp(preds, temperature=1.0):\n",
        "  preds = np.asarray(preds).astype('float64')\n",
        "  preds = np.log(preds) / temperature\n",
        "  exp_preds = np.exp(preds)\n",
        "  preds = exp_preds / np.sum(exp_preds)\n",
        "  probas = np.random.multinomial(1, preds, 1)\n",
        "  return np.argmax(probas)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "AyH6U3er5fis",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 53
        },
        "outputId": "c7d6e470-604f-4b85-e6bd-433f88e0069e"
      },
      "cell_type": "code",
      "source": [
        "def shift_and_append(test_arr, next_item):\n",
        "  '''Returns a copy of test_arr with items shifted one position to the left and \n",
        "     next_item appended.\n",
        "  '''\n",
        "  tmp = np.empty_like(test_arr)\n",
        "  tmp[:,:-1] = test_arr[:,1:]\n",
        "  tmp[:,-1] = next_item\n",
        "  return tmp\n",
        "\n",
        "## TEST the above function:\n",
        "test_arr = np.array([[1,2,3,4]])\n",
        "\n",
        "print(\"test_arr = {}\".format(test_arr))\n",
        "test_arr = shift_and_append(test_arr, 5)\n",
        "print(\"roll_arr(test_arr, 5) = {}\".format(test_arr))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "test_arr = [[1 2 3 4]]\n",
            "roll_arr(test_arr, 5) = [[2 3 4 5]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "q9le2p0YxeYD",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def sample_from_model(model, \n",
        "                      num_generate=400, \n",
        "                      prev_text=None,  # the text used to condition the model\n",
        "                      temperatures=[0.2, 0.5, 1.0, 1.2]):\n",
        "    \n",
        "    if not prev_text:\n",
        "      # Select a text seed at random\n",
        "      start_index = random.randint(0, len(training_text) - max_len - 1)\n",
        "      while ((start_index < (len(training_text) - max_len - 1)) and (\n",
        "        training_text[start_index - 1] is not ' ')):\n",
        "        start_index += 1  # Advance to beginning of new word\n",
        "      prev_text = training_text[start_index: start_index + max_len]\n",
        "    \n",
        "    if len(prev_text) != max_len:\n",
        "      print(\"`prev_text` must be of length `max_len`.\")\n",
        "      return\n",
        "\n",
        "    print('GENERATING TEXT WITH SEED: \\n\"' + prev_text + '\"')\n",
        "    prev_text_arr = np.array(\n",
        "      [[char2index[c] for c in prev_text]], dtype=np.int64) \n",
        "    \n",
        "    for temp in temperatures:\n",
        "        print('==TEMPERATURE:', temp)\n",
        "        sys.stdout.write(prev_text)\n",
        "\n",
        "        # Start with the same sampled text for all temperatures\n",
        "        generated_text = prev_text \n",
        "        generated_text_arr = prev_text_arr\n",
        "\n",
        "        # Now generate this many characters\n",
        "        for i in range(num_generate):         \n",
        "            \n",
        "            # Get the output softmax given the conditioning text\n",
        "            #prev_text = generated_text_enc[np.newaxis,:]\n",
        "            preds = model.predict(generated_text_arr, verbose=0)[0]\n",
        "            \n",
        "            next_index = sample_with_temp(preds, temp)\n",
        "            next_char = index2char[next_index]\n",
        "            generated_text += next_char\n",
        "            generated_text = generated_text[1:]\n",
        "\n",
        "            # Left-shift and add into encoded array\n",
        "            generated_text_arr = shift_and_append(generated_text_arr, next_index)\n",
        "\n",
        "            sys.stdout.write(next_char)\n",
        "            sys.stdout.flush()\n",
        "        print()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "a78CrsaMiKrA",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###Train the model\n",
        "\n",
        "Let's train the model! The code below will train the model on a subset of the available data, and then generate from the model every `sample_every` number of batches.\n",
        "\n",
        "To generate from the model, we use `model.predict()` on a sequence of `max_len` conditioning characters to produce an output distribution over `vocab_size` characters. We then sample one character from this distribution and shift everything up by one and append the new characters. By repeating this, we can generate text from the (partially-trained) model.\n",
        "\n",
        "**NOTE**: \n",
        "* It takes a while to train a model that starts generating anything resembling the Shapespeare text! In general it should start getting the rough structure in place around the 100K training example mark (examples, not batches). But to generate any meaningful words will need several hundred thousand examples.\n",
        "* We sample with *temperature*. This is a way to sharpen or flatten the probabilities produced by the model. By lowering the temperature, we emphasize the modes of the predicted distribution, and by increasing the temperature, we flatten the modes (tends towards uniform). Higher temperatures therefore encourage the model to be more 'creative', instead of always choosing the most likely next character."
      ]
    },
    {
      "metadata": {
        "id": "0RcFKgkdPXuI",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 8976
        },
        "outputId": "bb20fa69-1ab1-48d0-843f-b8cc9ed2c77c"
      },
      "cell_type": "code",
      "source": [
        "batch_size = 128\n",
        "total_num_batches = X.shape[0] // batch_size\n",
        "sample_every = 256  # Train on this many batches, then generate something\n",
        "\n",
        "print(\"Training on {} batches in total.\".format(total_num_batches))\n",
        "\n",
        "for cur_batch in range(0, total_num_batches, sample_every):\n",
        "    print('TRAINING ON BATCH {} to {} (example {} to {})'.format(\n",
        "      cur_batch, cur_batch + sample_every,\n",
        "      cur_batch * batch_size, (cur_batch + sample_every) * batch_size)\n",
        "    )\n",
        "    \n",
        "    X_batch = X[batch_size * cur_batch : batch_size * (cur_batch + sample_every), :]\n",
        "    Y_batch = Y[batch_size * cur_batch : batch_size * (cur_batch + sample_every), :]\n",
        "    \n",
        "    '''\n",
        "    # Show the first 5 examples to make sure we're not training on garbage\n",
        "    print(\"X_batch.shape = {}\".format(X_batch.shape))\n",
        "    print(\"Y_batch.shape = {}\".format(Y_batch.shape))\n",
        "    print(\"FIRST 5 EXAMPLES:\")\n",
        "    for num in range(5):\n",
        "      in_seq = [index2char[int(indx)] for indx in np.nditer(X_batch[num, :])]\n",
        "      next_char = index2char[Y_batch[num, 0]]\n",
        "      print(str(num) + '. ' + ''.join(in_seq) + '-->' + next_char)\n",
        "    '''\n",
        "       \n",
        "    model.fit(X_batch, Y_batch,\n",
        "              batch_size=batch_size,\n",
        "              epochs=1,\n",
        "              verbose=1)\n",
        "\n",
        "    print(\"GENERATING SOME RANDOM TEXT FROM THE MODEL\")\n",
        "    sample_from_model(model)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Training on 2604 batches in total.\n",
            "TRAINING ON BATCH 0 to 256 (example 0 to 32768)\n",
            "Epoch 1/1\n",
            "32768/32768 [==============================] - 124s 4ms/step - loss: 2.7223\n",
            "GENERATING SOME RANDOM TEXT FROM THE MODEL\n",
            "=====GENERATING TEXT WITH SEED: \n",
            "\"else thou art suborn'd against\"\n",
            "==========TEMPERATURE: 0.2\n",
            "else thou art suborn'd against sous the the the and the and wou the we hand the hand the and the our the han the ane the the ar the an the the and hand the hat he hand the the and and the the ser mous and the and the wous wou sert and the sou the wour yous wou whe wou the the the and the sou are he than the the mat ane the the wous an the the hand wou the hand the hat ane in the and the wind the the we the hare he and the and \n",
            "==========TEMPERATURE: 0.5\n",
            "else thou art suborn'd against then wous tha collld me wall oucon isen sous, the ve i ande thene ase ak on? he the dens harendene and ous has the pour me mand an thean you ha co hat the and hene the se the or oure hare ias and the uin wo the anan han sor hane wol leond couto angus the ar hor hand fous the mone an donene sous you vore the bed at fof se that ani wee thes hous gome arius fowe boue thi that oous mat ind and har bo\n",
            "==========TEMPERATURE: 1.0\n",
            "else thou art suborn'd againste bounds\n",
            "rhmes;\n",
            "ourhe wcuthys !o wgarhin'vghanl inhabgle thestons t bograg gavcan ' swtm thenucesu\n",
            "golde sy',\n",
            "\n",
            "eat ak spwou, dcint tod fo ujan fou, puaf rou harr utharng:\n",
            "eanhs; pow ir aomem ce; enrhes pavir ws.\n",
            "yoteusgoxua! ay pom rofori .,r, hor londemen\n",
            "thoag yor d hcas aats h, nehhas ial y hayan nt on di\n",
            "mll tous.\n",
            "; avtoly recetinlas walom\n",
            "menstcom\n",
            "thet,\n",
            "mnc eapa? darm aay\n",
            "tond monarpd hes for\n",
            "==========TEMPERATURE: 1.2\n",
            "else thou art suborn'd againsth ac\n",
            "nfed\n",
            "nysefhekcf,\n",
            "bbmwat$\n",
            "eat helstrcrec nor. oour .\n",
            "witin wysinew:\n",
            "yary bom.\n",
            "w yitg\n",
            "ec,?\n",
            "whuade, war:,, lke, ty re foen tlaniulh,a\n",
            "yk thee puyby houcf'gmmend\n",
            "y'g wri;te se'de wine beant cm moudut bo kreliwe bswtot wwwapkntth\n",
            "win yole l het ofin- dbelim,l ctais, sfwin hyf\n",
            "d'!sesilevani? tharm yor bkafo btgn whe anuya\n",
            "urest hapadru.\n",
            "sfat houkt\n",
            "knd\n",
            "n:\n",
            "recict gordisio'\n",
            "lacoauf,s\n",
            "bte ro sivorsr fo\n",
            "TRAINING ON BATCH 256 to 512 (example 32768 to 65536)\n",
            "Epoch 1/1\n",
            "32768/32768 [==============================] - 124s 4ms/step - loss: 2.3506\n",
            "GENERATING SOME RANDOM TEXT FROM THE MODEL\n",
            "=====GENERATING TEXT WITH SEED: \n",
            "\"would he for the momentary tri\"\n",
            "==========TEMPERATURE: 0.2\n",
            "would he for the momentary trit the the mes the hat the the and in the the dour hare sour and and the the the the the the the the the be the mere hare he the hare hare he the sengere the the the the the the wour the the seare thare hare the the sour and sour the the the the and and meand the the the sour har the the me the he the the the the the me deat mour hare the the the the the the be the bere and the the the the sere the\n",
            "==========TEMPERATURE: 0.5\n",
            "would he for the momentary tringe meand mane erde thane sinke and the sour athe come bils got fast wour wour wis the forere be the moces a and to and mope hat blevere cous me nod ar the bout rengane dot the the the dos, hins me hare to the the sthe et ford.\n",
            "\n",
            "me butest thou the bere buther ante he pat hace the the wis mee de of mer and the sof and the the douer war the corte boured and beer me the meres sthe the wou hare be to \n",
            "==========TEMPERATURE: 1.0\n",
            "would he for the momentary tripthep kase norel,\n",
            " shis sace hame wad wath , misus fue, pirce yofarw bely wusidlas,:\n",
            "\n",
            "yout, hamu mem; try frothot hanoe, foule the snatobe.\n",
            "\n",
            "biyidour ymh fath mond hawh lfy aid hare at in it cotus of an roue h; hame osagterhh\n",
            "voium angert.\n",
            "\n",
            "benvin oupe arincorend amerer fisnd frecome gadram!\n",
            "froce; thurediug\n",
            "cire:\n",
            "whapltj 'le ai inds ot the ad dowh math srand auty ay, id athet coure grato conoth e\n",
            "==========TEMPERATURE: 1.2\n",
            "would he for the momentary trin ennh\n",
            "ih thartor tfsas camrncas thin w-d'yhe har, cionll.\n",
            "\n",
            "youtgsre on lvoiet pfea\n",
            "ancumy gooe hasy butan hysec fore huthe dore y!\n",
            "aknompt thexlos enly.\n",
            "svo utsply,\n",
            "ais stour, harky\n",
            "whe gy'ens nrod buthovirs gardst\n",
            "sorhbt tlule ifade.\n",
            "\n",
            "raver, com mes lovt, the nomvor erill, mim. go thetes oponve ut the buke bu .ome ay ntith\n",
            "at slef, havi,\n",
            ", thoux.\n",
            "\n",
            "ucinds husf'l as! slres hrkh$ inu?\n",
            "cbent? hec, g\n",
            "TRAINING ON BATCH 512 to 768 (example 65536 to 98304)\n",
            "Epoch 1/1\n",
            "32768/32768 [==============================] - 125s 4ms/step - loss: 2.1718\n",
            "GENERATING SOME RANDOM TEXT FROM THE MODEL\n",
            "=====GENERATING TEXT WITH SEED: \n",
            "\"myself and thought\n",
            "this was so\"\n",
            "==========TEMPERATURE: 0.2\n",
            "myself and thought\n",
            "this was son the the the the the the the the his the the the the the the the the the the the your the hing the the the bere in the the the sond the the the the the the the the the the the the the souster the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the the sond the the the the the me the the the the the the the the the the the the\n",
            "==========TEMPERATURE: 0.5\n",
            "myself and thought\n",
            "this was soour my frath the diplour to mo the to the the the hast that wist i sing the to enged whe the mid the seling of the beren the tin the the soncest the thor souve wher thart my my to the my maing thes to ing and my wing of the beat i coun momend caren thes i so derenghas that in this of bore distus beme deat the hang the and to and sod realll as thith the the sout that the cain the deare wain the him\n",
            "==========TEMPERATURE: 1.0\n",
            "myself and thought\n",
            "this was songhy m melo this, thoa, wriinghat teaitor radll.\n",
            "\n",
            "pktilg:\n",
            "haat, theyr arnd 'rcad kim morty dourves.\n",
            "\n",
            "meisg ongter:\n",
            "am; wife berad hat, til the youce?\n",
            "akkikus,,\n",
            "and daindke:\n",
            "wthist dellad nle iit, thy poutid.\n",
            "\n",
            "bllorthi he remenk, porling hikench, whoth bild thors dye mysy take houth\n",
            "thils livery, the, rimlust;\n",
            "and thy i diznhi'.\n",
            "\n",
            "loucen:gus mens:\n",
            "any -thule ho knod any, to gravcut thouls enctou epi\n",
            "==========TEMPERATURE: 1.2\n",
            "myself and thought\n",
            "this was so mysolethon in miib.\n",
            "thing him has and ay yoy grea; ha ltorker sifrresy nlorert, hus, as kins,, tie havy, asus\n",
            "wut kha tye cinglm stalt, erives fod deady if, my nothind wiord osead me\n",
            "gpoupnten'et figenot ent.\n",
            "of.\n",
            "\n",
            "uckind prepiize:t'\n",
            "nova gith?\n",
            ", scave, mmy lovel.\n",
            "\n",
            "m'te racquu, steate tourels:\n",
            "mukths ryof i ghedse we ands neith\n",
            "yoused\n",
            "\n",
            "qucinys:\n",
            "ateruvand he yod bict?\n",
            "\n",
            "bicchelies:\n",
            "a\n",
            "lack, bupeald o\n",
            "TRAINING ON BATCH 768 to 1024 (example 98304 to 131072)\n",
            "Epoch 1/1\n",
            "32768/32768 [==============================] - 127s 4ms/step - loss: 2.1314\n",
            "GENERATING SOME RANDOM TEXT FROM THE MODEL\n",
            "=====GENERATING TEXT WITH SEED: \n",
            "\"my head and all this famous la\"\n",
            "==========TEMPERATURE: 0.2\n",
            "my head and all this famous land the rall the the with the sours and the ford the reand the the hast lord the the the be the the sour the the the the the hard the gare and my the that the the the the sord sour the hard the reand the the the the the sour the the with sour to bere the the the the come the frous the the the the the the the i with my the the the the the the the the hard the the the the the the my the with the sill\n",
            "==========TEMPERATURE: 0.5\n",
            "my head and all this famous land and to gare that the mard thour for hast the come your comes the is ar wich come he dare the flour to this of to the that but the in the row shar the sear, the with make the all the breces the cours sord a the langon then and my grouchard the leard four the for both the brome the hand, i the proke of wath rister his and and thou his the and the soun hard the grold wat ronder the serd is my my t\n",
            "==========TEMPERATURE: 1.0\n",
            "my head and all this famous lane.\n",
            "\n",
            "lorth, i ig hars y ag.r rish?yirusbal fite angy ther ioy proly to graves your kevi. focme\n",
            "hill htmy \n",
            "ratcrele.\n",
            "sing's at osssand, ands,\n",
            "bgerosem;\n",
            "gomus maims allpward ind grorbours.\n",
            "bule:\n",
            "aded dark warveans:\n",
            "to ficlir\n",
            "what to buth rok withrt ar watin:\n",
            "a thigh to fanorvet of buce.\n",
            "\n",
            "and hin yougegevh, goagells your boll that aus blerf;\n",
            "not moe lor to broke:\n",
            "bregerky hat arst?\n",
            "\n",
            "kins to kingass:\n",
            "\n",
            "==========TEMPERATURE: 1.2\n",
            "my head and all this famous lan.\n",
            "\n",
            "king thy murom:\n",
            "sor hige chrmut to mosd's i\n",
            "tou?\n",
            "\n",
            "lking-alle:\n",
            "whake jore, el witu tof of af r.iig?\n",
            "\n",
            "ing your in mond, on:\n",
            "thy' rakr'ce my and, and foty to spl-evce.\n",
            "manes:\n",
            "liz.\n",
            "\n",
            "qbukin:\n",
            "fov takn he is lewe i t och land in.\n",
            "\n",
            "engfr, nomek;\n",
            "than the tragrw\n",
            "ing aldcage the-artingnmand,\n",
            "werrstingresg farc! cos ar i an gheralm ther to green sfok.\n",
            "\n",
            "revesar'far: rawe good foun siff?\n",
            "\n",
            "meard:\n",
            "rol-sen;\n",
            "k\n",
            "TRAINING ON BATCH 1024 to 1280 (example 131072 to 163840)\n",
            "Epoch 1/1\n",
            "32768/32768 [==============================] - 125s 4ms/step - loss: 2.0923\n",
            "GENERATING SOME RANDOM TEXT FROM THE MODEL\n",
            "=====GENERATING TEXT WITH SEED: \n",
            "\"bed-chamber.\n",
            "\n",
            "lady anne:\n",
            "i'll \"\n",
            "==========TEMPERATURE: 0.2\n",
            "bed-chamber.\n",
            "\n",
            "lady anne:\n",
            "i'll of come the so so and so the the so son the love the that the the sare the so the will the love the so the the sear the so the so the sear the well the the sond and and with the gold and with the the come the the sear the so the the the sond the the the the so me the seard the the mare the son the love the ford the love i will the son the so the the will the hear the so the the the love the seard \n",
            "==========TEMPERATURE: 0.5\n",
            "bed-chamber.\n",
            "\n",
            "lady anne:\n",
            "i'll thou shat is the men that not the the mand swout thee thou the lond cime thou the leave the he some you the will not the the so com, lome a king lot sith sout, to hard leat of with that i with here of some of cother of the mane sher lewst a shall the and, with thou the and the he so the shall my so thou well soul the end waill with light be lace stour deas that so mang dom you lieg the for hat she\n",
            "==========TEMPERATURE: 1.0\n",
            "bed-chamber.\n",
            "\n",
            "lady anne:\n",
            "i'll thematomenclhe he theer dogbongart i's.\n",
            "\n",
            "uken,\n",
            "yol slich hell vear of con unoy ther tele-tace as suplown sprepe,\n",
            "a porse; i la wl bet i, ov fove to so winr cive oupke:\n",
            "that enw-n do sthat limesop the.\n",
            "conk,\n",
            "thye ion, dererule, swere.\n",
            "\n",
            "herioo:\n",
            "\n",
            "ithel:\n",
            "the me.\n",
            "\n",
            "qukeare yor ceichins graivet:\n",
            "you the qutgul -yald me?\n",
            "\n",
            "rreow\n",
            "macicheliotine:\n",
            "o xome at is mar.\n",
            "\n",
            "bint loire::\n",
            "you whe to you pare\n",
            "ast speak \n",
            "==========TEMPERATURE: 1.2\n",
            "bed-chamber.\n",
            "\n",
            "lady anne:\n",
            "i'll gugathit oi an w'reme thasy a paed-ait he owh: vary herver that you lovim,\n",
            "and wilrl'y sot mere the kulls?\n",
            "that, of is sfack yo ay if cufes, thes lack!\n",
            "a japcmel a not my my wagu.\n",
            "\n",
            "hidf riad.\n",
            "\n",
            "jolin:\n",
            "dhat whind liacem,\n",
            "i widse!\n",
            "\n",
            "rosk:\n",
            "mi eposrughe ho dont 'lutkid' of proof, 't\n",
            "honer''dughy thit.\n",
            "\n",
            "butpade iosiny.\n",
            "\n",
            "ouvook:\n",
            "my non.\n",
            "\n",
            "grinct, his: ovh hera teles have thattratcum.\n",
            "hacd, that thak maye'r\n",
            "TRAINING ON BATCH 1280 to 1536 (example 163840 to 196608)\n",
            "Epoch 1/1\n",
            "32768/32768 [==============================] - 122s 4ms/step - loss: 2.0355\n",
            "GENERATING SOME RANDOM TEXT FROM THE MODEL\n",
            "=====GENERATING TEXT WITH SEED: \n",
            "\"watchman:\n",
            "stay, or thou diest!\"\n",
            "==========TEMPERATURE: 0.2\n",
            "watchman:\n",
            "stay, or thou diest!\n",
            "\n",
            "rome:\n",
            "and i will the will the will the will the will the for the love with the will the case and the dear the the some my so with the will the frome the will the sone the bear the come the sill the will the cand the will the shall the will the will the and the will the with so the frome and with the will the will the will the will the will with the will the word and and the will the the so the w\n",
            "==========TEMPERATURE: 0.5\n",
            "watchman:\n",
            "stay, or thou diest!\n",
            "\n",
            "frist:\n",
            "and whou thou so the the the sone the pring her and deast son at of on the pear hat the sill the mare lay with so supe so me reeth\n",
            "and and thou cose the with and lead the bedsean the marsen and wull cith the raunded will the a thou so the with the thou prad to shat in frome the were well thou shall well whis and she hare ence courter, your his a sone, and the were the prease and the frien\n",
            "==========TEMPERATURE: 1.0\n",
            "watchman:\n",
            "stay, or thou diest! lad fencrem's and loven, stall for o: goped your daingoord dave,\n",
            "hen thoullly your for wist han sunbe and's gae the part.\n",
            "\n",
            "lant:\n",
            "and sing;\n",
            "i wam.\n",
            "\n",
            "seluve:\n",
            "nom ford hengorkaty thy nove:\n",
            "for a sorf came,\n",
            "buts say the wall aling withond, to uar oup.\n",
            "\n",
            "irrrom:\n",
            "isw whe forrursafs o'dt-meing and thou alling\n",
            "and gear the badtlofofo, yok dewer so'e.\n",
            "\n",
            "roaghour?\n",
            "wore is 'trid bewiss;\n",
            "a mecuribinsseang ny mr\n",
            "==========TEMPERATURE: 1.2\n",
            "watchman:\n",
            "stay, or thou diest!\n",
            "\n",
            "juvecifoo, gorse:\n",
            "werl iste comughtie thram of yoar tayrajen's fyom..o-\n",
            "myuld:\n",
            "love:\n",
            "on fithawinguss-either\n",
            "of culeonoig;\n",
            "anw\n",
            "an ve sheen-adnet'ce.\n",
            "\n",
            "jupmry:\n",
            "jeadqher:\n",
            "anv so grome, is darnes! and is aidt,\n",
            "and by crourspder', me oi ky,\n",
            "joon,\n",
            "thir therse,\n",
            "avend--anens ulfsocch flawdlace to thase apleing?\n",
            "\n",
            "nrocbaver:\n",
            "sels-m\n",
            "an? yon!\n",
            "\n",
            "kprowe:\n",
            "ir you lesty on ast,s purplay.\n",
            "erd:\n",
            "and king and, brave t\n",
            "TRAINING ON BATCH 1536 to 1792 (example 196608 to 229376)\n",
            "Epoch 1/1\n",
            "24320/32768 [=====================>........] - ETA: 31s - loss: 1.9848"
          ],
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0mTraceback (most recent call last)",
            "\u001b[0;32m<ipython-input-72-ff3e59dfe5f3>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     28\u001b[0m               \u001b[0mbatch_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mbatch_size\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     29\u001b[0m               \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 30\u001b[0;31m               verbose=1)\n\u001b[0m\u001b[1;32m     31\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     32\u001b[0m     \u001b[0;32mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"GENERATING SOME RANDOM TEXT FROM THE MODEL\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/training.pyc\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)\u001b[0m\n\u001b[1;32m   1346\u001b[0m           \u001b[0minitial_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0minitial_epoch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1347\u001b[0m           \u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1348\u001b[0;31m           validation_steps=validation_steps)\n\u001b[0m\u001b[1;32m   1349\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1350\u001b[0m       return training_arrays.fit_loop(\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/training_eager.pyc\u001b[0m in \u001b[0;36mfit_loop\u001b[0;34m(model, inputs, targets, sample_weights, class_weight, val_inputs, val_targets, val_sample_weights, batch_size, epochs, verbose, callbacks, shuffle, callback_metrics, initial_epoch, steps_per_epoch, validation_steps)\u001b[0m\n\u001b[1;32m   1056\u001b[0m             \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mshuffle\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1057\u001b[0m             \u001b[0mnum_train_samples\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mnum_train_samples\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1058\u001b[0;31m             do_validation=do_validation)\n\u001b[0m\u001b[1;32m   1059\u001b[0m       \u001b[0mcallbacks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mon_epoch_end\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mepoch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepoch_logs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1060\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0mcallback_model\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstop_training\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/training_eager.pyc\u001b[0m in \u001b[0;36mbatch_fit_loop\u001b[0;34m(model, inputs, targets, epoch_logs, index_array, out_labels, callback_model, batch_size, sample_weights, val_inputs, val_targets, val_sample_weights, callbacks, shuffle, num_train_samples, do_validation)\u001b[0m\n\u001b[1;32m    397\u001b[0m         \u001b[0mtargets_batch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    398\u001b[0m         \u001b[0msample_weights\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msample_weights_batch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 399\u001b[0;31m         training=True)\n\u001b[0m\u001b[1;32m    400\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    401\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlist\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/training_eager.pyc\u001b[0m in \u001b[0;36m_process_single_batch\u001b[0;34m(model, inputs, targets, sample_weights, training)\u001b[0m\n\u001b[1;32m    792\u001b[0m       outs, loss, loss_metrics = _model_loss(model, inputs, targets,\n\u001b[1;32m    793\u001b[0m                                              \u001b[0msample_weights\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msample_weights\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 794\u001b[0;31m                                              training=training)\n\u001b[0m\u001b[1;32m    795\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0mloss\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    796\u001b[0m         raise ValueError('The model cannot be run '\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/training_eager.pyc\u001b[0m in \u001b[0;36m_model_loss\u001b[0;34m(model, inputs, targets, sample_weights, training)\u001b[0m\n\u001b[1;32m    126\u001b[0m       \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcall\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtraining\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtraining\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    127\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 128\u001b[0;31m       \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcall\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    129\u001b[0m   \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    130\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_expects_training_arg\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/network.pyc\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, inputs, training, mask)\u001b[0m\n\u001b[1;32m    775\u001b[0m     outputs, _ = self._run_internal_graph(inputs,\n\u001b[1;32m    776\u001b[0m                                           \u001b[0mtraining\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mtraining\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 777\u001b[0;31m                                           mask=masks)\n\u001b[0m\u001b[1;32m    778\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0moutputs\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    779\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/engine/network.pyc\u001b[0m in \u001b[0;36m_run_internal_graph\u001b[0;34m(self, inputs, training, mask)\u001b[0m\n\u001b[1;32m    946\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    947\u001b[0m               output_tensors = nest.flatten(\n\u001b[0;32m--> 948\u001b[0;31m                   layer.call(computed_tensor, **kwargs))\n\u001b[0m\u001b[1;32m    949\u001b[0m               \u001b[0;32mif\u001b[0m \u001b[0mhasattr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlayer\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'compute_mask'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    950\u001b[0m                 output_masks = layer.compute_mask(computed_tensor,\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/layers/recurrent.pyc\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, inputs, mask, training, initial_state)\u001b[0m\n\u001b[1;32m   2097\u001b[0m     \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcell\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_recurrent_dropout_mask\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2098\u001b[0m     return super(LSTM, self).call(\n\u001b[0;32m-> 2099\u001b[0;31m         inputs, mask=mask, training=training, initial_state=initial_state)\n\u001b[0m\u001b[1;32m   2100\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2101\u001b[0m   \u001b[0;34m@\u001b[0m\u001b[0mproperty\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/layers/recurrent.pyc\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, inputs, mask, training, initial_state, constants)\u001b[0m\n\u001b[1;32m    637\u001b[0m         \u001b[0mmask\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mmask\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    638\u001b[0m         \u001b[0munroll\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munroll\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 639\u001b[0;31m         input_length=timesteps)\n\u001b[0m\u001b[1;32m    640\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstateful\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    641\u001b[0m       \u001b[0mupdates\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/backend.pyc\u001b[0m in \u001b[0;36mrnn\u001b[0;34m(step_function, inputs, initial_states, go_backwards, mask, constants, unroll, input_length)\u001b[0m\n\u001b[1;32m   3218\u001b[0m         \u001b[0mmaximum_iterations\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0minput_length\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3219\u001b[0m         \u001b[0mparallel_iterations\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m32\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3220\u001b[0;31m         swap_memory=True)\n\u001b[0m\u001b[1;32m   3221\u001b[0m     \u001b[0mlast_time\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfinal_outputs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3222\u001b[0m     \u001b[0moutput_ta\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfinal_outputs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/control_flow_ops.pyc\u001b[0m in \u001b[0;36mwhile_loop\u001b[0;34m(cond, body, loop_vars, shape_invariants, parallel_iterations, back_prop, swap_memory, name, maximum_iterations, return_same_structure)\u001b[0m\n\u001b[1;32m   3206\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3207\u001b[0m       \u001b[0;32mwhile\u001b[0m \u001b[0mcond\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mloop_vars\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3208\u001b[0;31m         \u001b[0mloop_vars\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mbody\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mloop_vars\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3209\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mtry_to_pack\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloop_vars\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mlist\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0m_basetuple\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3210\u001b[0m           \u001b[0mpacked\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mTrue\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/control_flow_ops.pyc\u001b[0m in \u001b[0;36m<lambda>\u001b[0;34m(i, lv)\u001b[0m\n\u001b[1;32m   3199\u001b[0m         cond = lambda i, lv: (  # pylint: disable=g-long-lambda\n\u001b[1;32m   3200\u001b[0m             math_ops.logical_and(i < maximum_iterations, orig_cond(*lv)))\n\u001b[0;32m-> 3201\u001b[0;31m         \u001b[0mbody\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mlambda\u001b[0m \u001b[0mi\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlv\u001b[0m\u001b[0;34m:\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mi\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0morig_body\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mlv\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3202\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3203\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mcontext\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexecuting_eagerly\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/backend.pyc\u001b[0m in \u001b[0;36m_step\u001b[0;34m(time, output_ta_t, *states)\u001b[0m\n\u001b[1;32m   3203\u001b[0m         \u001b[0mcurrent_input\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0minput_ta\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mread\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtime\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3204\u001b[0m         output, new_states = step_function(current_input,\n\u001b[0;32m-> 3205\u001b[0;31m                                            tuple(states) + tuple(constants))\n\u001b[0m\u001b[1;32m   3206\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mgetattr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0moutput\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'_uses_learning_phase'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3207\u001b[0m           \u001b[0;32mglobal\u001b[0m \u001b[0muses_learning_phase\u001b[0m  \u001b[0;31m# pylint: disable=global-variable-undefined\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/layers/recurrent.pyc\u001b[0m in \u001b[0;36mstep\u001b[0;34m(inputs, states)\u001b[0m\n\u001b[1;32m    627\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    628\u001b[0m       \u001b[0;32mdef\u001b[0m \u001b[0mstep\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mstates\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 629\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcell\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcall\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mstates\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    630\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    631\u001b[0m     last_output, outputs, states = K.rnn(\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/layers/recurrent.pyc\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, inputs, states, training)\u001b[0m\n\u001b[1;32m   1870\u001b[0m       \u001b[0mx_f\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mK\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs_f\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mkernel\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munits\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munits\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1871\u001b[0m       \u001b[0mx_c\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mK\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs_c\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mkernel\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munits\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munits\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0;36m3\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1872\u001b[0;31m       \u001b[0mx_o\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mK\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minputs_o\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mkernel\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munits\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0;36m3\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1873\u001b[0m       \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0muse_bias\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1874\u001b[0m         \u001b[0mx_i\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mK\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbias_add\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_i\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbias\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munits\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/backend.pyc\u001b[0m in \u001b[0;36mdot\u001b[0;34m(x, y)\u001b[0m\n\u001b[1;32m   1336\u001b[0m   \u001b[0;34m`\u001b[0m\u001b[0;34m`\u001b[0m\u001b[0;34m`\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1337\u001b[0m   \"\"\"\n\u001b[0;32m-> 1338\u001b[0;31m   \u001b[0;32mif\u001b[0m \u001b[0mndim\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mNone\u001b[0m \u001b[0;32mand\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mndim\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0;36m2\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0mndim\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1339\u001b[0m     \u001b[0mx_shape\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1340\u001b[0m     \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0ms\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mzip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mint_shape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0marray_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0munstack\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0marray_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/keras/backend.pyc\u001b[0m in \u001b[0;36mndim\u001b[0;34m(x)\u001b[0m\n\u001b[1;32m    881\u001b[0m   \u001b[0;34m`\u001b[0m\u001b[0;34m`\u001b[0m\u001b[0;34m`\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    882\u001b[0m   \"\"\"\n\u001b[0;32m--> 883\u001b[0;31m   \u001b[0mdims\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_shape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_dims\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    884\u001b[0m   \u001b[0;32mif\u001b[0m \u001b[0mdims\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    885\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdims\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.pyc\u001b[0m in \u001b[0;36mget_shape\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    819\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0mget_shape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    820\u001b[0m     \u001b[0;34m\"\"\"Alias of Tensor.shape.\"\"\"\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 821\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    822\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    823\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_shape_as_list\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.pyc\u001b[0m in \u001b[0;36mshape\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    814\u001b[0m       \u001b[0;31m# `_tensor_shape` is declared and defined in the definition of\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    815\u001b[0m       \u001b[0;31m# `EagerTensor`, in C.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 816\u001b[0;31m       \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_tensor_shape\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtensor_shape\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mTensorShape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_shape_tuple\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    817\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_tensor_shape\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    818\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    },
    {
      "metadata": {
        "id": "w3C-AIU18HFe",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "**NOTE**: Even after training has stopped you can still generate from the (partially trained) model as follows:"
      ]
    },
    {
      "metadata": {
        "id": "7HR4JXD3ZnX0",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "82fbdad8-0586-4096-f5a1-7c1cc9bfcee4"
      },
      "cell_type": "code",
      "source": [
        "my_text = \"       the meaning of life is:\"  # Needs to be max_len characters\n",
        "print(len(my_text))\n",
        "sample_from_model(model, prev_text=my_text)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "30\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "UbdLbn_xRe21",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "###IMPORTANT NOTES\n",
        "* Even if you stop training the model weights are persistant. If you resume training it will start where you left off. \n",
        "* To reset the weights, you need to recompile the model.\n",
        "* Sampling is **stochastic** (random), so you'll get new outputs every time you rerun the sampling code.\n",
        "\n",
        "### YOUR TASKS: \n",
        "* [**ALL**] Read the generations from your model in a funny voice to your neighbour.\n",
        "* [**ALL**] Increase `max_len` and regenerate the data and retrain the model.\n",
        " * What's the effect on training speed as you double `max_len`. Can you explain why?\n",
        " * Do you notice any effect on the quality of the model? Can you explain why?\n",
        "* [**ALL**] Set the max_len to be roughly the length of half a word; one word; two words... What kind of samples do these models generate? Explain how they go wrong.\n",
        "* [**ALL**] Change `embedding_dim` and the hidden size of the LSTM and observe the effect on training speed and quality.\n",
        "* [**INTERMEDIATE**] Change the dropout rates & retrain the model. \n",
        "  * What types of dropout do we get for recurrent models? \n",
        "  * What's the effect on the text quality?\n",
        "* [**ADVANCED**] Implement the \"teacher forcing\" training methodology, where the net must predict the entire output sequence shifted forward by one character, instead of just the next character. Compare the output of a model trained with teacher forcing versus the per-character model, given a similar training time. Is it fair to say that teacher forcing is a more efficient training methodology?\n"
      ]
    },
    {
      "metadata": {
        "id": "REZzoUwjHkZW",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "##Further reading\n",
        "\n",
        "* https://distill.pub/2016/augmented-rnns/\n",
        "* https://distill.pub/2017/ctc/\n",
        "* https://algotravelling.com/en/machine-learning-fun-part-5/"
      ]
    }
  ]
}