2_keras_mnist_advanced.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Copy of keras_mnist_2.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/peterjsadowski/keras_tutorial/blob/master/2_keras_mnist_advanced.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "metadata": {
        "id": "RPbBpk1NzpQ1",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Advanced Keras Tutorial\n",
        "This tutorial introduces:\n",
        "\n",
        "1) Using the GPU\n",
        "\n",
        "2) Data generators in Keras.\n",
        "\n",
        "3) Functional API for complex models.\n",
        " \n"
      ]
    },
    {
      "metadata": {
        "id": "B8zOyh7XzpQ3",
        "colab_type": "code",
        "outputId": "29492927-cead-4f65-ce42-8de7b0064886",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 50
        }
      },
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "import numpy.random\n",
        "import matplotlib.pyplot as plt\n",
        "import os, sys, socket\n",
        "\n",
        "import tensorflow as tf\n",
        "try:\n",
        "    device_name = tf.test.gpu_device_name()\n",
        "    if device_name != '/device:GPU:0':\n",
        "        raise SystemError('GPU device not found')\n",
        "    print('Found GPU at: {}'.format(device_name))\n",
        "except:\n",
        "    print('No GPU found.')\n",
        "\n",
        "import keras\n",
        "from keras.datasets import mnist\n",
        "from keras.models import Model\n",
        "from keras.layers import Dense, Dropout, Activation, Input, Conv2D, Flatten\n",
        "import keras.layers.merge as merge\n",
        "from keras.layers.merge import Concatenate\n",
        "from keras.utils import np_utils\n",
        "from keras.optimizers import SGD, Adam"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "No GPU found.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "metadata": {
        "id": "lrm9TejgzpQ9",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Define data generators for training and validation."
      ]
    },
    {
      "metadata": {
        "id": "vBOZOXe_zpQ-",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# The following function is used to serve up both train and validation data.\n",
        "def data_generator(X, Y, batchsize):\n",
        "    nb_classes = 10\n",
        "    N = X.shape[0]\n",
        "    while True:\n",
        "        indices1 = np.random.randint(low=0, high=N, size=(batchsize,))\n",
        "        indices2 = np.random.randint(low=0, high=N, size=(batchsize,))\n",
        "        \n",
        "        X1 = X[indices1,...].astype('float32') / 255.0\n",
        "        X2 = X[indices2,...].astype('float32') / 255.0\n",
        "        Y1 = Y[indices1]\n",
        "        Y2 = Y[indices2]\n",
        "        T  = (Y1 + Y2) # Sum of values.\n",
        "        \n",
        "        X1 = np.expand_dims(X1, axis=1) # For conv with theano, shape=(batchsize, channels, row, col).\n",
        "        X2 = np.expand_dims(X2, axis=1) # We are just adding a dummy dimension saying that there is one channel.\n",
        "        \n",
        "        Y1 = np_utils.to_categorical(Y1, nb_classes)\n",
        "        Y2 = np_utils.to_categorical(Y2, nb_classes)\n",
        "        T  = np_utils.to_categorical(T, 19) # 19 possible values.\n",
        "        \n",
        "        yield {'input1':X1 , 'input2':X2},  {'out':T, 'aux1':Y1, 'aux2':Y2}\n",
        "        \n",
        "# Load data.\n",
        "(X_train, Y_train), (X_valid, Y_valid) = mnist.load_data() # Shape = (N,28,28), (N,)\n",
        "# Create generators.\n",
        "batchsize  = 200 \n",
        "data_train = data_generator(X_train, Y_train, batchsize)\n",
        "data_valid = data_generator(X_valid, Y_valid, batchsize)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Gui8KjR-zpRB",
        "colab_type": "code",
        "outputId": "817d21fc-447c-4536-f1b6-5ffb0c16c78e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 155
        }
      },
      "cell_type": "code",
      "source": [
        "# Plot examples of the data.\n",
        "inputs, outputs = data_train.__next__()\n",
        "Xbatch = inputs['input1']\n",
        "plt.figure(1, figsize=(14,2))\n",
        "for i in range(10):\n",
        "    plt.subplot(2,10,i+1)\n",
        "    plt.imshow(inputs['input1'][i].reshape(28,28), cmap='gray', interpolation='nearest')\n",
        "    plt.xticks([])\n",
        "    plt.yticks([])\n",
        "    plt.title('Target=%d' % np.argmax(outputs['out'][i]))\n",
        "    plt.subplot(2,10,10+i+1)\n",
        "    plt.imshow(inputs['input2'][i].reshape(28,28), cmap='gray', interpolation='nearest')\n",
        "    plt.xticks([])\n",
        "    plt.yticks([])"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAw0AAACKCAYAAADovKYdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrkJggg==\n",
            "text/plain": [
              "<Figure size 1008x144 with 20 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "rNfItrsTzpRF",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Define Keras Model with the Functional API\n"
      ]
    },
    {
      "metadata": {
        "id": "dGg_p1h2zpRH",
        "colab_type": "code",
        "outputId": "2ac5fbdc-837f-49f6-ea0f-0e008078877f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 558
        }
      },
      "cell_type": "code",
      "source": [
        "# Input data.\n",
        "input1   = Input(shape=(1, 28,28), dtype='float32', name='input1') # Argument 'name' must match name in dictionary.\n",
        "input2   = Input(shape=(1, 28,28), dtype='float32', name='input2')\n",
        "\n",
        "# Layer one: Single convolutional layer applied to each input independently, with \"shared weights.\"\n",
        "nb_filter      = 32     # Number of convolutional kernels.\n",
        "nb_row, nb_col = 7,7   # Convolution kernel size.\n",
        "strides        = (3,3) # Step size for convolution kernels.\n",
        "conv  =  Conv2D(nb_filter, (nb_row, nb_col), activation='relu', padding='same', strides=strides)\n",
        "x1    = conv(input1) # Layer object conv transforms data.\n",
        "x2    = conv(input2)\n",
        "\n",
        "# Flatten data: transform from (28,28) to (784,)\n",
        "x1 = Flatten()(x1)\n",
        "x2 = Flatten()(x2)\n",
        "\n",
        "# Layer two: Single fully-connected layer applied to each input independently, with shared weights.\n",
        "layer = Dense(256, activation='relu')\n",
        "x1 = layer(x1)\n",
        "x2 = layer(x2)\n",
        "\n",
        "# Auxiliary softmax layers try to classify digits. \n",
        "# The output of these layers aren't fed into the next layer.\n",
        "#layer = Dense(10) # These weights are shared.\n",
        "#aux1  = Activation(activation='softmax', name='aux1')(layer(x1)) # Output layers must be named.\n",
        "#aux2  = Activation(activation='softmax', name='aux2')(layer(x2))\n",
        "\n",
        "# Merge hidden representations.\n",
        "#x = merge.Concatenate()([x1, x2])\n",
        "x = Concatenate()([x1, x2])\n",
        "\n",
        "# More dense layers then output.\n",
        "x   = Dense(512, activation='relu')(x)\n",
        "out = Dense(19, activation='softmax', name='out')(x) # Output layers must be named.\n",
        "\n",
        "# Create model object that puts it all together.\n",
        "#model = Model(inputs=[input1, input2], outputs=[out, aux1, aux2])\n",
        "model = Model(inputs=[input1, input2], outputs=[out])\n",
        "model.summary()"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Colocations handled automatically by placer.\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input1 (InputLayer)             (None, 1, 28, 28)    0                                            \n",
            "__________________________________________________________________________________________________\n",
            "input2 (InputLayer)             (None, 1, 28, 28)    0                                            \n",
            "__________________________________________________________________________________________________\n",
            "conv2d_1 (Conv2D)               (None, 1, 10, 32)    43936       input1[0][0]                     \n",
            "                                                                 input2[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "flatten_1 (Flatten)             (None, 320)          0           conv2d_1[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "flatten_2 (Flatten)             (None, 320)          0           conv2d_1[1][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dense_1 (Dense)                 (None, 256)          82176       flatten_1[0][0]                  \n",
            "                                                                 flatten_2[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "concatenate_1 (Concatenate)     (None, 512)          0           dense_1[0][0]                    \n",
            "                                                                 dense_1[1][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "dense_2 (Dense)                 (None, 512)          262656      concatenate_1[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "out (Dense)                     (None, 19)           9747        dense_2[0][0]                    \n",
            "==================================================================================================\n",
            "Total params: 398,515\n",
            "Trainable params: 398,515\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "nXH5jZGHzpRL",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Optimization"
      ]
    },
    {
      "metadata": {
        "id": "j_KuS3jSzpRM",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "optimizer = Adam(lr=0.001, beta_1=0.99, beta_2=0.999, epsilon=1e-08) # Optimization hyperparameters.\n",
        "\n",
        "model.compile(optimizer=optimizer,\n",
        "              loss={'out':'categorical_crossentropy'},\n",
        "              loss_weights={'out': 1.0}, # \n",
        "              metrics=['accuracy'])\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "HMuCALu9zpRP",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Train the model."
      ]
    },
    {
      "metadata": {
        "id": "hkgjaFS-zpRQ",
        "colab_type": "code",
        "outputId": "4520f2a4-2f8f-40cc-867e-5735593a4e3b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 423
        }
      },
      "cell_type": "code",
      "source": [
        "# Callbacks can be used to stop early, decrease learning rate, checkpoint the model, etc.\n",
        "#from keras.callbacks import EarlyStopping\n",
        "#stopping  = keras.callbacks.EarlyStopping(monitor='val_loss', patience=0, verbose=0, mode='auto')\n",
        "callbacks = []#[stopping]\n",
        "\n",
        "# The fit_generator function loads data batches on the fly, instead of transfering entire data set to the gpu.\n",
        "history   = model.fit_generator(generator=data_train, steps_per_epoch=100, # 500\n",
        "                                epochs=10, verbose=1,\n",
        "                                callbacks=callbacks, \n",
        "                                validation_data=data_valid, validation_steps=100)"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Use tf.cast instead.\n",
            "Epoch 1/10\n",
            "100/100 [==============================] - 7s 74ms/step - loss: 2.5282 - acc: 0.1630 - val_loss: 2.1615 - val_acc: 0.2457\n",
            "Epoch 2/10\n",
            "100/100 [==============================] - 8s 75ms/step - loss: 1.8280 - acc: 0.3599 - val_loss: 1.4547 - val_acc: 0.5298\n",
            "Epoch 3/10\n",
            "100/100 [==============================] - 7s 75ms/step - loss: 1.1171 - acc: 0.6488 - val_loss: 0.8004 - val_acc: 0.7674\n",
            "Epoch 4/10\n",
            "100/100 [==============================] - 8s 77ms/step - loss: 0.6590 - acc: 0.8048 - val_loss: 0.5271 - val_acc: 0.8530\n",
            "Epoch 5/10\n",
            "100/100 [==============================] - 8s 79ms/step - loss: 0.4533 - acc: 0.8708 - val_loss: 0.3860 - val_acc: 0.8888\n",
            "Epoch 6/10\n",
            "100/100 [==============================] - 8s 77ms/step - loss: 0.3450 - acc: 0.9023 - val_loss: 0.3087 - val_acc: 0.9102\n",
            "Epoch 7/10\n",
            "100/100 [==============================] - 8s 75ms/step - loss: 0.2938 - acc: 0.9156 - val_loss: 0.2604 - val_acc: 0.9231\n",
            "Epoch 8/10\n",
            "100/100 [==============================] - 8s 76ms/step - loss: 0.2277 - acc: 0.9346 - val_loss: 0.2181 - val_acc: 0.9338\n",
            "Epoch 9/10\n",
            "100/100 [==============================] - 8s 77ms/step - loss: 0.2012 - acc: 0.9410 - val_loss: 0.1984 - val_acc: 0.9396\n",
            "Epoch 10/10\n",
            "100/100 [==============================] - 8s 77ms/step - loss: 0.1710 - acc: 0.9509 - val_loss: 0.1725 - val_acc: 0.9469\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "Vgj-AWt_zpRU",
        "colab_type": "code",
        "outputId": "11d4ce30-4db3-4d79-a1b4-d57ff2e14855",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        }
      },
      "cell_type": "code",
      "source": [
        "# Plot loss trajectory throughout training.\n",
        "plt.figure(1, figsize=(10,5))\n",
        "plt.plot(history.history['acc'], label='primary')\n",
        "plt.xlabel('Epoch')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.legend()"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7f66a249b550>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 11
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAl8AAAE9CAYAAAA1awfRAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ninmo56IkAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 720x360 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "L6doGuvdzpRY",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Post-training analysis."
      ]
    },
    {
      "metadata": {
        "id": "s1hFaHsOzpRZ",
        "colab_type": "code",
        "outputId": "cf859a38-9ca7-4e5b-be04-a4ea7d5b74b4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 224
        }
      },
      "cell_type": "code",
      "source": [
        "# Plot example pairs for which the network made a mistake.\n",
        "inputs, outputs = data_valid.__next__()\n",
        "plt.figure(1, figsize=(14,3))\n",
        "i = 0\n",
        "while i < 10:\n",
        "    inputs, outputs = data_valid.__next__()\n",
        "    predictions = model.predict(inputs)\n",
        "    j = 0\n",
        "    while j < 100:\n",
        "        yhat   = predictions[j].argmax(axis=-1) # predictions is list: [out, aux1, aux2]\n",
        "        y      = np.argmax(outputs['out'][j])\n",
        "        input1 = inputs['input1'][j].reshape(28,28)\n",
        "        input2 = inputs['input2'][j].reshape(28,28)\n",
        "        j += 1\n",
        "        if yhat != y:\n",
        "            # Prediction was correct. Ignore this example.\n",
        "            continue\n",
        "        else:\n",
        "            # Prediction was incorrect. Show example.\n",
        "            plt.subplot(2,10,i+1)\n",
        "            plt.imshow(input1, cmap='gray', interpolation='nearest')\n",
        "            plt.xticks([])\n",
        "            plt.yticks([])\n",
        "            plt.title('Pred=%d' % yhat)\n",
        "            plt.title('$y=%d$\\n$\\hat{y}=%d$' % (y,yhat))\n",
        "            plt.subplot(2,10,10+i+1)\n",
        "            plt.imshow(input2, cmap='gray', interpolation='nearest')\n",
        "            plt.xticks([])\n",
        "            plt.yticks([])\n",
        "            i += 1\n",
        "        if not i<10:\n",
        "            break\n",
        "        "
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAxsAAADPCAYAAACdmcxtAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nQjcbIiIiIiISCN1siIiIiIhIIP4fyAp0bNGpbI4AAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 1008x216 with 20 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}