workflow/pred_temperature_LSTM.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predict 2 meter temperature with sea surface temperature using LSTM\n",
    "This notebook serves as an example of a basic workflow of data driven forecasting using deep learning with `s2spy` & `lilio` packages. <br>\n",
    "We will predict temperature in US at seasonal time scales using ERA5 dataset with LSTM network. <br>\n",
    "\n",
    "<img src=\"../assets/concept_test_case.png\" alt=\"usecase\" width=\"500\"/>"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This recipe includes the following steps:\n",
    "- Define a calendar (`lilio`)\n",
    "- Download/load input data (`era5cli`) (test data, accessible via `era5cli`)\n",
    "- Map the calendar to the data (`lilio`)\n",
    "- Train-validate-test split (60%/20%/20%)\n",
    "- Preprocessing based on the training set (`s2spy`)\n",
    "- Resample data to the calendar (`lilio`)\n",
    "- Create LSTM model (`torch`)\n",
    "- Specify hyper-parameters (`wandb`)\n",
    "- Train model (`torch`)\n",
    "- Evaludate model"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The workflow is illustrated below:\n",
    "\n",
    "<img src=\"../assets/dl.PNG\" alt=\"Transformer\" width=\"900\"/>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import lilio\n",
    "import numpy as np\n",
    "import time as tt\n",
    "import wandb\n",
    "import sys\n",
    "import urllib\n",
    "import xarray as xr\n",
    "from pathlib import Path\n",
    "from s2spy import preprocess\n",
    "import torch\n",
    "from torch import nn\n",
    "from torch.autograd import Variable\n",
    "\n",
    "# import utils function to check the statistics of parameters\n",
    "sys.path.append(\"../src/\")\n",
    "import utils\n"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Define a calendar with `lilio` to specify time range for targets and precursors."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create custom calendar based on the time of interest\n",
    "calendar = lilio.Calendar(anchor=\"08-01\")\n",
    "# add target periods\n",
    "calendar.add_intervals(\"target\", length=\"30d\")\n",
    "# add precursor periods\n",
    "periods_of_interest = 4\n",
    "calendar.add_intervals(\"precursor\", \"1M\", gap=\"1M\", n=periods_of_interest)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Calendar(\n",
       "    anchor='08-01',\n",
       "    allow_overlap=False,\n",
       "    mapping=None,\n",
       "    intervals=[\n",
       "        Interval(role='target', length='30d', gap='0d'),\n",
       "        Interval(role='precursor', length='1M', gap='1M'),\n",
       "        Interval(role='precursor', length='1M', gap='1M'),\n",
       "        Interval(role='precursor', length='1M', gap='1M'),\n",
       "        Interval(role='precursor', length='1M', gap='1M')\n",
       "    ]\n",
       ")"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# check calendar\n",
    "calendar"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Load test data SST and (clustered) T2M\n",
    "For the sake of batch size, we use 61 years (1961-2021) of data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('t2m_daily_1959-2021_2deg_clustered_226_300E_30_70N.nc',\n",
       " <http.client.HTTPMessage at 0x2a36283d0>)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# URL of the dataset from zenodo\n",
    "sst_url = \"https://zenodo.org/record/8186914/files/sst_daily_1959-2021_5deg_Pacific_175_240E_25_50N.nc\"\n",
    "t2m_url = \"https://zenodo.org/record/8186914/files/t2m_daily_1959-2021_2deg_clustered_226_300E_30_70N.nc\"\n",
    "sst_field = \"sst_daily_1959-2021_5deg_Pacific_175_240E_25_50N.nc\"\n",
    "t2m_field = \"t2m_daily_1959-2021_2deg_clustered_226_300E_30_70N.nc\"\n",
    "\n",
    "urllib.request.urlretrieve(sst_url, sst_field)\n",
    "urllib.request.urlretrieve(t2m_url, t2m_field)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load data\n",
    "precursor_field = xr.open_dataset(sst_field)\n",
    "precursor_field = precursor_field.sel(time=slice(\"19610101\",\"20211231\"))\n",
    "target_field = xr.open_dataset(t2m_field)\n",
    "target_field = target_field.sel(time=slice(\"19610101\",\"20211231\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Convert Kelvin to Celcius\n",
    "precursor_field[\"sst\"] = precursor_field[\"sst\"] - 273.15\n",
    "target_field[\"t2m\"] = target_field[\"t2m\"] - 273.15"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Map the calendar to the data\n",
    "After mapping the calendar to the field, we can visualize our calendar by calling the `visualize` method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 700x400 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# map calendar to data\n",
    "calendar.map_to_data(precursor_field)\n",
    "calendar.visualize(show_length=True)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Also, we can get a list of all intervals by running the following line. There, you will find the intervals `-1` and `1`, which corresponds to the creation of a precursor interval (negative integer(s)) and a target interval (positive integer(s)), respectively. <br>\n",
    "\n",
    "For more information about the definition of intervals, and how `lilio` works, please check the [README](https://github.com/AI4S2S/lilio) of `lilio`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>i_interval</th>\n",
       "      <th>-4</th>\n",
       "      <th>-3</th>\n",
       "      <th>-2</th>\n",
       "      <th>-1</th>\n",
       "      <th>1</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>anchor_year</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2021</th>\n",
       "      <td>[2020-12-01, 2021-01-01)</td>\n",
       "      <td>[2021-02-01, 2021-03-01)</td>\n",
       "      <td>[2021-04-01, 2021-05-01)</td>\n",
       "      <td>[2021-06-01, 2021-07-01)</td>\n",
       "      <td>[2021-08-01, 2021-08-31)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2020</th>\n",
       "      <td>[2019-12-01, 2020-01-01)</td>\n",
       "      <td>[2020-02-01, 2020-03-01)</td>\n",
       "      <td>[2020-04-01, 2020-05-01)</td>\n",
       "      <td>[2020-06-01, 2020-07-01)</td>\n",
       "      <td>[2020-08-01, 2020-08-31)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2019</th>\n",
       "      <td>[2018-12-01, 2019-01-01)</td>\n",
       "      <td>[2019-02-01, 2019-03-01)</td>\n",
       "      <td>[2019-04-01, 2019-05-01)</td>\n",
       "      <td>[2019-06-01, 2019-07-01)</td>\n",
       "      <td>[2019-08-01, 2019-08-31)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "i_interval                         -4                        -3  \\\n",
       "anchor_year                                                       \n",
       "2021         [2020-12-01, 2021-01-01)  [2021-02-01, 2021-03-01)   \n",
       "2020         [2019-12-01, 2020-01-01)  [2020-02-01, 2020-03-01)   \n",
       "2019         [2018-12-01, 2019-01-01)  [2019-02-01, 2019-03-01)   \n",
       "\n",
       "i_interval                         -2                        -1  \\\n",
       "anchor_year                                                       \n",
       "2021         [2021-04-01, 2021-05-01)  [2021-06-01, 2021-07-01)   \n",
       "2020         [2020-04-01, 2020-05-01)  [2020-06-01, 2020-07-01)   \n",
       "2019         [2019-04-01, 2019-05-01)  [2019-06-01, 2019-07-01)   \n",
       "\n",
       "i_interval                          1  \n",
       "anchor_year                            \n",
       "2021         [2021-08-01, 2021-08-31)  \n",
       "2020         [2020-08-01, 2020-08-31)  \n",
       "2019         [2019-08-01, 2019-08-31)  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "calendar.show()[:3]"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Train-validate-test split based on the anchor years (60%/20%/20% split)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get 60% of instance as training\n",
    "years = sorted(calendar.get_intervals().index)\n",
    "train_samples = round(len(years) * 0.6)\n",
    "test_samples = round(len(years) * 0.2)\n",
    "start_year = years[0]"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Fit preprocessor with training samples and preprocess data\n",
    "In this step, we remove trend and take anomalies for the precursor field. Note that here we use raw daily data for detrending and taking anomalies. <br>\n",
    "\n",
    "In general, there are many \"flavors\" of preprocessing, like when to perform this operation, and in which order do we want to preprocess the data. To improve the transparency and reproducibility of our work, we think it is necessary to standardize these steps. To stick to the best practices, we suggest to preprocess your data in the following way."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create preprocessor\n",
    "preprocessor = preprocess.Preprocessor(\n",
    "    rolling_window_size=25,\n",
    "    timescale=\"monthly\",\n",
    "    detrend=\"linear\",\n",
    "    subtract_climatology=True,\n",
    ")\n",
    "\n",
    "# fit preprocessor with training data\n",
    "preprocessor.fit(\n",
    "    precursor_field.sel(\n",
    "        time=slice(str(start_year), str(start_year + train_samples - 1))\n",
    "    )\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# preprocess the whole precursor field\n",
    "precursor_field_prep = preprocessor.transform(precursor_field)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Resample data to the calendar"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "precursor_field_resample = lilio.resample(calendar, precursor_field_prep)\n",
    "target_field_resample = lilio.resample(calendar, target_field)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# select variables and intervals\n",
    "precursor_field_sel = precursor_field_resample['sst']\n",
    "target_series_sel = target_field_resample['t2m'].sel(cluster=3)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We need to convert our data to `torch.Tensor`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# slice and reshape input desired by transformer\n",
    "sequence_precursor = len(precursor_field_sel.i_interval) - 1 # we only take precursor parts of i intervals\n",
    "lat_precursor = len(precursor_field_sel.latitude)\n",
    "lon_precursor = len(precursor_field_sel.longitude)\n",
    "\n",
    "X_torch = torch.from_numpy(precursor_field_sel[:,:-1,:,:].data).type(torch.FloatTensor)\n",
    "y_torch = torch.from_numpy(target_series_sel[:,-1].data).type(torch.FloatTensor)\n",
    "\n",
    "X_torch = X_torch.view(-1, sequence_precursor, lat_precursor*lon_precursor)\n",
    "\n",
    "# turn nan to 0.0\n",
    "X_torch = torch.nan_to_num(X_torch, 0.0)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We split our data into train/cross-validate/test sets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# train/validate/test split and use pytorch dataloader\n",
    "train_X_torch = X_torch[:train_samples]\n",
    "train_y_torch = y_torch[:train_samples]\n",
    "\n",
    "valid_X_torch = X_torch[train_samples:train_samples + test_samples]\n",
    "valid_y_torch = y_torch[train_samples:train_samples + test_samples]\n",
    "\n",
    "test_X_torch = X_torch[-test_samples:]\n",
    "test_y_torch = y_torch[-test_samples:]\n",
    "\n",
    "# pytorch train and test sets\n",
    "train_set = torch.utils.data.TensorDataset(train_X_torch, train_y_torch)\n",
    "valid_set = torch.utils.data.TensorDataset(valid_X_torch, valid_y_torch)\n",
    "test_set = torch.utils.data.TensorDataset(test_X_torch, test_y_torch)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Build LSTM model\n",
    "Build a LSTM model with `nn.LSTM` module.\n",
    "\n",
    "The architecture of the autoencoder used here is shown in the figure below.\n",
    "\n",
    "<img src=\"../assets/lstm.png\" alt=\"LSTM\" width=\"500\"/>\n",
    "\n",
    "(source of image: https://colah.github.io/posts/2015-08-Understanding-LSTMs/)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "class LSTM(nn.Module):\n",
    "    def __init__(self, input_dim, hidden_dim, output_dim=1,\n",
    "                 batch_size=1, num_layers=1, dropout=0.1):\n",
    "        \"\"\"\n",
    "        Initialize the LSTM model in Pytorch and specify the basic model structure.\n",
    "        Expected input timeseries dimension [batch_size, sequence, channels]\n",
    "        \"\"\"\n",
    "        super().__init__()\n",
    "        self.hidden_dim = hidden_dim\n",
    "        self.batch_size = batch_size\n",
    "        self.num_layers = num_layers\n",
    "        # Define the LSTM layer\n",
    "        self.lstm = nn.LSTM(input_size = input_dim, hidden_size = hidden_dim,\n",
    "                            num_layers = num_layers, batch_first = True, dropout = dropout)\n",
    "\n",
    "        # Define the output layer\n",
    "        self.linear = nn.Linear(hidden_dim, output_dim)\n",
    "        \n",
    "    def init_hidden(self):\n",
    "        \"\"\"Initialize hidden state with random values.\"\"\"\n",
    "        return (torch.randn(self.num_layers, self.batch_size, self.hidden_dim),\n",
    "                torch.randn(self.num_layers, self.batch_size, self.hidden_dim))\n",
    "        \n",
    "    def forward(self, input):\n",
    "        (h_0, c_0) = self.init_hidden()\n",
    "        x, _ = self.lstm(input, (h_0, c_0))\n",
    "        x = self.linear(x)\n",
    "            \n",
    "        return x"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Hyper-parameter tuning with W&B\n",
    "We use Weight&Biases to monitor the training process. It is very simple to integrate it into our workflow and more information about how to set it up can be found at https://docs.wandb.ai/quickstart. <br>"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Print system info."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pytorch version 2.1.2\n",
      "Is CUDA available? False\n",
      "Device to be used for computation: cpu\n"
     ]
    }
   ],
   "source": [
    "print (\"Pytorch version {}\".format(torch.__version__))\n",
    "use_cuda = torch.cuda.is_available()\n",
    "print(\"Is CUDA available? {}\".format(use_cuda))\n",
    "# use GPU if possible\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(\"Device to be used for computation: {}\".format(device))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Define hyperparameters, initialize config for wandb and syncronize training information with W&B server. If you have an API key and a login, you can uncomment the lines `wandb.login()`  and change `mode = 'disabled'` to `mode = 'online'`, to synchronise training. Otherwise, keep those lines commented.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# call weights & biases service\n",
    "# wandb.login()\n",
    "\n",
    "# define hyperparameters and the \n",
    "hyperparameters = dict(\n",
    "    epoch = 150,\n",
    "    input_dim = lat_precursor*lon_precursor,\n",
    "    hidden_dim = lat_precursor*lon_precursor*2,\n",
    "    output_dim = 1,\n",
    "    batch_size = 4, \n",
    "    num_layers = 2,\n",
    "    dropout = 0.0,\n",
    "    learning_rate = 0.02,\n",
    "    dataset = 'Weather',\n",
    "    architecture = 'LSTM'\n",
    ")\n",
    "\n",
    "# initialize weights & biases service\n",
    "# mode = 'online'\n",
    "mode = 'disabled'\n",
    "wandb.init(config=hyperparameters, project='test-LSTM', entity='ai4s2s', mode=mode)\n",
    "config = wandb.config"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create data loaders with chosen batch size. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create data loader and use batch \n",
    "train_loader = torch.utils.data.DataLoader(train_set, batch_size = config.batch_size, shuffle = False)\n",
    "valid_loader = torch.utils.data.DataLoader(valid_set, batch_size = config.batch_size, shuffle = False)\n",
    "test_loader = torch.utils.data.DataLoader(test_set, batch_size = config.batch_size, shuffle = False)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Initialize and train model\n",
    "Create model using specified hyperparameter. Initialize model and choose loss function and optimizer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model details:\n",
      " LSTM(\n",
      "  (lstm): LSTM(65, 130, num_layers=2, batch_first=True, dropout=0.1)\n",
      "  (linear): Linear(in_features=130, out_features=1, bias=True)\n",
      ")\n",
      "Optimizer details:\n",
      " Adam (\n",
      "Parameter Group 0\n",
      "    amsgrad: False\n",
      "    betas: (0.9, 0.999)\n",
      "    capturable: False\n",
      "    differentiable: False\n",
      "    eps: 1e-08\n",
      "    foreach: None\n",
      "    fused: None\n",
      "    lr: 0.02\n",
      "    maximize: False\n",
      "    weight_decay: 0\n",
      ")\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Initialize model\n",
    "model = LSTM(input_dim = config[\"input_dim\"],\n",
    "             hidden_dim = config[\"hidden_dim\"],\n",
    "             output_dim = config[\"output_dim\"], \n",
    "             batch_size = config[\"batch_size\"], \n",
    "             num_layers = config[\"num_layers\"]\n",
    ")\n",
    "# Specify loss function\n",
    "criterion = nn.MSELoss()\n",
    "# Choose optimizer\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=config.learning_rate)\n",
    "# Print model and optimizer details\n",
    "print('Model details:\\n', model)\n",
    "print('Optimizer details:\\n',optimizer)\n",
    "wandb.watch(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "238811\n"
     ]
    }
   ],
   "source": [
    "# display the total number of parameters\n",
    "utils.total_num_param(model)\n",
    "# for more details about the trainable parameter in each layer\n",
    "#utils.param_trainable(model)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Start the training and cross validation loop."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch : 0 [0/36(0%)]\tLoss: 508.584259\n",
      "Epoch : 0 [4/36(11%)]\tLoss: 489.579163\n",
      "Epoch : 0 [8/36(22%)]\tLoss: 484.424500\n",
      "Epoch : 0 [12/36(33%)]\tLoss: 396.097168\n",
      "Epoch : 0 [16/36(44%)]\tLoss: 299.434753\n",
      "Epoch : 0 [20/36(56%)]\tLoss: 171.694519\n",
      "Epoch : 0 [24/36(67%)]\tLoss: 97.885445\n",
      "Epoch : 0 [28/36(78%)]\tLoss: 49.746807\n",
      "Epoch : 0 [32/36(89%)]\tLoss: 21.667461\n",
      "Epoch : 1 [0/36(0%)]\tLoss: 4.278190\n",
      "Epoch : 1 [4/36(11%)]\tLoss: 0.842780\n",
      "Epoch : 1 [8/36(22%)]\tLoss: 3.558572\n",
      "Epoch : 1 [12/36(33%)]\tLoss: 14.213568\n",
      "Epoch : 1 [16/36(44%)]\tLoss: 18.502754\n",
      "Epoch : 1 [20/36(56%)]\tLoss: 22.304293\n",
      "Epoch : 1 [24/36(67%)]\tLoss: 27.259897\n",
      "Epoch : 1 [28/36(78%)]\tLoss: 98.888191\n",
      "Epoch : 1 [32/36(89%)]\tLoss: 23.050077\n",
      "Epoch : 2 [0/36(0%)]\tLoss: 23.876320\n",
      "Epoch : 2 [4/36(11%)]\tLoss: 17.615000\n",
      "Epoch : 2 [8/36(22%)]\tLoss: 7.877542\n",
      "Epoch : 2 [12/36(33%)]\tLoss: 5.795005\n",
      "Epoch : 2 [16/36(44%)]\tLoss: 1.190840\n",
      "Epoch : 2 [20/36(56%)]\tLoss: 1.911593\n",
      "Epoch : 2 [24/36(67%)]\tLoss: 2.109590\n",
      "Epoch : 2 [28/36(78%)]\tLoss: 2.889643\n",
      "Epoch : 2 [32/36(89%)]\tLoss: 7.070022\n",
      "Epoch : 3 [0/36(0%)]\tLoss: 4.037026\n",
      "Epoch : 3 [4/36(11%)]\tLoss: 5.566755\n",
      "Epoch : 3 [8/36(22%)]\tLoss: 7.413876\n",
      "Epoch : 3 [12/36(33%)]\tLoss: 4.555263\n",
      "Epoch : 3 [16/36(44%)]\tLoss: 6.288348\n",
      "Epoch : 3 [20/36(56%)]\tLoss: 6.952232\n",
      "Epoch : 3 [24/36(67%)]\tLoss: 3.299424\n",
      "Epoch : 3 [28/36(78%)]\tLoss: 2.170640\n",
      "Epoch : 3 [32/36(89%)]\tLoss: 27.657698\n",
      "Epoch : 4 [0/36(0%)]\tLoss: 1.177925\n",
      "Epoch : 4 [4/36(11%)]\tLoss: 3.040742\n",
      "Epoch : 4 [8/36(22%)]\tLoss: 2.325803\n",
      "Epoch : 4 [12/36(33%)]\tLoss: 4.379526\n",
      "Epoch : 4 [16/36(44%)]\tLoss: 3.289202\n",
      "Epoch : 4 [20/36(56%)]\tLoss: 2.637097\n",
      "Epoch : 4 [24/36(67%)]\tLoss: 3.250897\n",
      "Epoch : 4 [28/36(78%)]\tLoss: 3.662415\n",
      "Epoch : 4 [32/36(89%)]\tLoss: 2.874640\n",
      "Epoch : 5 [0/36(0%)]\tLoss: 2.667332\n",
      "Epoch : 5 [4/36(11%)]\tLoss: 2.149035\n",
      "Epoch : 5 [8/36(22%)]\tLoss: 0.297704\n",
      "Epoch : 5 [12/36(33%)]\tLoss: 0.363631\n",
      "Epoch : 5 [16/36(44%)]\tLoss: 1.337354\n",
      "Epoch : 5 [20/36(56%)]\tLoss: 3.352650\n",
      "Epoch : 5 [24/36(67%)]\tLoss: 2.523152\n",
      "Epoch : 5 [28/36(78%)]\tLoss: 2.578851\n",
      "Epoch : 5 [32/36(89%)]\tLoss: 2.733651\n",
      "Epoch : 6 [0/36(0%)]\tLoss: 0.437017\n",
      "Epoch : 6 [4/36(11%)]\tLoss: 0.763395\n",
      "Epoch : 6 [8/36(22%)]\tLoss: 0.775450\n",
      "Epoch : 6 [12/36(33%)]\tLoss: 0.376000\n",
      "Epoch : 6 [16/36(44%)]\tLoss: 0.948935\n",
      "Epoch : 6 [20/36(56%)]\tLoss: 2.390343\n",
      "Epoch : 6 [24/36(67%)]\tLoss: 1.146480\n",
      "Epoch : 6 [28/36(78%)]\tLoss: 1.848136\n",
      "Epoch : 6 [32/36(89%)]\tLoss: 2.219945\n",
      "Epoch : 7 [0/36(0%)]\tLoss: 1.158904\n",
      "Epoch : 7 [4/36(11%)]\tLoss: 2.085171\n",
      "Epoch : 7 [8/36(22%)]\tLoss: 0.731972\n",
      "Epoch : 7 [12/36(33%)]\tLoss: 0.729763\n",
      "Epoch : 7 [16/36(44%)]\tLoss: 0.586955\n",
      "Epoch : 7 [20/36(56%)]\tLoss: 2.005870\n",
      "Epoch : 7 [24/36(67%)]\tLoss: 1.103463\n",
      "Epoch : 7 [28/36(78%)]\tLoss: 1.231708\n",
      "Epoch : 7 [32/36(89%)]\tLoss: 1.629164\n",
      "Epoch : 8 [0/36(0%)]\tLoss: 0.437819\n",
      "Epoch : 8 [4/36(11%)]\tLoss: 0.836587\n",
      "Epoch : 8 [8/36(22%)]\tLoss: 0.408744\n",
      "Epoch : 8 [12/36(33%)]\tLoss: 0.285194\n",
      "Epoch : 8 [16/36(44%)]\tLoss: 1.012265\n",
      "Epoch : 8 [20/36(56%)]\tLoss: 2.258360\n",
      "Epoch : 8 [24/36(67%)]\tLoss: 1.213454\n",
      "Epoch : 8 [28/36(78%)]\tLoss: 1.099410\n",
      "Epoch : 8 [32/36(89%)]\tLoss: 1.800122\n",
      "Epoch : 9 [0/36(0%)]\tLoss: 0.502236\n",
      "Epoch : 9 [4/36(11%)]\tLoss: 1.257139\n",
      "Epoch : 9 [8/36(22%)]\tLoss: 0.280901\n",
      "Epoch : 9 [12/36(33%)]\tLoss: 0.614581\n",
      "Epoch : 9 [16/36(44%)]\tLoss: 0.836018\n",
      "Epoch : 9 [20/36(56%)]\tLoss: 1.800052\n",
      "Epoch : 9 [24/36(67%)]\tLoss: 1.123045\n",
      "Epoch : 9 [28/36(78%)]\tLoss: 1.503552\n",
      "Epoch : 9 [32/36(89%)]\tLoss: 1.571692\n",
      "Epoch : 10 [0/36(0%)]\tLoss: 0.638223\n",
      "Epoch : 10 [4/36(11%)]\tLoss: 1.202439\n",
      "Epoch : 10 [8/36(22%)]\tLoss: 0.351088\n",
      "Epoch : 10 [12/36(33%)]\tLoss: 0.272203\n",
      "Epoch : 10 [16/36(44%)]\tLoss: 0.809111\n",
      "Epoch : 10 [20/36(56%)]\tLoss: 2.077389\n",
      "Epoch : 10 [24/36(67%)]\tLoss: 1.366327\n",
      "Epoch : 10 [28/36(78%)]\tLoss: 1.298341\n",
      "Epoch : 10 [32/36(89%)]\tLoss: 1.423017\n",
      "Epoch : 11 [0/36(0%)]\tLoss: 0.412729\n",
      "Epoch : 11 [4/36(11%)]\tLoss: 1.066908\n",
      "Epoch : 11 [8/36(22%)]\tLoss: 0.314077\n",
      "Epoch : 11 [12/36(33%)]\tLoss: 0.544457\n",
      "Epoch : 11 [16/36(44%)]\tLoss: 0.762939\n",
      "Epoch : 11 [20/36(56%)]\tLoss: 1.853525\n",
      "Epoch : 11 [24/36(67%)]\tLoss: 1.093112\n",
      "Epoch : 11 [28/36(78%)]\tLoss: 0.972522\n",
      "Epoch : 11 [32/36(89%)]\tLoss: 1.370434\n",
      "Epoch : 12 [0/36(0%)]\tLoss: 0.579455\n",
      "Epoch : 12 [4/36(11%)]\tLoss: 1.175184\n",
      "Epoch : 12 [8/36(22%)]\tLoss: 0.260135\n",
      "Epoch : 12 [12/36(33%)]\tLoss: 0.459266\n",
      "Epoch : 12 [16/36(44%)]\tLoss: 2.220437\n",
      "Epoch : 12 [20/36(56%)]\tLoss: 1.667756\n",
      "Epoch : 12 [24/36(67%)]\tLoss: 0.944162\n",
      "Epoch : 12 [28/36(78%)]\tLoss: 1.087148\n",
      "Epoch : 12 [32/36(89%)]\tLoss: 1.285422\n",
      "Epoch : 13 [0/36(0%)]\tLoss: 0.700278\n",
      "Epoch : 13 [4/36(11%)]\tLoss: 1.417995\n",
      "Epoch : 13 [8/36(22%)]\tLoss: 0.382121\n",
      "Epoch : 13 [12/36(33%)]\tLoss: 0.622434\n",
      "Epoch : 13 [16/36(44%)]\tLoss: 0.992100\n",
      "Epoch : 13 [20/36(56%)]\tLoss: 2.076142\n",
      "Epoch : 13 [24/36(67%)]\tLoss: 1.157655\n",
      "Epoch : 13 [28/36(78%)]\tLoss: 1.450197\n",
      "Epoch : 13 [32/36(89%)]\tLoss: 1.486493\n",
      "Epoch : 14 [0/36(0%)]\tLoss: 0.383499\n",
      "Epoch : 14 [4/36(11%)]\tLoss: 1.214674\n",
      "Epoch : 14 [8/36(22%)]\tLoss: 0.401538\n",
      "Epoch : 14 [12/36(33%)]\tLoss: 0.517807\n",
      "Epoch : 14 [16/36(44%)]\tLoss: 0.746554\n",
      "Epoch : 14 [20/36(56%)]\tLoss: 1.917388\n",
      "Epoch : 14 [24/36(67%)]\tLoss: 1.132636\n",
      "Epoch : 14 [28/36(78%)]\tLoss: 1.559286\n",
      "Epoch : 14 [32/36(89%)]\tLoss: 1.217656\n",
      "Epoch : 15 [0/36(0%)]\tLoss: 0.505799\n",
      "Epoch : 15 [4/36(11%)]\tLoss: 1.102555\n",
      "Epoch : 15 [8/36(22%)]\tLoss: 0.286813\n",
      "Epoch : 15 [12/36(33%)]\tLoss: 0.435927\n",
      "Epoch : 15 [16/36(44%)]\tLoss: 0.656406\n",
      "Epoch : 15 [20/36(56%)]\tLoss: 1.893523\n",
      "Epoch : 15 [24/36(67%)]\tLoss: 1.034207\n",
      "Epoch : 15 [28/36(78%)]\tLoss: 1.606460\n",
      "Epoch : 15 [32/36(89%)]\tLoss: 1.204749\n",
      "Epoch : 16 [0/36(0%)]\tLoss: 0.591076\n",
      "Epoch : 16 [4/36(11%)]\tLoss: 1.097042\n",
      "Epoch : 16 [8/36(22%)]\tLoss: 0.365289\n",
      "Epoch : 16 [12/36(33%)]\tLoss: 0.621078\n",
      "Epoch : 16 [16/36(44%)]\tLoss: 0.773648\n",
      "Epoch : 16 [20/36(56%)]\tLoss: 2.002934\n",
      "Epoch : 16 [24/36(67%)]\tLoss: 0.985409\n",
      "Epoch : 16 [28/36(78%)]\tLoss: 1.361294\n",
      "Epoch : 16 [32/36(89%)]\tLoss: 1.388230\n",
      "Epoch : 17 [0/36(0%)]\tLoss: 0.608375\n",
      "Epoch : 17 [4/36(11%)]\tLoss: 1.079986\n",
      "Epoch : 17 [8/36(22%)]\tLoss: 0.228396\n",
      "Epoch : 17 [12/36(33%)]\tLoss: 0.541713\n",
      "Epoch : 17 [16/36(44%)]\tLoss: 15.403788\n",
      "Epoch : 17 [20/36(56%)]\tLoss: 1.549975\n",
      "Epoch : 17 [24/36(67%)]\tLoss: 0.990663\n",
      "Epoch : 17 [28/36(78%)]\tLoss: 1.399802\n",
      "Epoch : 17 [32/36(89%)]\tLoss: 1.247961\n",
      "Epoch : 18 [0/36(0%)]\tLoss: 1.129281\n",
      "Epoch : 18 [4/36(11%)]\tLoss: 1.744506\n",
      "Epoch : 18 [8/36(22%)]\tLoss: 0.430318\n",
      "Epoch : 18 [12/36(33%)]\tLoss: 0.567855\n",
      "Epoch : 18 [16/36(44%)]\tLoss: 0.732278\n",
      "Epoch : 18 [20/36(56%)]\tLoss: 2.068665\n",
      "Epoch : 18 [24/36(67%)]\tLoss: 1.507936\n",
      "Epoch : 18 [28/36(78%)]\tLoss: 1.225059\n",
      "Epoch : 18 [32/36(89%)]\tLoss: 1.590272\n",
      "Epoch : 19 [0/36(0%)]\tLoss: 0.250360\n",
      "Epoch : 19 [4/36(11%)]\tLoss: 0.949434\n",
      "Epoch : 19 [8/36(22%)]\tLoss: 0.364347\n",
      "Epoch : 19 [12/36(33%)]\tLoss: 0.885778\n",
      "Epoch : 19 [16/36(44%)]\tLoss: 0.766354\n",
      "Epoch : 19 [20/36(56%)]\tLoss: 1.680036\n",
      "Epoch : 19 [24/36(67%)]\tLoss: 0.999250\n",
      "Epoch : 19 [28/36(78%)]\tLoss: 0.956252\n",
      "Epoch : 19 [32/36(89%)]\tLoss: 1.265652\n",
      "Epoch : 20 [0/36(0%)]\tLoss: 0.537306\n",
      "Epoch : 20 [4/36(11%)]\tLoss: 1.385282\n",
      "Epoch : 20 [8/36(22%)]\tLoss: 0.354683\n",
      "Epoch : 20 [12/36(33%)]\tLoss: 0.467504\n",
      "Epoch : 20 [16/36(44%)]\tLoss: 0.823383\n",
      "Epoch : 20 [20/36(56%)]\tLoss: 2.091824\n",
      "Epoch : 20 [24/36(67%)]\tLoss: 1.060751\n",
      "Epoch : 20 [28/36(78%)]\tLoss: 0.959582\n",
      "Epoch : 20 [32/36(89%)]\tLoss: 1.754925\n",
      "Epoch : 21 [0/36(0%)]\tLoss: 0.354787\n",
      "Epoch : 21 [4/36(11%)]\tLoss: 1.486068\n",
      "Epoch : 21 [8/36(22%)]\tLoss: 0.413150\n",
      "Epoch : 21 [12/36(33%)]\tLoss: 0.696335\n",
      "Epoch : 21 [16/36(44%)]\tLoss: 0.816269\n",
      "Epoch : 21 [20/36(56%)]\tLoss: 1.928941\n",
      "Epoch : 21 [24/36(67%)]\tLoss: 1.065933\n",
      "Epoch : 21 [28/36(78%)]\tLoss: 1.492135\n",
      "Epoch : 21 [32/36(89%)]\tLoss: 1.272583\n",
      "Epoch : 22 [0/36(0%)]\tLoss: 0.449633\n",
      "Epoch : 22 [4/36(11%)]\tLoss: 1.111781\n",
      "Epoch : 22 [8/36(22%)]\tLoss: 0.362148\n",
      "Epoch : 22 [12/36(33%)]\tLoss: 0.365892\n",
      "Epoch : 22 [16/36(44%)]\tLoss: 0.685664\n",
      "Epoch : 22 [20/36(56%)]\tLoss: 1.761396\n",
      "Epoch : 22 [24/36(67%)]\tLoss: 0.993014\n",
      "Epoch : 22 [28/36(78%)]\tLoss: 0.656901\n",
      "Epoch : 22 [32/36(89%)]\tLoss: 1.486815\n",
      "Epoch : 23 [0/36(0%)]\tLoss: 0.614337\n",
      "Epoch : 23 [4/36(11%)]\tLoss: 1.176307\n",
      "Epoch : 23 [8/36(22%)]\tLoss: 0.287041\n",
      "Epoch : 23 [12/36(33%)]\tLoss: 0.348178\n",
      "Epoch : 23 [16/36(44%)]\tLoss: 0.640980\n",
      "Epoch : 23 [20/36(56%)]\tLoss: 1.814524\n",
      "Epoch : 23 [24/36(67%)]\tLoss: 0.785374\n",
      "Epoch : 23 [28/36(78%)]\tLoss: 1.290595\n",
      "Epoch : 23 [32/36(89%)]\tLoss: 1.378169\n",
      "Epoch : 24 [0/36(0%)]\tLoss: 0.492574\n",
      "Epoch : 24 [4/36(11%)]\tLoss: 0.988177\n",
      "Epoch : 24 [8/36(22%)]\tLoss: 0.438381\n",
      "Epoch : 24 [12/36(33%)]\tLoss: 0.808166\n",
      "Epoch : 24 [16/36(44%)]\tLoss: 0.755162\n",
      "Epoch : 24 [20/36(56%)]\tLoss: 1.714777\n",
      "Epoch : 24 [24/36(67%)]\tLoss: 0.813457\n",
      "Epoch : 24 [28/36(78%)]\tLoss: 0.886818\n",
      "Epoch : 24 [32/36(89%)]\tLoss: 1.164585\n",
      "Epoch : 25 [0/36(0%)]\tLoss: 0.597715\n",
      "Epoch : 25 [4/36(11%)]\tLoss: 1.452986\n",
      "Epoch : 25 [8/36(22%)]\tLoss: 0.432988\n",
      "Epoch : 25 [12/36(33%)]\tLoss: 0.549594\n",
      "Epoch : 25 [16/36(44%)]\tLoss: 0.805746\n",
      "Epoch : 25 [20/36(56%)]\tLoss: 1.575088\n",
      "Epoch : 25 [24/36(67%)]\tLoss: 1.070147\n",
      "Epoch : 25 [28/36(78%)]\tLoss: 0.931112\n",
      "Epoch : 25 [32/36(89%)]\tLoss: 1.014211\n",
      "Epoch : 26 [0/36(0%)]\tLoss: 0.538299\n",
      "Epoch : 26 [4/36(11%)]\tLoss: 1.431323\n",
      "Epoch : 26 [8/36(22%)]\tLoss: 0.458226\n",
      "Epoch : 26 [12/36(33%)]\tLoss: 0.588073\n",
      "Epoch : 26 [16/36(44%)]\tLoss: 0.776289\n",
      "Epoch : 26 [20/36(56%)]\tLoss: 1.632388\n",
      "Epoch : 26 [24/36(67%)]\tLoss: 1.299865\n",
      "Epoch : 26 [28/36(78%)]\tLoss: 1.235039\n",
      "Epoch : 26 [32/36(89%)]\tLoss: 2.361759\n",
      "Epoch : 27 [0/36(0%)]\tLoss: 0.503370\n",
      "Epoch : 27 [4/36(11%)]\tLoss: 0.787307\n",
      "Epoch : 27 [8/36(22%)]\tLoss: 0.390411\n",
      "Epoch : 27 [12/36(33%)]\tLoss: 0.518691\n",
      "Epoch : 27 [16/36(44%)]\tLoss: 0.507706\n",
      "Epoch : 27 [20/36(56%)]\tLoss: 1.486015\n",
      "Epoch : 27 [24/36(67%)]\tLoss: 1.007744\n",
      "Epoch : 27 [28/36(78%)]\tLoss: 1.124909\n",
      "Epoch : 27 [32/36(89%)]\tLoss: 1.310836\n",
      "Epoch : 28 [0/36(0%)]\tLoss: 0.724389\n",
      "Epoch : 28 [4/36(11%)]\tLoss: 1.445935\n",
      "Epoch : 28 [8/36(22%)]\tLoss: 0.198445\n",
      "Epoch : 28 [12/36(33%)]\tLoss: 0.280480\n",
      "Epoch : 28 [16/36(44%)]\tLoss: 0.857233\n",
      "Epoch : 28 [20/36(56%)]\tLoss: 2.163746\n",
      "Epoch : 28 [24/36(67%)]\tLoss: 1.104517\n",
      "Epoch : 28 [28/36(78%)]\tLoss: 1.154333\n",
      "Epoch : 28 [32/36(89%)]\tLoss: 0.951360\n",
      "Epoch : 29 [0/36(0%)]\tLoss: 0.836381\n",
      "Epoch : 29 [4/36(11%)]\tLoss: 1.432677\n",
      "Epoch : 29 [8/36(22%)]\tLoss: 0.505042\n",
      "Epoch : 29 [12/36(33%)]\tLoss: 0.887347\n",
      "Epoch : 29 [16/36(44%)]\tLoss: 0.716298\n",
      "Epoch : 29 [20/36(56%)]\tLoss: 1.673971\n",
      "Epoch : 29 [24/36(67%)]\tLoss: 1.170848\n",
      "Epoch : 29 [28/36(78%)]\tLoss: 1.306377\n",
      "Epoch : 29 [32/36(89%)]\tLoss: 1.645008\n",
      "Epoch : 30 [0/36(0%)]\tLoss: 0.366824\n",
      "Epoch : 30 [4/36(11%)]\tLoss: 0.499150\n",
      "Epoch : 30 [8/36(22%)]\tLoss: 0.265904\n",
      "Epoch : 30 [12/36(33%)]\tLoss: 0.465497\n",
      "Epoch : 30 [16/36(44%)]\tLoss: 0.571813\n",
      "Epoch : 30 [20/36(56%)]\tLoss: 1.589559\n",
      "Epoch : 30 [24/36(67%)]\tLoss: 0.993832\n",
      "Epoch : 30 [28/36(78%)]\tLoss: 1.207466\n",
      "Epoch : 30 [32/36(89%)]\tLoss: 1.210111\n",
      "Epoch : 31 [0/36(0%)]\tLoss: 0.355208\n",
      "Epoch : 31 [4/36(11%)]\tLoss: 0.793420\n",
      "Epoch : 31 [8/36(22%)]\tLoss: 0.184402\n",
      "Epoch : 31 [12/36(33%)]\tLoss: 0.417205\n",
      "Epoch : 31 [16/36(44%)]\tLoss: 0.868806\n",
      "Epoch : 31 [20/36(56%)]\tLoss: 1.704289\n",
      "Epoch : 31 [24/36(67%)]\tLoss: 1.019739\n",
      "Epoch : 31 [28/36(78%)]\tLoss: 0.753782\n",
      "Epoch : 31 [32/36(89%)]\tLoss: 1.028763\n",
      "Epoch : 32 [0/36(0%)]\tLoss: 0.438362\n",
      "Epoch : 32 [4/36(11%)]\tLoss: 1.387437\n",
      "Epoch : 32 [8/36(22%)]\tLoss: 0.301998\n",
      "Epoch : 32 [12/36(33%)]\tLoss: 0.527436\n",
      "Epoch : 32 [16/36(44%)]\tLoss: 0.419992\n",
      "Epoch : 32 [20/36(56%)]\tLoss: 1.410710\n",
      "Epoch : 32 [24/36(67%)]\tLoss: 0.979486\n",
      "Epoch : 32 [28/36(78%)]\tLoss: 1.023432\n",
      "Epoch : 32 [32/36(89%)]\tLoss: 1.318741\n",
      "Epoch : 33 [0/36(0%)]\tLoss: 0.408260\n",
      "Epoch : 33 [4/36(11%)]\tLoss: 1.255412\n",
      "Epoch : 33 [8/36(22%)]\tLoss: 0.396476\n",
      "Epoch : 33 [12/36(33%)]\tLoss: 0.424917\n",
      "Epoch : 33 [16/36(44%)]\tLoss: 0.531891\n",
      "Epoch : 33 [20/36(56%)]\tLoss: 1.569926\n",
      "Epoch : 33 [24/36(67%)]\tLoss: 0.930528\n",
      "Epoch : 33 [28/36(78%)]\tLoss: 1.037493\n",
      "Epoch : 33 [32/36(89%)]\tLoss: 1.100780\n",
      "Epoch : 34 [0/36(0%)]\tLoss: 0.354256\n",
      "Epoch : 34 [4/36(11%)]\tLoss: 1.314515\n",
      "Epoch : 34 [8/36(22%)]\tLoss: 0.206112\n",
      "Epoch : 34 [12/36(33%)]\tLoss: 0.593593\n",
      "Epoch : 34 [16/36(44%)]\tLoss: 0.903699\n",
      "Epoch : 34 [20/36(56%)]\tLoss: 1.324670\n",
      "Epoch : 34 [24/36(67%)]\tLoss: 0.778266\n",
      "Epoch : 34 [28/36(78%)]\tLoss: 0.705398\n",
      "Epoch : 34 [32/36(89%)]\tLoss: 1.020614\n",
      "Epoch : 35 [0/36(0%)]\tLoss: 0.333391\n",
      "Epoch : 35 [4/36(11%)]\tLoss: 1.395265\n",
      "Epoch : 35 [8/36(22%)]\tLoss: 0.131507\n",
      "Epoch : 35 [12/36(33%)]\tLoss: 0.537471\n",
      "Epoch : 35 [16/36(44%)]\tLoss: 0.652337\n",
      "Epoch : 35 [20/36(56%)]\tLoss: 1.225708\n",
      "Epoch : 35 [24/36(67%)]\tLoss: 0.661133\n",
      "Epoch : 35 [28/36(78%)]\tLoss: 1.139045\n",
      "Epoch : 35 [32/36(89%)]\tLoss: 1.339713\n",
      "Epoch : 36 [0/36(0%)]\tLoss: 0.315643\n",
      "Epoch : 36 [4/36(11%)]\tLoss: 1.441062\n",
      "Epoch : 36 [8/36(22%)]\tLoss: 0.270850\n",
      "Epoch : 36 [12/36(33%)]\tLoss: 0.419435\n",
      "Epoch : 36 [16/36(44%)]\tLoss: 0.537357\n",
      "Epoch : 36 [20/36(56%)]\tLoss: 1.046800\n",
      "Epoch : 36 [24/36(67%)]\tLoss: 0.560395\n",
      "Epoch : 36 [28/36(78%)]\tLoss: 1.159633\n",
      "Epoch : 36 [32/36(89%)]\tLoss: 1.465568\n",
      "Epoch : 37 [0/36(0%)]\tLoss: 0.326660\n",
      "Epoch : 37 [4/36(11%)]\tLoss: 0.658771\n",
      "Epoch : 37 [8/36(22%)]\tLoss: 0.309746\n",
      "Epoch : 37 [12/36(33%)]\tLoss: 0.543800\n",
      "Epoch : 37 [16/36(44%)]\tLoss: 0.450595\n",
      "Epoch : 37 [20/36(56%)]\tLoss: 1.316446\n",
      "Epoch : 37 [24/36(67%)]\tLoss: 0.744958\n",
      "Epoch : 37 [28/36(78%)]\tLoss: 0.759218\n",
      "Epoch : 37 [32/36(89%)]\tLoss: 0.904929\n",
      "Epoch : 38 [0/36(0%)]\tLoss: 0.358062\n",
      "Epoch : 38 [4/36(11%)]\tLoss: 1.498037\n",
      "Epoch : 38 [8/36(22%)]\tLoss: 0.181726\n",
      "Epoch : 38 [12/36(33%)]\tLoss: 0.220401\n",
      "Epoch : 38 [16/36(44%)]\tLoss: 0.924105\n",
      "Epoch : 38 [20/36(56%)]\tLoss: 2.004258\n",
      "Epoch : 38 [24/36(67%)]\tLoss: 0.578513\n",
      "Epoch : 38 [28/36(78%)]\tLoss: 1.056301\n",
      "Epoch : 38 [32/36(89%)]\tLoss: 1.061478\n",
      "Epoch : 39 [0/36(0%)]\tLoss: 0.480924\n",
      "Epoch : 39 [4/36(11%)]\tLoss: 1.865398\n",
      "Epoch : 39 [8/36(22%)]\tLoss: 0.321774\n",
      "Epoch : 39 [12/36(33%)]\tLoss: 0.480000\n",
      "Epoch : 39 [16/36(44%)]\tLoss: 1.622866\n",
      "Epoch : 39 [20/36(56%)]\tLoss: 1.484055\n",
      "Epoch : 39 [24/36(67%)]\tLoss: 0.643541\n",
      "Epoch : 39 [28/36(78%)]\tLoss: 1.022989\n",
      "Epoch : 39 [32/36(89%)]\tLoss: 0.573010\n",
      "Epoch : 40 [0/36(0%)]\tLoss: 0.240988\n",
      "Epoch : 40 [4/36(11%)]\tLoss: 0.934603\n",
      "Epoch : 40 [8/36(22%)]\tLoss: 0.127332\n",
      "Epoch : 40 [12/36(33%)]\tLoss: 0.511201\n",
      "Epoch : 40 [16/36(44%)]\tLoss: 0.641673\n",
      "Epoch : 40 [20/36(56%)]\tLoss: 1.189185\n",
      "Epoch : 40 [24/36(67%)]\tLoss: 0.762764\n",
      "Epoch : 40 [28/36(78%)]\tLoss: 0.714797\n",
      "Epoch : 40 [32/36(89%)]\tLoss: 0.745611\n",
      "Epoch : 41 [0/36(0%)]\tLoss: 0.261947\n",
      "Epoch : 41 [4/36(11%)]\tLoss: 1.161531\n",
      "Epoch : 41 [8/36(22%)]\tLoss: 0.156456\n",
      "Epoch : 41 [12/36(33%)]\tLoss: 0.469099\n",
      "Epoch : 41 [16/36(44%)]\tLoss: 0.624814\n",
      "Epoch : 41 [20/36(56%)]\tLoss: 1.295788\n",
      "Epoch : 41 [24/36(67%)]\tLoss: 0.682808\n",
      "Epoch : 41 [28/36(78%)]\tLoss: 0.859398\n",
      "Epoch : 41 [32/36(89%)]\tLoss: 0.640153\n",
      "Epoch : 42 [0/36(0%)]\tLoss: 0.359450\n",
      "Epoch : 42 [4/36(11%)]\tLoss: 1.317981\n",
      "Epoch : 42 [8/36(22%)]\tLoss: 0.189891\n",
      "Epoch : 42 [12/36(33%)]\tLoss: 0.541890\n",
      "Epoch : 42 [16/36(44%)]\tLoss: 0.660854\n",
      "Epoch : 42 [20/36(56%)]\tLoss: 1.630389\n",
      "Epoch : 42 [24/36(67%)]\tLoss: 0.479220\n",
      "Epoch : 42 [28/36(78%)]\tLoss: 1.119197\n",
      "Epoch : 42 [32/36(89%)]\tLoss: 0.700291\n",
      "Epoch : 43 [0/36(0%)]\tLoss: 0.330202\n",
      "Epoch : 43 [4/36(11%)]\tLoss: 0.963364\n",
      "Epoch : 43 [8/36(22%)]\tLoss: 0.074339\n",
      "Epoch : 43 [12/36(33%)]\tLoss: 0.321596\n",
      "Epoch : 43 [16/36(44%)]\tLoss: 0.702669\n",
      "Epoch : 43 [20/36(56%)]\tLoss: 1.461693\n",
      "Epoch : 43 [24/36(67%)]\tLoss: 0.717811\n",
      "Epoch : 43 [28/36(78%)]\tLoss: 0.648568\n",
      "Epoch : 43 [32/36(89%)]\tLoss: 0.692827\n",
      "Epoch : 44 [0/36(0%)]\tLoss: 0.390700\n",
      "Epoch : 44 [4/36(11%)]\tLoss: 1.149426\n",
      "Epoch : 44 [8/36(22%)]\tLoss: 0.168986\n",
      "Epoch : 44 [12/36(33%)]\tLoss: 0.562529\n",
      "Epoch : 44 [16/36(44%)]\tLoss: 0.609913\n",
      "Epoch : 44 [20/36(56%)]\tLoss: 1.734709\n",
      "Epoch : 44 [24/36(67%)]\tLoss: 0.340055\n",
      "Epoch : 44 [28/36(78%)]\tLoss: 1.080092\n",
      "Epoch : 44 [32/36(89%)]\tLoss: 0.697234\n",
      "Epoch : 45 [0/36(0%)]\tLoss: 0.285021\n",
      "Epoch : 45 [4/36(11%)]\tLoss: 1.133306\n",
      "Epoch : 45 [8/36(22%)]\tLoss: 0.345929\n",
      "Epoch : 45 [12/36(33%)]\tLoss: 0.247243\n",
      "Epoch : 45 [16/36(44%)]\tLoss: 0.622414\n",
      "Epoch : 45 [20/36(56%)]\tLoss: 0.957461\n",
      "Epoch : 45 [24/36(67%)]\tLoss: 0.386344\n",
      "Epoch : 45 [28/36(78%)]\tLoss: 0.921096\n",
      "Epoch : 45 [32/36(89%)]\tLoss: 0.515546\n",
      "Epoch : 46 [0/36(0%)]\tLoss: 0.337347\n",
      "Epoch : 46 [4/36(11%)]\tLoss: 0.819388\n",
      "Epoch : 46 [8/36(22%)]\tLoss: 0.118760\n",
      "Epoch : 46 [12/36(33%)]\tLoss: 0.403043\n",
      "Epoch : 46 [16/36(44%)]\tLoss: 0.589820\n",
      "Epoch : 46 [20/36(56%)]\tLoss: 1.215221\n",
      "Epoch : 46 [24/36(67%)]\tLoss: 0.420268\n",
      "Epoch : 46 [28/36(78%)]\tLoss: 0.631678\n",
      "Epoch : 46 [32/36(89%)]\tLoss: 0.865619\n",
      "Epoch : 47 [0/36(0%)]\tLoss: 0.277459\n",
      "Epoch : 47 [4/36(11%)]\tLoss: 1.072624\n",
      "Epoch : 47 [8/36(22%)]\tLoss: 0.069972\n",
      "Epoch : 47 [12/36(33%)]\tLoss: 0.403553\n",
      "Epoch : 47 [16/36(44%)]\tLoss: 0.539215\n",
      "Epoch : 47 [20/36(56%)]\tLoss: 1.585401\n",
      "Epoch : 47 [24/36(67%)]\tLoss: 0.484923\n",
      "Epoch : 47 [28/36(78%)]\tLoss: 1.048617\n",
      "Epoch : 47 [32/36(89%)]\tLoss: 0.472647\n",
      "Epoch : 48 [0/36(0%)]\tLoss: 0.249743\n",
      "Epoch : 48 [4/36(11%)]\tLoss: 1.233549\n",
      "Epoch : 48 [8/36(22%)]\tLoss: 0.133145\n",
      "Epoch : 48 [12/36(33%)]\tLoss: 0.487102\n",
      "Epoch : 48 [16/36(44%)]\tLoss: 0.772336\n",
      "Epoch : 48 [20/36(56%)]\tLoss: 1.075320\n",
      "Epoch : 48 [24/36(67%)]\tLoss: 0.419837\n",
      "Epoch : 48 [28/36(78%)]\tLoss: 1.099843\n",
      "Epoch : 48 [32/36(89%)]\tLoss: 0.690398\n",
      "Epoch : 49 [0/36(0%)]\tLoss: 0.529672\n",
      "Epoch : 49 [4/36(11%)]\tLoss: 1.377880\n",
      "Epoch : 49 [8/36(22%)]\tLoss: 0.338583\n",
      "Epoch : 49 [12/36(33%)]\tLoss: 0.457722\n",
      "Epoch : 49 [16/36(44%)]\tLoss: 0.794459\n",
      "Epoch : 49 [20/36(56%)]\tLoss: 0.758587\n",
      "Epoch : 49 [24/36(67%)]\tLoss: 0.364898\n",
      "Epoch : 49 [28/36(78%)]\tLoss: 1.617512\n",
      "Epoch : 49 [32/36(89%)]\tLoss: 0.905486\n",
      "Epoch : 50 [0/36(0%)]\tLoss: 0.194277\n",
      "Epoch : 50 [4/36(11%)]\tLoss: 1.276311\n",
      "Epoch : 50 [8/36(22%)]\tLoss: 0.395778\n",
      "Epoch : 50 [12/36(33%)]\tLoss: 0.800589\n",
      "Epoch : 50 [16/36(44%)]\tLoss: 0.193169\n",
      "Epoch : 50 [20/36(56%)]\tLoss: 0.342214\n",
      "Epoch : 50 [24/36(67%)]\tLoss: 0.348514\n",
      "Epoch : 50 [28/36(78%)]\tLoss: 1.669420\n",
      "Epoch : 50 [32/36(89%)]\tLoss: 0.981101\n",
      "Epoch : 51 [0/36(0%)]\tLoss: 0.436034\n",
      "Epoch : 51 [4/36(11%)]\tLoss: 0.919327\n",
      "Epoch : 51 [8/36(22%)]\tLoss: 0.141251\n",
      "Epoch : 51 [12/36(33%)]\tLoss: 0.334751\n",
      "Epoch : 51 [16/36(44%)]\tLoss: 0.380790\n",
      "Epoch : 51 [20/36(56%)]\tLoss: 0.365853\n",
      "Epoch : 51 [24/36(67%)]\tLoss: 0.508698\n",
      "Epoch : 51 [28/36(78%)]\tLoss: 0.782394\n",
      "Epoch : 51 [32/36(89%)]\tLoss: 0.907044\n",
      "Epoch : 52 [0/36(0%)]\tLoss: 0.341293\n",
      "Epoch : 52 [4/36(11%)]\tLoss: 0.938866\n",
      "Epoch : 52 [8/36(22%)]\tLoss: 0.081437\n",
      "Epoch : 52 [12/36(33%)]\tLoss: 0.391152\n",
      "Epoch : 52 [16/36(44%)]\tLoss: 0.901419\n",
      "Epoch : 52 [20/36(56%)]\tLoss: 0.445237\n",
      "Epoch : 52 [24/36(67%)]\tLoss: 0.443016\n",
      "Epoch : 52 [28/36(78%)]\tLoss: 1.168984\n",
      "Epoch : 52 [32/36(89%)]\tLoss: 0.332686\n",
      "Epoch : 53 [0/36(0%)]\tLoss: 0.411747\n",
      "Epoch : 53 [4/36(11%)]\tLoss: 0.751605\n",
      "Epoch : 53 [8/36(22%)]\tLoss: 0.057952\n",
      "Epoch : 53 [12/36(33%)]\tLoss: 0.340813\n",
      "Epoch : 53 [16/36(44%)]\tLoss: 0.237320\n",
      "Epoch : 53 [20/36(56%)]\tLoss: 0.387541\n",
      "Epoch : 53 [24/36(67%)]\tLoss: 0.378422\n",
      "Epoch : 53 [28/36(78%)]\tLoss: 0.954479\n",
      "Epoch : 53 [32/36(89%)]\tLoss: 0.344902\n",
      "Epoch : 54 [0/36(0%)]\tLoss: 0.283856\n",
      "Epoch : 54 [4/36(11%)]\tLoss: 1.144521\n",
      "Epoch : 54 [8/36(22%)]\tLoss: 0.094913\n",
      "Epoch : 54 [12/36(33%)]\tLoss: 0.408707\n",
      "Epoch : 54 [16/36(44%)]\tLoss: 0.732193\n",
      "Epoch : 54 [20/36(56%)]\tLoss: 0.565221\n",
      "Epoch : 54 [24/36(67%)]\tLoss: 0.268218\n",
      "Epoch : 54 [28/36(78%)]\tLoss: 0.806647\n",
      "Epoch : 54 [32/36(89%)]\tLoss: 0.272619\n",
      "Epoch : 55 [0/36(0%)]\tLoss: 0.236451\n",
      "Epoch : 55 [4/36(11%)]\tLoss: 0.978926\n",
      "Epoch : 55 [8/36(22%)]\tLoss: 0.205416\n",
      "Epoch : 55 [12/36(33%)]\tLoss: 0.282229\n",
      "Epoch : 55 [16/36(44%)]\tLoss: 0.585025\n",
      "Epoch : 55 [20/36(56%)]\tLoss: 0.961256\n",
      "Epoch : 55 [24/36(67%)]\tLoss: 0.373155\n",
      "Epoch : 55 [28/36(78%)]\tLoss: 0.511703\n",
      "Epoch : 55 [32/36(89%)]\tLoss: 0.986082\n",
      "Epoch : 56 [0/36(0%)]\tLoss: 0.304324\n",
      "Epoch : 56 [4/36(11%)]\tLoss: 0.540398\n",
      "Epoch : 56 [8/36(22%)]\tLoss: 0.031518\n",
      "Epoch : 56 [12/36(33%)]\tLoss: 0.778907\n",
      "Epoch : 56 [16/36(44%)]\tLoss: 0.936378\n",
      "Epoch : 56 [20/36(56%)]\tLoss: 0.409590\n",
      "Epoch : 56 [24/36(67%)]\tLoss: 0.367014\n",
      "Epoch : 56 [28/36(78%)]\tLoss: 0.341347\n",
      "Epoch : 56 [32/36(89%)]\tLoss: 1.171724\n",
      "Epoch : 57 [0/36(0%)]\tLoss: 0.457374\n",
      "Epoch : 57 [4/36(11%)]\tLoss: 0.379317\n",
      "Epoch : 57 [8/36(22%)]\tLoss: 0.079970\n",
      "Epoch : 57 [12/36(33%)]\tLoss: 0.362772\n",
      "Epoch : 57 [16/36(44%)]\tLoss: 0.569463\n",
      "Epoch : 57 [20/36(56%)]\tLoss: 0.283171\n",
      "Epoch : 57 [24/36(67%)]\tLoss: 0.517740\n",
      "Epoch : 57 [28/36(78%)]\tLoss: 0.577444\n",
      "Epoch : 57 [32/36(89%)]\tLoss: 0.555228\n",
      "Epoch : 58 [0/36(0%)]\tLoss: 0.309039\n",
      "Epoch : 58 [4/36(11%)]\tLoss: 0.256726\n",
      "Epoch : 58 [8/36(22%)]\tLoss: 0.103449\n",
      "Epoch : 58 [12/36(33%)]\tLoss: 0.075231\n",
      "Epoch : 58 [16/36(44%)]\tLoss: 0.548213\n",
      "Epoch : 58 [20/36(56%)]\tLoss: 0.724548\n",
      "Epoch : 58 [24/36(67%)]\tLoss: 0.187811\n",
      "Epoch : 58 [28/36(78%)]\tLoss: 0.500180\n",
      "Epoch : 58 [32/36(89%)]\tLoss: 0.326950\n",
      "Epoch : 59 [0/36(0%)]\tLoss: 0.405652\n",
      "Epoch : 59 [4/36(11%)]\tLoss: 0.666810\n",
      "Epoch : 59 [8/36(22%)]\tLoss: 0.180725\n",
      "Epoch : 59 [12/36(33%)]\tLoss: 0.016212\n",
      "Epoch : 59 [16/36(44%)]\tLoss: 0.499729\n",
      "Epoch : 59 [20/36(56%)]\tLoss: 0.505925\n",
      "Epoch : 59 [24/36(67%)]\tLoss: 0.140059\n",
      "Epoch : 59 [28/36(78%)]\tLoss: 0.546197\n",
      "Epoch : 59 [32/36(89%)]\tLoss: 0.425007\n",
      "Epoch : 60 [0/36(0%)]\tLoss: 0.219418\n",
      "Epoch : 60 [4/36(11%)]\tLoss: 0.421457\n",
      "Epoch : 60 [8/36(22%)]\tLoss: 0.598579\n",
      "Epoch : 60 [12/36(33%)]\tLoss: 0.514790\n",
      "Epoch : 60 [16/36(44%)]\tLoss: 0.605864\n",
      "Epoch : 60 [20/36(56%)]\tLoss: 0.251507\n",
      "Epoch : 60 [24/36(67%)]\tLoss: 0.352219\n",
      "Epoch : 60 [28/36(78%)]\tLoss: 0.506136\n",
      "Epoch : 60 [32/36(89%)]\tLoss: 1.117898\n",
      "Epoch : 61 [0/36(0%)]\tLoss: 0.513959\n",
      "Epoch : 61 [4/36(11%)]\tLoss: 0.529024\n",
      "Epoch : 61 [8/36(22%)]\tLoss: 0.437208\n",
      "Epoch : 61 [12/36(33%)]\tLoss: 0.276267\n",
      "Epoch : 61 [16/36(44%)]\tLoss: 0.940994\n",
      "Epoch : 61 [20/36(56%)]\tLoss: 0.144337\n",
      "Epoch : 61 [24/36(67%)]\tLoss: 0.451732\n",
      "Epoch : 61 [28/36(78%)]\tLoss: 0.633429\n",
      "Epoch : 61 [32/36(89%)]\tLoss: 0.248102\n",
      "Epoch : 62 [0/36(0%)]\tLoss: 0.298392\n",
      "Epoch : 62 [4/36(11%)]\tLoss: 0.136540\n",
      "Epoch : 62 [8/36(22%)]\tLoss: 0.349791\n",
      "Epoch : 62 [12/36(33%)]\tLoss: 0.453364\n",
      "Epoch : 62 [16/36(44%)]\tLoss: 0.753284\n",
      "Epoch : 62 [20/36(56%)]\tLoss: 0.859750\n",
      "Epoch : 62 [24/36(67%)]\tLoss: 0.717212\n",
      "Epoch : 62 [28/36(78%)]\tLoss: 1.509548\n",
      "Epoch : 62 [32/36(89%)]\tLoss: 1.077371\n",
      "Epoch : 63 [0/36(0%)]\tLoss: 0.718926\n",
      "Epoch : 63 [4/36(11%)]\tLoss: 0.562798\n",
      "Epoch : 63 [8/36(22%)]\tLoss: 0.068516\n",
      "Epoch : 63 [12/36(33%)]\tLoss: 0.053405\n",
      "Epoch : 63 [16/36(44%)]\tLoss: 0.954837\n",
      "Epoch : 63 [20/36(56%)]\tLoss: 1.849401\n",
      "Epoch : 63 [24/36(67%)]\tLoss: 0.620505\n",
      "Epoch : 63 [28/36(78%)]\tLoss: 0.780675\n",
      "Epoch : 63 [32/36(89%)]\tLoss: 0.403958\n",
      "Epoch : 64 [0/36(0%)]\tLoss: 0.343337\n",
      "Epoch : 64 [4/36(11%)]\tLoss: 3.238760\n",
      "Epoch : 64 [8/36(22%)]\tLoss: 1.898067\n",
      "Epoch : 64 [12/36(33%)]\tLoss: 2.115777\n",
      "Epoch : 64 [16/36(44%)]\tLoss: 0.848143\n",
      "Epoch : 64 [20/36(56%)]\tLoss: 0.806305\n",
      "Epoch : 64 [24/36(67%)]\tLoss: 0.452060\n",
      "Epoch : 64 [28/36(78%)]\tLoss: 2.463549\n",
      "Epoch : 64 [32/36(89%)]\tLoss: 2.090080\n",
      "Epoch : 65 [0/36(0%)]\tLoss: 0.594378\n",
      "Epoch : 65 [4/36(11%)]\tLoss: 0.849742\n",
      "Epoch : 65 [8/36(22%)]\tLoss: 0.874032\n",
      "Epoch : 65 [12/36(33%)]\tLoss: 1.964415\n",
      "Epoch : 65 [16/36(44%)]\tLoss: 0.642051\n",
      "Epoch : 65 [20/36(56%)]\tLoss: 0.477361\n",
      "Epoch : 65 [24/36(67%)]\tLoss: 0.120294\n",
      "Epoch : 65 [28/36(78%)]\tLoss: 1.078532\n",
      "Epoch : 65 [32/36(89%)]\tLoss: 0.903756\n",
      "Epoch : 66 [0/36(0%)]\tLoss: 0.514669\n",
      "Epoch : 66 [4/36(11%)]\tLoss: 0.824379\n",
      "Epoch : 66 [8/36(22%)]\tLoss: 1.023315\n",
      "Epoch : 66 [12/36(33%)]\tLoss: 0.381200\n",
      "Epoch : 66 [16/36(44%)]\tLoss: 1.037312\n",
      "Epoch : 66 [20/36(56%)]\tLoss: 0.903365\n",
      "Epoch : 66 [24/36(67%)]\tLoss: 0.774563\n",
      "Epoch : 66 [28/36(78%)]\tLoss: 2.397466\n",
      "Epoch : 66 [32/36(89%)]\tLoss: 1.076071\n",
      "Epoch : 67 [0/36(0%)]\tLoss: 0.852245\n",
      "Epoch : 67 [4/36(11%)]\tLoss: 0.449035\n",
      "Epoch : 67 [8/36(22%)]\tLoss: 1.569142\n",
      "Epoch : 67 [12/36(33%)]\tLoss: 1.346696\n",
      "Epoch : 67 [16/36(44%)]\tLoss: 3.071082\n",
      "Epoch : 67 [20/36(56%)]\tLoss: 1.506607\n",
      "Epoch : 67 [24/36(67%)]\tLoss: 0.260695\n",
      "Epoch : 67 [28/36(78%)]\tLoss: 1.698182\n",
      "Epoch : 67 [32/36(89%)]\tLoss: 0.899683\n",
      "Epoch : 68 [0/36(0%)]\tLoss: 1.478061\n",
      "Epoch : 68 [4/36(11%)]\tLoss: 1.039874\n",
      "Epoch : 68 [8/36(22%)]\tLoss: 0.334089\n",
      "Epoch : 68 [12/36(33%)]\tLoss: 0.411375\n",
      "Epoch : 68 [16/36(44%)]\tLoss: 0.879305\n",
      "Epoch : 68 [20/36(56%)]\tLoss: 0.183512\n",
      "Epoch : 68 [24/36(67%)]\tLoss: 0.230084\n",
      "Epoch : 68 [28/36(78%)]\tLoss: 1.500132\n",
      "Epoch : 68 [32/36(89%)]\tLoss: 0.340366\n",
      "Epoch : 69 [0/36(0%)]\tLoss: 0.258355\n",
      "Epoch : 69 [4/36(11%)]\tLoss: 0.589765\n",
      "Epoch : 69 [8/36(22%)]\tLoss: 0.319899\n",
      "Epoch : 69 [12/36(33%)]\tLoss: 0.374496\n",
      "Epoch : 69 [16/36(44%)]\tLoss: 0.551007\n",
      "Epoch : 69 [20/36(56%)]\tLoss: 0.225988\n",
      "Epoch : 69 [24/36(67%)]\tLoss: 0.099364\n",
      "Epoch : 69 [28/36(78%)]\tLoss: 1.631904\n",
      "Epoch : 69 [32/36(89%)]\tLoss: 0.466498\n",
      "Epoch : 70 [0/36(0%)]\tLoss: 0.407466\n",
      "Epoch : 70 [4/36(11%)]\tLoss: 0.420138\n",
      "Epoch : 70 [8/36(22%)]\tLoss: 0.267836\n",
      "Epoch : 70 [12/36(33%)]\tLoss: 0.253649\n",
      "Epoch : 70 [16/36(44%)]\tLoss: 0.286114\n",
      "Epoch : 70 [20/36(56%)]\tLoss: 0.232113\n",
      "Epoch : 70 [24/36(67%)]\tLoss: 0.371424\n",
      "Epoch : 70 [28/36(78%)]\tLoss: 1.118470\n",
      "Epoch : 70 [32/36(89%)]\tLoss: 0.191333\n",
      "Epoch : 71 [0/36(0%)]\tLoss: 0.264010\n",
      "Epoch : 71 [4/36(11%)]\tLoss: 0.657681\n",
      "Epoch : 71 [8/36(22%)]\tLoss: 0.514348\n",
      "Epoch : 71 [12/36(33%)]\tLoss: 0.322641\n",
      "Epoch : 71 [16/36(44%)]\tLoss: 0.126045\n",
      "Epoch : 71 [20/36(56%)]\tLoss: 0.078214\n",
      "Epoch : 71 [24/36(67%)]\tLoss: 0.275537\n",
      "Epoch : 71 [28/36(78%)]\tLoss: 1.317531\n",
      "Epoch : 71 [32/36(89%)]\tLoss: 0.173586\n",
      "Epoch : 72 [0/36(0%)]\tLoss: 0.129451\n",
      "Epoch : 72 [4/36(11%)]\tLoss: 0.097578\n",
      "Epoch : 72 [8/36(22%)]\tLoss: 0.393987\n",
      "Epoch : 72 [12/36(33%)]\tLoss: 0.435156\n",
      "Epoch : 72 [16/36(44%)]\tLoss: 0.338715\n",
      "Epoch : 72 [20/36(56%)]\tLoss: 0.152991\n",
      "Epoch : 72 [24/36(67%)]\tLoss: 0.142891\n",
      "Epoch : 72 [28/36(78%)]\tLoss: 0.721927\n",
      "Epoch : 72 [32/36(89%)]\tLoss: 0.383034\n",
      "Epoch : 73 [0/36(0%)]\tLoss: 0.204317\n",
      "Epoch : 73 [4/36(11%)]\tLoss: 0.376913\n",
      "Epoch : 73 [8/36(22%)]\tLoss: 0.193951\n",
      "Epoch : 73 [12/36(33%)]\tLoss: 0.297517\n",
      "Epoch : 73 [16/36(44%)]\tLoss: 0.362754\n",
      "Epoch : 73 [20/36(56%)]\tLoss: 0.083240\n",
      "Epoch : 73 [24/36(67%)]\tLoss: 0.230178\n",
      "Epoch : 73 [28/36(78%)]\tLoss: 0.512106\n",
      "Epoch : 73 [32/36(89%)]\tLoss: 0.302204\n",
      "Epoch : 74 [0/36(0%)]\tLoss: 0.253307\n",
      "Epoch : 74 [4/36(11%)]\tLoss: 0.189126\n",
      "Epoch : 74 [8/36(22%)]\tLoss: 0.203764\n",
      "Epoch : 74 [12/36(33%)]\tLoss: 0.179353\n",
      "Epoch : 74 [16/36(44%)]\tLoss: 0.736090\n",
      "Epoch : 74 [20/36(56%)]\tLoss: 0.530914\n",
      "Epoch : 74 [24/36(67%)]\tLoss: 0.448075\n",
      "Epoch : 74 [28/36(78%)]\tLoss: 1.057136\n",
      "Epoch : 74 [32/36(89%)]\tLoss: 0.402945\n",
      "Epoch : 75 [0/36(0%)]\tLoss: 0.178742\n",
      "Epoch : 75 [4/36(11%)]\tLoss: 0.316218\n",
      "Epoch : 75 [8/36(22%)]\tLoss: 0.547905\n",
      "Epoch : 75 [12/36(33%)]\tLoss: 1.802518\n",
      "Epoch : 75 [16/36(44%)]\tLoss: 0.106681\n",
      "Epoch : 75 [20/36(56%)]\tLoss: 0.236099\n",
      "Epoch : 75 [24/36(67%)]\tLoss: 0.235181\n",
      "Epoch : 75 [28/36(78%)]\tLoss: 0.497176\n",
      "Epoch : 75 [32/36(89%)]\tLoss: 0.234107\n",
      "Epoch : 76 [0/36(0%)]\tLoss: 0.337848\n",
      "Epoch : 76 [4/36(11%)]\tLoss: 0.257877\n",
      "Epoch : 76 [8/36(22%)]\tLoss: 0.164606\n",
      "Epoch : 76 [12/36(33%)]\tLoss: 0.182177\n",
      "Epoch : 76 [16/36(44%)]\tLoss: 0.717128\n",
      "Epoch : 76 [20/36(56%)]\tLoss: 0.469179\n",
      "Epoch : 76 [24/36(67%)]\tLoss: 0.440587\n",
      "Epoch : 76 [28/36(78%)]\tLoss: 0.284575\n",
      "Epoch : 76 [32/36(89%)]\tLoss: 0.185150\n",
      "Epoch : 77 [0/36(0%)]\tLoss: 0.372343\n",
      "Epoch : 77 [4/36(11%)]\tLoss: 0.950824\n",
      "Epoch : 77 [8/36(22%)]\tLoss: 1.528706\n",
      "Epoch : 77 [12/36(33%)]\tLoss: 0.545178\n",
      "Epoch : 77 [16/36(44%)]\tLoss: 1.017160\n",
      "Epoch : 77 [20/36(56%)]\tLoss: 0.598669\n",
      "Epoch : 77 [24/36(67%)]\tLoss: 1.215366\n",
      "Epoch : 77 [28/36(78%)]\tLoss: 1.143411\n",
      "Epoch : 77 [32/36(89%)]\tLoss: 0.477086\n",
      "Epoch : 78 [0/36(0%)]\tLoss: 0.248793\n",
      "Epoch : 78 [4/36(11%)]\tLoss: 1.218502\n",
      "Epoch : 78 [8/36(22%)]\tLoss: 1.275512\n",
      "Epoch : 78 [12/36(33%)]\tLoss: 1.236294\n",
      "Epoch : 78 [16/36(44%)]\tLoss: 0.127997\n",
      "Epoch : 78 [20/36(56%)]\tLoss: 0.078595\n",
      "Epoch : 78 [24/36(67%)]\tLoss: 0.317337\n",
      "Epoch : 78 [28/36(78%)]\tLoss: 1.148836\n",
      "Epoch : 78 [32/36(89%)]\tLoss: 1.173112\n",
      "Epoch : 79 [0/36(0%)]\tLoss: 0.565380\n",
      "Epoch : 79 [4/36(11%)]\tLoss: 0.567879\n",
      "Epoch : 79 [8/36(22%)]\tLoss: 0.297542\n",
      "Epoch : 79 [12/36(33%)]\tLoss: 0.535070\n",
      "Epoch : 79 [16/36(44%)]\tLoss: 0.408744\n",
      "Epoch : 79 [20/36(56%)]\tLoss: 0.450857\n",
      "Epoch : 79 [24/36(67%)]\tLoss: 0.494538\n",
      "Epoch : 79 [28/36(78%)]\tLoss: 0.896840\n",
      "Epoch : 79 [32/36(89%)]\tLoss: 0.327353\n",
      "Epoch : 80 [0/36(0%)]\tLoss: 0.310786\n",
      "Epoch : 80 [4/36(11%)]\tLoss: 0.513216\n",
      "Epoch : 80 [8/36(22%)]\tLoss: 0.712861\n",
      "Epoch : 80 [12/36(33%)]\tLoss: 0.093425\n",
      "Epoch : 80 [16/36(44%)]\tLoss: 0.006736\n",
      "Epoch : 80 [20/36(56%)]\tLoss: 0.239421\n",
      "Epoch : 80 [24/36(67%)]\tLoss: 0.056736\n",
      "Epoch : 80 [28/36(78%)]\tLoss: 0.524854\n",
      "Epoch : 80 [32/36(89%)]\tLoss: 0.320070\n",
      "Epoch : 81 [0/36(0%)]\tLoss: 0.397217\n",
      "Epoch : 81 [4/36(11%)]\tLoss: 1.626582\n",
      "Epoch : 81 [8/36(22%)]\tLoss: 0.133639\n",
      "Epoch : 81 [12/36(33%)]\tLoss: 0.308262\n",
      "Epoch : 81 [16/36(44%)]\tLoss: 0.197422\n",
      "Epoch : 81 [20/36(56%)]\tLoss: 0.776269\n",
      "Epoch : 81 [24/36(67%)]\tLoss: 0.907187\n",
      "Epoch : 81 [28/36(78%)]\tLoss: 0.399915\n",
      "Epoch : 81 [32/36(89%)]\tLoss: 0.163212\n",
      "Epoch : 82 [0/36(0%)]\tLoss: 0.252094\n",
      "Epoch : 82 [4/36(11%)]\tLoss: 1.120031\n",
      "Epoch : 82 [8/36(22%)]\tLoss: 0.512468\n",
      "Epoch : 82 [12/36(33%)]\tLoss: 0.543376\n",
      "Epoch : 82 [16/36(44%)]\tLoss: 0.638372\n",
      "Epoch : 82 [20/36(56%)]\tLoss: 0.439703\n",
      "Epoch : 82 [24/36(67%)]\tLoss: 0.237680\n",
      "Epoch : 82 [28/36(78%)]\tLoss: 0.430955\n",
      "Epoch : 82 [32/36(89%)]\tLoss: 1.240114\n",
      "Epoch : 83 [0/36(0%)]\tLoss: 1.292049\n",
      "Epoch : 83 [4/36(11%)]\tLoss: 0.369022\n",
      "Epoch : 83 [8/36(22%)]\tLoss: 0.218486\n",
      "Epoch : 83 [12/36(33%)]\tLoss: 0.256118\n",
      "Epoch : 83 [16/36(44%)]\tLoss: 1.272993\n",
      "Epoch : 83 [20/36(56%)]\tLoss: 0.645443\n",
      "Epoch : 83 [24/36(67%)]\tLoss: 1.229238\n",
      "Epoch : 83 [28/36(78%)]\tLoss: 0.308838\n",
      "Epoch : 83 [32/36(89%)]\tLoss: 0.714647\n",
      "Epoch : 84 [0/36(0%)]\tLoss: 1.451609\n",
      "Epoch : 84 [4/36(11%)]\tLoss: 0.643531\n",
      "Epoch : 84 [8/36(22%)]\tLoss: 1.047234\n",
      "Epoch : 84 [12/36(33%)]\tLoss: 0.390438\n",
      "Epoch : 84 [16/36(44%)]\tLoss: 0.120879\n",
      "Epoch : 84 [20/36(56%)]\tLoss: 0.509526\n",
      "Epoch : 84 [24/36(67%)]\tLoss: 1.253399\n",
      "Epoch : 84 [28/36(78%)]\tLoss: 2.891810\n",
      "Epoch : 84 [32/36(89%)]\tLoss: 0.572470\n",
      "Epoch : 85 [0/36(0%)]\tLoss: 0.216476\n",
      "Epoch : 85 [4/36(11%)]\tLoss: 0.225592\n",
      "Epoch : 85 [8/36(22%)]\tLoss: 0.979569\n",
      "Epoch : 85 [12/36(33%)]\tLoss: 0.831647\n",
      "Epoch : 85 [16/36(44%)]\tLoss: 0.666355\n",
      "Epoch : 85 [20/36(56%)]\tLoss: 0.771376\n",
      "Epoch : 85 [24/36(67%)]\tLoss: 0.492060\n",
      "Epoch : 85 [28/36(78%)]\tLoss: 1.350610\n",
      "Epoch : 85 [32/36(89%)]\tLoss: 0.598870\n",
      "Epoch : 86 [0/36(0%)]\tLoss: 0.689926\n",
      "Epoch : 86 [4/36(11%)]\tLoss: 1.346202\n",
      "Epoch : 86 [8/36(22%)]\tLoss: 0.158690\n",
      "Epoch : 86 [12/36(33%)]\tLoss: 0.139311\n",
      "Epoch : 86 [16/36(44%)]\tLoss: 0.630269\n",
      "Epoch : 86 [20/36(56%)]\tLoss: 1.786239\n",
      "Epoch : 86 [24/36(67%)]\tLoss: 1.276744\n",
      "Epoch : 86 [28/36(78%)]\tLoss: 0.939706\n",
      "Epoch : 86 [32/36(89%)]\tLoss: 0.408783\n",
      "Epoch : 87 [0/36(0%)]\tLoss: 0.466741\n",
      "Epoch : 87 [4/36(11%)]\tLoss: 1.702614\n",
      "Epoch : 87 [8/36(22%)]\tLoss: 0.605949\n",
      "Epoch : 87 [12/36(33%)]\tLoss: 1.212143\n",
      "Epoch : 87 [16/36(44%)]\tLoss: 0.189703\n",
      "Epoch : 87 [20/36(56%)]\tLoss: 0.265069\n",
      "Epoch : 87 [24/36(67%)]\tLoss: 0.666991\n",
      "Epoch : 87 [28/36(78%)]\tLoss: 0.783599\n",
      "Epoch : 87 [32/36(89%)]\tLoss: 1.595242\n",
      "Epoch : 88 [0/36(0%)]\tLoss: 0.428605\n",
      "Epoch : 88 [4/36(11%)]\tLoss: 0.428047\n",
      "Epoch : 88 [8/36(22%)]\tLoss: 0.324079\n",
      "Epoch : 88 [12/36(33%)]\tLoss: 1.012025\n",
      "Epoch : 88 [16/36(44%)]\tLoss: 1.761783\n",
      "Epoch : 88 [20/36(56%)]\tLoss: 0.056252\n",
      "Epoch : 88 [24/36(67%)]\tLoss: 0.090969\n",
      "Epoch : 88 [28/36(78%)]\tLoss: 0.539378\n",
      "Epoch : 88 [32/36(89%)]\tLoss: 1.241442\n",
      "Epoch : 89 [0/36(0%)]\tLoss: 1.213862\n",
      "Epoch : 89 [4/36(11%)]\tLoss: 0.723440\n",
      "Epoch : 89 [8/36(22%)]\tLoss: 0.400293\n",
      "Epoch : 89 [12/36(33%)]\tLoss: 0.234032\n",
      "Epoch : 89 [16/36(44%)]\tLoss: 1.147678\n",
      "Epoch : 89 [20/36(56%)]\tLoss: 1.026823\n",
      "Epoch : 89 [24/36(67%)]\tLoss: 0.551719\n",
      "Epoch : 89 [28/36(78%)]\tLoss: 0.752412\n",
      "Epoch : 89 [32/36(89%)]\tLoss: 0.216397\n",
      "Epoch : 90 [0/36(0%)]\tLoss: 0.090137\n",
      "Epoch : 90 [4/36(11%)]\tLoss: 0.304487\n",
      "Epoch : 90 [8/36(22%)]\tLoss: 0.939765\n",
      "Epoch : 90 [12/36(33%)]\tLoss: 0.871449\n",
      "Epoch : 90 [16/36(44%)]\tLoss: 1.234147\n",
      "Epoch : 90 [20/36(56%)]\tLoss: 0.487888\n",
      "Epoch : 90 [24/36(67%)]\tLoss: 0.402278\n",
      "Epoch : 90 [28/36(78%)]\tLoss: 0.940601\n",
      "Epoch : 90 [32/36(89%)]\tLoss: 0.986107\n",
      "Epoch : 91 [0/36(0%)]\tLoss: 1.807013\n",
      "Epoch : 91 [4/36(11%)]\tLoss: 0.686301\n",
      "Epoch : 91 [8/36(22%)]\tLoss: 0.307709\n",
      "Epoch : 91 [12/36(33%)]\tLoss: 0.726971\n",
      "Epoch : 91 [16/36(44%)]\tLoss: 1.401892\n",
      "Epoch : 91 [20/36(56%)]\tLoss: 1.299923\n",
      "Epoch : 91 [24/36(67%)]\tLoss: 0.150487\n",
      "Epoch : 91 [28/36(78%)]\tLoss: 0.319018\n",
      "Epoch : 91 [32/36(89%)]\tLoss: 0.151229\n",
      "Epoch : 92 [0/36(0%)]\tLoss: 0.415080\n",
      "Epoch : 92 [4/36(11%)]\tLoss: 2.087612\n",
      "Epoch : 92 [8/36(22%)]\tLoss: 0.749638\n",
      "Epoch : 92 [12/36(33%)]\tLoss: 0.330674\n",
      "Epoch : 92 [16/36(44%)]\tLoss: 0.907063\n",
      "Epoch : 92 [20/36(56%)]\tLoss: 1.101265\n",
      "Epoch : 92 [24/36(67%)]\tLoss: 0.521061\n",
      "Epoch : 92 [28/36(78%)]\tLoss: 1.948056\n",
      "Epoch : 92 [32/36(89%)]\tLoss: 1.321537\n",
      "Epoch : 93 [0/36(0%)]\tLoss: 0.506191\n",
      "Epoch : 93 [4/36(11%)]\tLoss: 0.810635\n",
      "Epoch : 93 [8/36(22%)]\tLoss: 0.600879\n",
      "Epoch : 93 [12/36(33%)]\tLoss: 2.251687\n",
      "Epoch : 93 [16/36(44%)]\tLoss: 1.685602\n",
      "Epoch : 93 [20/36(56%)]\tLoss: 0.318683\n",
      "Epoch : 93 [24/36(67%)]\tLoss: 0.041152\n",
      "Epoch : 93 [28/36(78%)]\tLoss: 0.639703\n",
      "Epoch : 93 [32/36(89%)]\tLoss: 2.379704\n",
      "Epoch : 94 [0/36(0%)]\tLoss: 1.253242\n",
      "Epoch : 94 [4/36(11%)]\tLoss: 0.739708\n",
      "Epoch : 94 [8/36(22%)]\tLoss: 0.057108\n",
      "Epoch : 94 [12/36(33%)]\tLoss: 0.326923\n",
      "Epoch : 94 [16/36(44%)]\tLoss: 1.424239\n",
      "Epoch : 94 [20/36(56%)]\tLoss: 0.478328\n",
      "Epoch : 94 [24/36(67%)]\tLoss: 0.503424\n",
      "Epoch : 94 [28/36(78%)]\tLoss: 0.440383\n",
      "Epoch : 94 [32/36(89%)]\tLoss: 0.304647\n",
      "Epoch : 95 [0/36(0%)]\tLoss: 0.139066\n",
      "Epoch : 95 [4/36(11%)]\tLoss: 0.379277\n",
      "Epoch : 95 [8/36(22%)]\tLoss: 1.024751\n",
      "Epoch : 95 [12/36(33%)]\tLoss: 0.299449\n",
      "Epoch : 95 [16/36(44%)]\tLoss: 0.556933\n",
      "Epoch : 95 [20/36(56%)]\tLoss: 0.403408\n",
      "Epoch : 95 [24/36(67%)]\tLoss: 0.493916\n",
      "Epoch : 95 [28/36(78%)]\tLoss: 1.078770\n",
      "Epoch : 95 [32/36(89%)]\tLoss: 1.061820\n",
      "Epoch : 96 [0/36(0%)]\tLoss: 0.524650\n",
      "Epoch : 96 [4/36(11%)]\tLoss: 0.211743\n",
      "Epoch : 96 [8/36(22%)]\tLoss: 0.293448\n",
      "Epoch : 96 [12/36(33%)]\tLoss: 0.823350\n",
      "Epoch : 96 [16/36(44%)]\tLoss: 1.064321\n",
      "Epoch : 96 [20/36(56%)]\tLoss: 0.979206\n",
      "Epoch : 96 [24/36(67%)]\tLoss: 0.268117\n",
      "Epoch : 96 [28/36(78%)]\tLoss: 0.417737\n",
      "Epoch : 96 [32/36(89%)]\tLoss: 0.938825\n",
      "Epoch : 97 [0/36(0%)]\tLoss: 0.927347\n",
      "Epoch : 97 [4/36(11%)]\tLoss: 0.995706\n",
      "Epoch : 97 [8/36(22%)]\tLoss: 0.035987\n",
      "Epoch : 97 [12/36(33%)]\tLoss: 0.279080\n",
      "Epoch : 97 [16/36(44%)]\tLoss: 0.315906\n",
      "Epoch : 97 [20/36(56%)]\tLoss: 1.367435\n",
      "Epoch : 97 [24/36(67%)]\tLoss: 0.544782\n",
      "Epoch : 97 [28/36(78%)]\tLoss: 0.318281\n",
      "Epoch : 97 [32/36(89%)]\tLoss: 0.281093\n",
      "Epoch : 98 [0/36(0%)]\tLoss: 0.124036\n",
      "Epoch : 98 [4/36(11%)]\tLoss: 0.740108\n",
      "Epoch : 98 [8/36(22%)]\tLoss: 0.578126\n",
      "Epoch : 98 [12/36(33%)]\tLoss: 0.272565\n",
      "Epoch : 98 [16/36(44%)]\tLoss: 0.110221\n",
      "Epoch : 98 [20/36(56%)]\tLoss: 0.883819\n",
      "Epoch : 98 [24/36(67%)]\tLoss: 0.324939\n",
      "Epoch : 98 [28/36(78%)]\tLoss: 0.644794\n",
      "Epoch : 98 [32/36(89%)]\tLoss: 0.287560\n",
      "Epoch : 99 [0/36(0%)]\tLoss: 0.285752\n",
      "Epoch : 99 [4/36(11%)]\tLoss: 0.192318\n",
      "Epoch : 99 [8/36(22%)]\tLoss: 0.337973\n",
      "Epoch : 99 [12/36(33%)]\tLoss: 0.770591\n",
      "Epoch : 99 [16/36(44%)]\tLoss: 1.130126\n",
      "Epoch : 99 [20/36(56%)]\tLoss: 0.947934\n",
      "Epoch : 99 [24/36(67%)]\tLoss: 0.078783\n",
      "Epoch : 99 [28/36(78%)]\tLoss: 0.813531\n",
      "Epoch : 99 [32/36(89%)]\tLoss: 1.470579\n",
      "Epoch : 100 [0/36(0%)]\tLoss: 1.656416\n",
      "Epoch : 100 [4/36(11%)]\tLoss: 0.296373\n",
      "Epoch : 100 [8/36(22%)]\tLoss: 0.394930\n",
      "Epoch : 100 [12/36(33%)]\tLoss: 1.364865\n",
      "Epoch : 100 [16/36(44%)]\tLoss: 0.742887\n",
      "Epoch : 100 [20/36(56%)]\tLoss: 0.912956\n",
      "Epoch : 100 [24/36(67%)]\tLoss: 0.191225\n",
      "Epoch : 100 [28/36(78%)]\tLoss: 0.468934\n",
      "Epoch : 100 [32/36(89%)]\tLoss: 0.699587\n",
      "Epoch : 101 [0/36(0%)]\tLoss: 0.693993\n",
      "Epoch : 101 [4/36(11%)]\tLoss: 0.203357\n",
      "Epoch : 101 [8/36(22%)]\tLoss: 0.740956\n",
      "Epoch : 101 [12/36(33%)]\tLoss: 0.087016\n",
      "Epoch : 101 [16/36(44%)]\tLoss: 0.227309\n",
      "Epoch : 101 [20/36(56%)]\tLoss: 1.077742\n",
      "Epoch : 101 [24/36(67%)]\tLoss: 0.992182\n",
      "Epoch : 101 [28/36(78%)]\tLoss: 0.315228\n",
      "Epoch : 101 [32/36(89%)]\tLoss: 0.151752\n",
      "Epoch : 102 [0/36(0%)]\tLoss: 0.108489\n",
      "Epoch : 102 [4/36(11%)]\tLoss: 0.035975\n",
      "Epoch : 102 [8/36(22%)]\tLoss: 0.330176\n",
      "Epoch : 102 [12/36(33%)]\tLoss: 0.162562\n",
      "Epoch : 102 [16/36(44%)]\tLoss: 0.419001\n",
      "Epoch : 102 [20/36(56%)]\tLoss: 0.715099\n",
      "Epoch : 102 [24/36(67%)]\tLoss: 0.256195\n",
      "Epoch : 102 [28/36(78%)]\tLoss: 0.636999\n",
      "Epoch : 102 [32/36(89%)]\tLoss: 0.255898\n",
      "Epoch : 103 [0/36(0%)]\tLoss: 0.020008\n",
      "Epoch : 103 [4/36(11%)]\tLoss: 0.152856\n",
      "Epoch : 103 [8/36(22%)]\tLoss: 0.193847\n",
      "Epoch : 103 [12/36(33%)]\tLoss: 0.102329\n",
      "Epoch : 103 [16/36(44%)]\tLoss: 0.351786\n",
      "Epoch : 103 [20/36(56%)]\tLoss: 0.789241\n",
      "Epoch : 103 [24/36(67%)]\tLoss: 0.195083\n",
      "Epoch : 103 [28/36(78%)]\tLoss: 0.312350\n",
      "Epoch : 103 [32/36(89%)]\tLoss: 0.365818\n",
      "Epoch : 104 [0/36(0%)]\tLoss: 0.181411\n",
      "Epoch : 104 [4/36(11%)]\tLoss: 0.273911\n",
      "Epoch : 104 [8/36(22%)]\tLoss: 0.081439\n",
      "Epoch : 104 [12/36(33%)]\tLoss: 0.048998\n",
      "Epoch : 104 [16/36(44%)]\tLoss: 0.611410\n",
      "Epoch : 104 [20/36(56%)]\tLoss: 0.722970\n",
      "Epoch : 104 [24/36(67%)]\tLoss: 0.298369\n",
      "Epoch : 104 [28/36(78%)]\tLoss: 0.214827\n",
      "Epoch : 104 [32/36(89%)]\tLoss: 0.160282\n",
      "Epoch : 105 [0/36(0%)]\tLoss: 0.268735\n",
      "Epoch : 105 [4/36(11%)]\tLoss: 0.813517\n",
      "Epoch : 105 [8/36(22%)]\tLoss: 0.022932\n",
      "Epoch : 105 [12/36(33%)]\tLoss: 0.069016\n",
      "Epoch : 105 [16/36(44%)]\tLoss: 0.607992\n",
      "Epoch : 105 [20/36(56%)]\tLoss: 0.405917\n",
      "Epoch : 105 [24/36(67%)]\tLoss: 0.496228\n",
      "Epoch : 105 [28/36(78%)]\tLoss: 0.271842\n",
      "Epoch : 105 [32/36(89%)]\tLoss: 0.476250\n",
      "Epoch : 106 [0/36(0%)]\tLoss: 0.082213\n",
      "Epoch : 106 [4/36(11%)]\tLoss: 0.464349\n",
      "Epoch : 106 [8/36(22%)]\tLoss: 0.240906\n",
      "Epoch : 106 [12/36(33%)]\tLoss: 0.244132\n",
      "Epoch : 106 [16/36(44%)]\tLoss: 0.091099\n",
      "Epoch : 106 [20/36(56%)]\tLoss: 0.311333\n",
      "Epoch : 106 [24/36(67%)]\tLoss: 0.044099\n",
      "Epoch : 106 [28/36(78%)]\tLoss: 0.419591\n",
      "Epoch : 106 [32/36(89%)]\tLoss: 0.479224\n",
      "Epoch : 107 [0/36(0%)]\tLoss: 0.108300\n",
      "Epoch : 107 [4/36(11%)]\tLoss: 0.160818\n",
      "Epoch : 107 [8/36(22%)]\tLoss: 0.121201\n",
      "Epoch : 107 [12/36(33%)]\tLoss: 0.258312\n",
      "Epoch : 107 [16/36(44%)]\tLoss: 0.612813\n",
      "Epoch : 107 [20/36(56%)]\tLoss: 0.048103\n",
      "Epoch : 107 [24/36(67%)]\tLoss: 0.227457\n",
      "Epoch : 107 [28/36(78%)]\tLoss: 0.213894\n",
      "Epoch : 107 [32/36(89%)]\tLoss: 0.118591\n",
      "Epoch : 108 [0/36(0%)]\tLoss: 0.050458\n",
      "Epoch : 108 [4/36(11%)]\tLoss: 0.381968\n",
      "Epoch : 108 [8/36(22%)]\tLoss: 0.626733\n",
      "Epoch : 108 [12/36(33%)]\tLoss: 0.058010\n",
      "Epoch : 108 [16/36(44%)]\tLoss: 0.247354\n",
      "Epoch : 108 [20/36(56%)]\tLoss: 0.046174\n",
      "Epoch : 108 [24/36(67%)]\tLoss: 0.582927\n",
      "Epoch : 108 [28/36(78%)]\tLoss: 0.549752\n",
      "Epoch : 108 [32/36(89%)]\tLoss: 0.289972\n",
      "Epoch : 109 [0/36(0%)]\tLoss: 0.110717\n",
      "Epoch : 109 [4/36(11%)]\tLoss: 0.105906\n",
      "Epoch : 109 [8/36(22%)]\tLoss: 0.570869\n",
      "Epoch : 109 [12/36(33%)]\tLoss: 0.290837\n",
      "Epoch : 109 [16/36(44%)]\tLoss: 0.500940\n",
      "Epoch : 109 [20/36(56%)]\tLoss: 0.088384\n",
      "Epoch : 109 [24/36(67%)]\tLoss: 0.331014\n",
      "Epoch : 109 [28/36(78%)]\tLoss: 0.560172\n",
      "Epoch : 109 [32/36(89%)]\tLoss: 0.214552\n",
      "Epoch : 110 [0/36(0%)]\tLoss: 0.209961\n",
      "Epoch : 110 [4/36(11%)]\tLoss: 0.234349\n",
      "Epoch : 110 [8/36(22%)]\tLoss: 0.157858\n",
      "Epoch : 110 [12/36(33%)]\tLoss: 0.059408\n",
      "Epoch : 110 [16/36(44%)]\tLoss: 0.632840\n",
      "Epoch : 110 [20/36(56%)]\tLoss: 0.787808\n",
      "Epoch : 110 [24/36(67%)]\tLoss: 0.182956\n",
      "Epoch : 110 [28/36(78%)]\tLoss: 0.079478\n",
      "Epoch : 110 [32/36(89%)]\tLoss: 0.133135\n",
      "Epoch : 111 [0/36(0%)]\tLoss: 0.520723\n",
      "Epoch : 111 [4/36(11%)]\tLoss: 0.221821\n",
      "Epoch : 111 [8/36(22%)]\tLoss: 0.095664\n",
      "Epoch : 111 [12/36(33%)]\tLoss: 0.043037\n",
      "Epoch : 111 [16/36(44%)]\tLoss: 0.205640\n",
      "Epoch : 111 [20/36(56%)]\tLoss: 0.311723\n",
      "Epoch : 111 [24/36(67%)]\tLoss: 0.153568\n",
      "Epoch : 111 [28/36(78%)]\tLoss: 0.625166\n",
      "Epoch : 111 [32/36(89%)]\tLoss: 0.091304\n",
      "Epoch : 112 [0/36(0%)]\tLoss: 0.190186\n",
      "Epoch : 112 [4/36(11%)]\tLoss: 0.436641\n",
      "Epoch : 112 [8/36(22%)]\tLoss: 0.110545\n",
      "Epoch : 112 [12/36(33%)]\tLoss: 0.226720\n",
      "Epoch : 112 [16/36(44%)]\tLoss: 0.127711\n",
      "Epoch : 112 [20/36(56%)]\tLoss: 0.346472\n",
      "Epoch : 112 [24/36(67%)]\tLoss: 0.123816\n",
      "Epoch : 112 [28/36(78%)]\tLoss: 0.423376\n",
      "Epoch : 112 [32/36(89%)]\tLoss: 0.543780\n",
      "Epoch : 113 [0/36(0%)]\tLoss: 0.173488\n",
      "Epoch : 113 [4/36(11%)]\tLoss: 0.164447\n",
      "Epoch : 113 [8/36(22%)]\tLoss: 0.090284\n",
      "Epoch : 113 [12/36(33%)]\tLoss: 0.272537\n",
      "Epoch : 113 [16/36(44%)]\tLoss: 0.406808\n",
      "Epoch : 113 [20/36(56%)]\tLoss: 0.193106\n",
      "Epoch : 113 [24/36(67%)]\tLoss: 0.170911\n",
      "Epoch : 113 [28/36(78%)]\tLoss: 0.033004\n",
      "Epoch : 113 [32/36(89%)]\tLoss: 0.121307\n",
      "Epoch : 114 [0/36(0%)]\tLoss: 0.105593\n",
      "Epoch : 114 [4/36(11%)]\tLoss: 0.091189\n",
      "Epoch : 114 [8/36(22%)]\tLoss: 0.072856\n",
      "Epoch : 114 [12/36(33%)]\tLoss: 0.016102\n",
      "Epoch : 114 [16/36(44%)]\tLoss: 0.156448\n",
      "Epoch : 114 [20/36(56%)]\tLoss: 0.059616\n",
      "Epoch : 114 [24/36(67%)]\tLoss: 0.165676\n",
      "Epoch : 114 [28/36(78%)]\tLoss: 0.316213\n",
      "Epoch : 114 [32/36(89%)]\tLoss: 0.066291\n",
      "Epoch : 115 [0/36(0%)]\tLoss: 0.172965\n",
      "Epoch : 115 [4/36(11%)]\tLoss: 0.268435\n",
      "Epoch : 115 [8/36(22%)]\tLoss: 0.058343\n",
      "Epoch : 115 [12/36(33%)]\tLoss: 0.059689\n",
      "Epoch : 115 [16/36(44%)]\tLoss: 0.276091\n",
      "Epoch : 115 [20/36(56%)]\tLoss: 0.150128\n",
      "Epoch : 115 [24/36(67%)]\tLoss: 0.206634\n",
      "Epoch : 115 [28/36(78%)]\tLoss: 0.198630\n",
      "Epoch : 115 [32/36(89%)]\tLoss: 0.101946\n",
      "Epoch : 116 [0/36(0%)]\tLoss: 0.106101\n",
      "Epoch : 116 [4/36(11%)]\tLoss: 0.244015\n",
      "Epoch : 116 [8/36(22%)]\tLoss: 0.066768\n",
      "Epoch : 116 [12/36(33%)]\tLoss: 0.191041\n",
      "Epoch : 116 [16/36(44%)]\tLoss: 0.247495\n",
      "Epoch : 116 [20/36(56%)]\tLoss: 0.145999\n",
      "Epoch : 116 [24/36(67%)]\tLoss: 0.178439\n",
      "Epoch : 116 [28/36(78%)]\tLoss: 0.086537\n",
      "Epoch : 116 [32/36(89%)]\tLoss: 0.129315\n",
      "Epoch : 117 [0/36(0%)]\tLoss: 0.174346\n",
      "Epoch : 117 [4/36(11%)]\tLoss: 0.074372\n",
      "Epoch : 117 [8/36(22%)]\tLoss: 0.086500\n",
      "Epoch : 117 [12/36(33%)]\tLoss: 0.059157\n",
      "Epoch : 117 [16/36(44%)]\tLoss: 0.086128\n",
      "Epoch : 117 [20/36(56%)]\tLoss: 0.173520\n",
      "Epoch : 117 [24/36(67%)]\tLoss: 0.578364\n",
      "Epoch : 117 [28/36(78%)]\tLoss: 0.235941\n",
      "Epoch : 117 [32/36(89%)]\tLoss: 0.091722\n",
      "Epoch : 118 [0/36(0%)]\tLoss: 0.363392\n",
      "Epoch : 118 [4/36(11%)]\tLoss: 0.410303\n",
      "Epoch : 118 [8/36(22%)]\tLoss: 0.097592\n",
      "Epoch : 118 [12/36(33%)]\tLoss: 0.029516\n",
      "Epoch : 118 [16/36(44%)]\tLoss: 0.106401\n",
      "Epoch : 118 [20/36(56%)]\tLoss: 0.059144\n",
      "Epoch : 118 [24/36(67%)]\tLoss: 0.169211\n",
      "Epoch : 118 [28/36(78%)]\tLoss: 0.650244\n",
      "Epoch : 118 [32/36(89%)]\tLoss: 0.082858\n",
      "Epoch : 119 [0/36(0%)]\tLoss: 0.140430\n",
      "Epoch : 119 [4/36(11%)]\tLoss: 0.075207\n",
      "Epoch : 119 [8/36(22%)]\tLoss: 0.053820\n",
      "Epoch : 119 [12/36(33%)]\tLoss: 0.028375\n",
      "Epoch : 119 [16/36(44%)]\tLoss: 0.398878\n",
      "Epoch : 119 [20/36(56%)]\tLoss: 0.122729\n",
      "Epoch : 119 [24/36(67%)]\tLoss: 0.142674\n",
      "Epoch : 119 [28/36(78%)]\tLoss: 0.092062\n",
      "Epoch : 119 [32/36(89%)]\tLoss: 0.036081\n",
      "Epoch : 120 [0/36(0%)]\tLoss: 0.283199\n",
      "Epoch : 120 [4/36(11%)]\tLoss: 0.485780\n",
      "Epoch : 120 [8/36(22%)]\tLoss: 0.041864\n",
      "Epoch : 120 [12/36(33%)]\tLoss: 0.326766\n",
      "Epoch : 120 [16/36(44%)]\tLoss: 0.291629\n",
      "Epoch : 120 [20/36(56%)]\tLoss: 0.756765\n",
      "Epoch : 120 [24/36(67%)]\tLoss: 0.048102\n",
      "Epoch : 120 [28/36(78%)]\tLoss: 0.151921\n",
      "Epoch : 120 [32/36(89%)]\tLoss: 0.113426\n",
      "Epoch : 121 [0/36(0%)]\tLoss: 0.308737\n",
      "Epoch : 121 [4/36(11%)]\tLoss: 0.129108\n",
      "Epoch : 121 [8/36(22%)]\tLoss: 0.032633\n",
      "Epoch : 121 [12/36(33%)]\tLoss: 0.043534\n",
      "Epoch : 121 [16/36(44%)]\tLoss: 0.106559\n",
      "Epoch : 121 [20/36(56%)]\tLoss: 0.479028\n",
      "Epoch : 121 [24/36(67%)]\tLoss: 0.289511\n",
      "Epoch : 121 [28/36(78%)]\tLoss: 0.339022\n",
      "Epoch : 121 [32/36(89%)]\tLoss: 0.135735\n",
      "Epoch : 122 [0/36(0%)]\tLoss: 0.159383\n",
      "Epoch : 122 [4/36(11%)]\tLoss: 0.329505\n",
      "Epoch : 122 [8/36(22%)]\tLoss: 0.163652\n",
      "Epoch : 122 [12/36(33%)]\tLoss: 0.516763\n",
      "Epoch : 122 [16/36(44%)]\tLoss: 0.334249\n",
      "Epoch : 122 [20/36(56%)]\tLoss: 0.495905\n",
      "Epoch : 122 [24/36(67%)]\tLoss: 0.101896\n",
      "Epoch : 122 [28/36(78%)]\tLoss: 0.305475\n",
      "Epoch : 122 [32/36(89%)]\tLoss: 0.609204\n",
      "Epoch : 123 [0/36(0%)]\tLoss: 0.117383\n",
      "Epoch : 123 [4/36(11%)]\tLoss: 0.407645\n",
      "Epoch : 123 [8/36(22%)]\tLoss: 0.418990\n",
      "Epoch : 123 [12/36(33%)]\tLoss: 0.382046\n",
      "Epoch : 123 [16/36(44%)]\tLoss: 0.303330\n",
      "Epoch : 123 [20/36(56%)]\tLoss: 0.306819\n",
      "Epoch : 123 [24/36(67%)]\tLoss: 0.163441\n",
      "Epoch : 123 [28/36(78%)]\tLoss: 0.562952\n",
      "Epoch : 123 [32/36(89%)]\tLoss: 0.508310\n",
      "Epoch : 124 [0/36(0%)]\tLoss: 0.065835\n",
      "Epoch : 124 [4/36(11%)]\tLoss: 0.281677\n",
      "Epoch : 124 [8/36(22%)]\tLoss: 0.181778\n",
      "Epoch : 124 [12/36(33%)]\tLoss: 0.297746\n",
      "Epoch : 124 [16/36(44%)]\tLoss: 0.451363\n",
      "Epoch : 124 [20/36(56%)]\tLoss: 0.222398\n",
      "Epoch : 124 [24/36(67%)]\tLoss: 0.049337\n",
      "Epoch : 124 [28/36(78%)]\tLoss: 0.162715\n",
      "Epoch : 124 [32/36(89%)]\tLoss: 0.281520\n",
      "Epoch : 125 [0/36(0%)]\tLoss: 0.227619\n",
      "Epoch : 125 [4/36(11%)]\tLoss: 0.115717\n",
      "Epoch : 125 [8/36(22%)]\tLoss: 0.078361\n",
      "Epoch : 125 [12/36(33%)]\tLoss: 0.182788\n",
      "Epoch : 125 [16/36(44%)]\tLoss: 0.514610\n",
      "Epoch : 125 [20/36(56%)]\tLoss: 0.047526\n",
      "Epoch : 125 [24/36(67%)]\tLoss: 0.386758\n",
      "Epoch : 125 [28/36(78%)]\tLoss: 0.249806\n",
      "Epoch : 125 [32/36(89%)]\tLoss: 0.046495\n",
      "Epoch : 126 [0/36(0%)]\tLoss: 0.067463\n",
      "Epoch : 126 [4/36(11%)]\tLoss: 0.316769\n",
      "Epoch : 126 [8/36(22%)]\tLoss: 0.555038\n",
      "Epoch : 126 [12/36(33%)]\tLoss: 0.075643\n",
      "Epoch : 126 [16/36(44%)]\tLoss: 0.097208\n",
      "Epoch : 126 [20/36(56%)]\tLoss: 0.103073\n",
      "Epoch : 126 [24/36(67%)]\tLoss: 0.656007\n",
      "Epoch : 126 [28/36(78%)]\tLoss: 0.512177\n",
      "Epoch : 126 [32/36(89%)]\tLoss: 0.355737\n",
      "Epoch : 127 [0/36(0%)]\tLoss: 0.111695\n",
      "Epoch : 127 [4/36(11%)]\tLoss: 0.291507\n",
      "Epoch : 127 [8/36(22%)]\tLoss: 0.879242\n",
      "Epoch : 127 [12/36(33%)]\tLoss: 0.139404\n",
      "Epoch : 127 [16/36(44%)]\tLoss: 0.328845\n",
      "Epoch : 127 [20/36(56%)]\tLoss: 0.146543\n",
      "Epoch : 127 [24/36(67%)]\tLoss: 0.437699\n",
      "Epoch : 127 [28/36(78%)]\tLoss: 0.918302\n",
      "Epoch : 127 [32/36(89%)]\tLoss: 0.232255\n",
      "Epoch : 128 [0/36(0%)]\tLoss: 0.195201\n",
      "Epoch : 128 [4/36(11%)]\tLoss: 0.159555\n",
      "Epoch : 128 [8/36(22%)]\tLoss: 0.288615\n",
      "Epoch : 128 [12/36(33%)]\tLoss: 0.182769\n",
      "Epoch : 128 [16/36(44%)]\tLoss: 0.486691\n",
      "Epoch : 128 [20/36(56%)]\tLoss: 1.186558\n",
      "Epoch : 128 [24/36(67%)]\tLoss: 0.281083\n",
      "Epoch : 128 [28/36(78%)]\tLoss: 0.510006\n",
      "Epoch : 128 [32/36(89%)]\tLoss: 0.301064\n",
      "Epoch : 129 [0/36(0%)]\tLoss: 0.571860\n",
      "Epoch : 129 [4/36(11%)]\tLoss: 0.511510\n",
      "Epoch : 129 [8/36(22%)]\tLoss: 0.098110\n",
      "Epoch : 129 [12/36(33%)]\tLoss: 0.044901\n",
      "Epoch : 129 [16/36(44%)]\tLoss: 0.228436\n",
      "Epoch : 129 [20/36(56%)]\tLoss: 1.121233\n",
      "Epoch : 129 [24/36(67%)]\tLoss: 0.783711\n",
      "Epoch : 129 [28/36(78%)]\tLoss: 0.098817\n",
      "Epoch : 129 [32/36(89%)]\tLoss: 0.074182\n",
      "Epoch : 130 [0/36(0%)]\tLoss: 0.266093\n",
      "Epoch : 130 [4/36(11%)]\tLoss: 1.053113\n",
      "Epoch : 130 [8/36(22%)]\tLoss: 0.543912\n",
      "Epoch : 130 [12/36(33%)]\tLoss: 0.340454\n",
      "Epoch : 130 [16/36(44%)]\tLoss: 0.086426\n",
      "Epoch : 130 [20/36(56%)]\tLoss: 0.334075\n",
      "Epoch : 130 [24/36(67%)]\tLoss: 1.035887\n",
      "Epoch : 130 [28/36(78%)]\tLoss: 0.815006\n",
      "Epoch : 130 [32/36(89%)]\tLoss: 0.362916\n",
      "Epoch : 131 [0/36(0%)]\tLoss: 0.267535\n",
      "Epoch : 131 [4/36(11%)]\tLoss: 0.685511\n",
      "Epoch : 131 [8/36(22%)]\tLoss: 1.568968\n",
      "Epoch : 131 [12/36(33%)]\tLoss: 1.392822\n",
      "Epoch : 131 [16/36(44%)]\tLoss: 0.698108\n",
      "Epoch : 131 [20/36(56%)]\tLoss: 0.776229\n",
      "Epoch : 131 [24/36(67%)]\tLoss: 0.738864\n",
      "Epoch : 131 [28/36(78%)]\tLoss: 1.364831\n",
      "Epoch : 131 [32/36(89%)]\tLoss: 0.598560\n",
      "Epoch : 132 [0/36(0%)]\tLoss: 0.245823\n",
      "Epoch : 132 [4/36(11%)]\tLoss: 0.541241\n",
      "Epoch : 132 [8/36(22%)]\tLoss: 1.268045\n",
      "Epoch : 132 [12/36(33%)]\tLoss: 1.452606\n",
      "Epoch : 132 [16/36(44%)]\tLoss: 0.255852\n",
      "Epoch : 132 [20/36(56%)]\tLoss: 0.110474\n",
      "Epoch : 132 [24/36(67%)]\tLoss: 0.283971\n",
      "Epoch : 132 [28/36(78%)]\tLoss: 0.393676\n",
      "Epoch : 132 [32/36(89%)]\tLoss: 0.726203\n",
      "Epoch : 133 [0/36(0%)]\tLoss: 0.732823\n",
      "Epoch : 133 [4/36(11%)]\tLoss: 0.341891\n",
      "Epoch : 133 [8/36(22%)]\tLoss: 0.143929\n",
      "Epoch : 133 [12/36(33%)]\tLoss: 0.971662\n",
      "Epoch : 133 [16/36(44%)]\tLoss: 0.703579\n",
      "Epoch : 133 [20/36(56%)]\tLoss: 0.114067\n",
      "Epoch : 133 [24/36(67%)]\tLoss: 0.067433\n",
      "Epoch : 133 [28/36(78%)]\tLoss: 0.316554\n",
      "Epoch : 133 [32/36(89%)]\tLoss: 0.215482\n",
      "Epoch : 134 [0/36(0%)]\tLoss: 0.400922\n",
      "Epoch : 134 [4/36(11%)]\tLoss: 0.271390\n",
      "Epoch : 134 [8/36(22%)]\tLoss: 0.517674\n",
      "Epoch : 134 [12/36(33%)]\tLoss: 0.082328\n",
      "Epoch : 134 [16/36(44%)]\tLoss: 0.578447\n",
      "Epoch : 134 [20/36(56%)]\tLoss: 0.156328\n",
      "Epoch : 134 [24/36(67%)]\tLoss: 0.823687\n",
      "Epoch : 134 [28/36(78%)]\tLoss: 1.452585\n",
      "Epoch : 134 [32/36(89%)]\tLoss: 0.254751\n",
      "Epoch : 135 [0/36(0%)]\tLoss: 0.096825\n",
      "Epoch : 135 [4/36(11%)]\tLoss: 0.577743\n",
      "Epoch : 135 [8/36(22%)]\tLoss: 1.216215\n",
      "Epoch : 135 [12/36(33%)]\tLoss: 0.266566\n",
      "Epoch : 135 [16/36(44%)]\tLoss: 0.866197\n",
      "Epoch : 135 [20/36(56%)]\tLoss: 0.116730\n",
      "Epoch : 135 [24/36(67%)]\tLoss: 0.214404\n",
      "Epoch : 135 [28/36(78%)]\tLoss: 0.908065\n",
      "Epoch : 135 [32/36(89%)]\tLoss: 0.497162\n",
      "Epoch : 136 [0/36(0%)]\tLoss: 0.791361\n",
      "Epoch : 136 [4/36(11%)]\tLoss: 0.070404\n",
      "Epoch : 136 [8/36(22%)]\tLoss: 0.332302\n",
      "Epoch : 136 [12/36(33%)]\tLoss: 0.530394\n",
      "Epoch : 136 [16/36(44%)]\tLoss: 1.177648\n",
      "Epoch : 136 [20/36(56%)]\tLoss: 0.361134\n",
      "Epoch : 136 [24/36(67%)]\tLoss: 0.029007\n",
      "Epoch : 136 [28/36(78%)]\tLoss: 0.378445\n",
      "Epoch : 136 [32/36(89%)]\tLoss: 0.352557\n",
      "Epoch : 137 [0/36(0%)]\tLoss: 0.514302\n",
      "Epoch : 137 [4/36(11%)]\tLoss: 0.514673\n",
      "Epoch : 137 [8/36(22%)]\tLoss: 0.078363\n",
      "Epoch : 137 [12/36(33%)]\tLoss: 0.248395\n",
      "Epoch : 137 [16/36(44%)]\tLoss: 0.690084\n",
      "Epoch : 137 [20/36(56%)]\tLoss: 0.556150\n",
      "Epoch : 137 [24/36(67%)]\tLoss: 0.692066\n",
      "Epoch : 137 [28/36(78%)]\tLoss: 0.395910\n",
      "Epoch : 137 [32/36(89%)]\tLoss: 0.067013\n",
      "Epoch : 138 [0/36(0%)]\tLoss: 0.321294\n",
      "Epoch : 138 [4/36(11%)]\tLoss: 1.200751\n",
      "Epoch : 138 [8/36(22%)]\tLoss: 0.760579\n",
      "Epoch : 138 [12/36(33%)]\tLoss: 0.454217\n",
      "Epoch : 138 [16/36(44%)]\tLoss: 0.490970\n",
      "Epoch : 138 [20/36(56%)]\tLoss: 0.479115\n",
      "Epoch : 138 [24/36(67%)]\tLoss: 1.204567\n",
      "Epoch : 138 [28/36(78%)]\tLoss: 1.369840\n",
      "Epoch : 138 [32/36(89%)]\tLoss: 0.630212\n",
      "Epoch : 139 [0/36(0%)]\tLoss: 0.209625\n",
      "Epoch : 139 [4/36(11%)]\tLoss: 1.160850\n",
      "Epoch : 139 [8/36(22%)]\tLoss: 1.128245\n",
      "Epoch : 139 [12/36(33%)]\tLoss: 0.566942\n",
      "Epoch : 139 [16/36(44%)]\tLoss: 0.885552\n",
      "Epoch : 139 [20/36(56%)]\tLoss: 0.500880\n",
      "Epoch : 139 [24/36(67%)]\tLoss: 0.443619\n",
      "Epoch : 139 [28/36(78%)]\tLoss: 1.193957\n",
      "Epoch : 139 [32/36(89%)]\tLoss: 1.347992\n",
      "Epoch : 140 [0/36(0%)]\tLoss: 0.146880\n",
      "Epoch : 140 [4/36(11%)]\tLoss: 0.348057\n",
      "Epoch : 140 [8/36(22%)]\tLoss: 0.522477\n",
      "Epoch : 140 [12/36(33%)]\tLoss: 0.750798\n",
      "Epoch : 140 [16/36(44%)]\tLoss: 0.270967\n",
      "Epoch : 140 [20/36(56%)]\tLoss: 0.239108\n",
      "Epoch : 140 [24/36(67%)]\tLoss: 0.286817\n",
      "Epoch : 140 [28/36(78%)]\tLoss: 0.199484\n",
      "Epoch : 140 [32/36(89%)]\tLoss: 0.407403\n",
      "Epoch : 141 [0/36(0%)]\tLoss: 0.119836\n",
      "Epoch : 141 [4/36(11%)]\tLoss: 0.653336\n",
      "Epoch : 141 [8/36(22%)]\tLoss: 0.602166\n",
      "Epoch : 141 [12/36(33%)]\tLoss: 0.043642\n",
      "Epoch : 141 [16/36(44%)]\tLoss: 0.449214\n",
      "Epoch : 141 [20/36(56%)]\tLoss: 0.047876\n",
      "Epoch : 141 [24/36(67%)]\tLoss: 0.246095\n",
      "Epoch : 141 [28/36(78%)]\tLoss: 0.635551\n",
      "Epoch : 141 [32/36(89%)]\tLoss: 0.452014\n",
      "Epoch : 142 [0/36(0%)]\tLoss: 0.281104\n",
      "Epoch : 142 [4/36(11%)]\tLoss: 0.052166\n",
      "Epoch : 142 [8/36(22%)]\tLoss: 0.722157\n",
      "Epoch : 142 [12/36(33%)]\tLoss: 0.463506\n",
      "Epoch : 142 [16/36(44%)]\tLoss: 0.912861\n",
      "Epoch : 142 [20/36(56%)]\tLoss: 0.157428\n",
      "Epoch : 142 [24/36(67%)]\tLoss: 0.461840\n",
      "Epoch : 142 [28/36(78%)]\tLoss: 0.538407\n",
      "Epoch : 142 [32/36(89%)]\tLoss: 1.155769\n",
      "Epoch : 143 [0/36(0%)]\tLoss: 1.060187\n",
      "Epoch : 143 [4/36(11%)]\tLoss: 0.190263\n",
      "Epoch : 143 [8/36(22%)]\tLoss: 0.251320\n",
      "Epoch : 143 [12/36(33%)]\tLoss: 0.135756\n",
      "Epoch : 143 [16/36(44%)]\tLoss: 0.769416\n",
      "Epoch : 143 [20/36(56%)]\tLoss: 0.557293\n",
      "Epoch : 143 [24/36(67%)]\tLoss: 0.234772\n",
      "Epoch : 143 [28/36(78%)]\tLoss: 0.224294\n",
      "Epoch : 143 [32/36(89%)]\tLoss: 0.162216\n",
      "Epoch : 144 [0/36(0%)]\tLoss: 0.885804\n",
      "Epoch : 144 [4/36(11%)]\tLoss: 1.064728\n",
      "Epoch : 144 [8/36(22%)]\tLoss: 0.577098\n",
      "Epoch : 144 [12/36(33%)]\tLoss: 0.186192\n",
      "Epoch : 144 [16/36(44%)]\tLoss: 0.295856\n",
      "Epoch : 144 [20/36(56%)]\tLoss: 1.072749\n",
      "Epoch : 144 [24/36(67%)]\tLoss: 0.777039\n",
      "Epoch : 144 [28/36(78%)]\tLoss: 0.861959\n",
      "Epoch : 144 [32/36(89%)]\tLoss: 0.847830\n",
      "Epoch : 145 [0/36(0%)]\tLoss: 0.100434\n",
      "Epoch : 145 [4/36(11%)]\tLoss: 0.881786\n",
      "Epoch : 145 [8/36(22%)]\tLoss: 1.652492\n",
      "Epoch : 145 [12/36(33%)]\tLoss: 1.100422\n",
      "Epoch : 145 [16/36(44%)]\tLoss: 0.500160\n",
      "Epoch : 145 [20/36(56%)]\tLoss: 0.017991\n",
      "Epoch : 145 [24/36(67%)]\tLoss: 0.924879\n",
      "Epoch : 145 [28/36(78%)]\tLoss: 1.047480\n",
      "Epoch : 145 [32/36(89%)]\tLoss: 1.900838\n",
      "Epoch : 146 [0/36(0%)]\tLoss: 0.604198\n",
      "Epoch : 146 [4/36(11%)]\tLoss: 0.511137\n",
      "Epoch : 146 [8/36(22%)]\tLoss: 0.564407\n",
      "Epoch : 146 [12/36(33%)]\tLoss: 2.628869\n",
      "Epoch : 146 [16/36(44%)]\tLoss: 0.726824\n",
      "Epoch : 146 [20/36(56%)]\tLoss: 0.216232\n",
      "Epoch : 146 [24/36(67%)]\tLoss: 0.148627\n",
      "Epoch : 146 [28/36(78%)]\tLoss: 1.096104\n",
      "Epoch : 146 [32/36(89%)]\tLoss: 2.876029\n",
      "Epoch : 147 [0/36(0%)]\tLoss: 0.751905\n",
      "Epoch : 147 [4/36(11%)]\tLoss: 0.070039\n",
      "Epoch : 147 [8/36(22%)]\tLoss: 0.395695\n",
      "Epoch : 147 [12/36(33%)]\tLoss: 0.712969\n",
      "Epoch : 147 [16/36(44%)]\tLoss: 1.569826\n",
      "Epoch : 147 [20/36(56%)]\tLoss: 0.929488\n",
      "Epoch : 147 [24/36(67%)]\tLoss: 1.048727\n",
      "Epoch : 147 [28/36(78%)]\tLoss: 0.314118\n",
      "Epoch : 147 [32/36(89%)]\tLoss: 1.024843\n",
      "Epoch : 148 [0/36(0%)]\tLoss: 0.580811\n",
      "Epoch : 148 [4/36(11%)]\tLoss: 0.778647\n",
      "Epoch : 148 [8/36(22%)]\tLoss: 0.985396\n",
      "Epoch : 148 [12/36(33%)]\tLoss: 0.664756\n",
      "Epoch : 148 [16/36(44%)]\tLoss: 0.741158\n",
      "Epoch : 148 [20/36(56%)]\tLoss: 0.389928\n",
      "Epoch : 148 [24/36(67%)]\tLoss: 1.165801\n",
      "Epoch : 148 [28/36(78%)]\tLoss: 2.472889\n",
      "Epoch : 148 [32/36(89%)]\tLoss: 0.595185\n",
      "Epoch : 149 [0/36(0%)]\tLoss: 0.162939\n",
      "Epoch : 149 [4/36(11%)]\tLoss: 1.400282\n",
      "Epoch : 149 [8/36(22%)]\tLoss: 3.115043\n",
      "Epoch : 149 [12/36(33%)]\tLoss: 1.705444\n",
      "Epoch : 149 [16/36(44%)]\tLoss: 0.784464\n",
      "Epoch : 149 [20/36(56%)]\tLoss: 0.263937\n",
      "Epoch : 149 [24/36(67%)]\tLoss: 2.393395\n",
      "Epoch : 149 [28/36(78%)]\tLoss: 5.546067\n",
      "Epoch : 149 [32/36(89%)]\tLoss: 1.594911\n",
      "--- 0.09205681482950846 minutes ---\n"
     ]
    }
   ],
   "source": [
    "# calculate the time for the code execution\n",
    "start_time = tt.time()\n",
    "\n",
    "# switch model into training mode\n",
    "model.train()\n",
    "\n",
    "hist_train = []\n",
    "hist_valid = []\n",
    "for epoch in range(config.epoch):\n",
    "    # training loop\n",
    "    # switch model into train mode\n",
    "    model.train()\n",
    "    hist_train_step = 0\n",
    "    for batch_idx, (X_batch, y_batch) in enumerate(train_loader):\n",
    "        var_X_batch = Variable(X_batch).to(device)\n",
    "        var_y_batch = Variable(y_batch).to(device)\n",
    "        optimizer.zero_grad()\n",
    "        # note: decoder input is the last instance of encoder input\n",
    "        output = model(var_X_batch)\n",
    "        loss = criterion(output[:,-1,:].squeeze(), var_y_batch) # we only need the last instance from output sequence\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        wandb.log({'train_loss': loss.item()})\n",
    "        print(f'Epoch : {epoch} [{batch_idx*len(X_batch)}/{len(train_loader.dataset)}'\n",
    "              f'({100.* batch_idx / len(train_loader):.0f}%)]\\tLoss: {loss.item():.6f}')\n",
    "        hist_train_step += loss.item()\n",
    "\n",
    "    hist_train.append(hist_train_step / len(train_loader.dataset))\n",
    "\n",
    "    # cross-validation loop\n",
    "    # switch model into evaluation mode\n",
    "    model.eval()\n",
    "    hist_valid_step = 0\n",
    "\n",
    "    for batch_idx, (X_batch, y_batch) in enumerate(valid_loader):\n",
    "        var_X_batch = Variable(X_batch).to(device)\n",
    "        var_y_batch = Variable(y_batch).to(device)\n",
    "        optimizer.zero_grad()\n",
    "        with torch.no_grad():\n",
    "            output = model(var_X_batch)\n",
    "            loss = criterion(output[:,-1,:].squeeze(), var_y_batch)\n",
    "        wandb.log({'validation_loss': loss.item()})\n",
    "        hist_valid_step += loss.item()\n",
    "\n",
    "    hist_valid.append(hist_valid_step / len(valid_loader.dataset))\n",
    "\n",
    "print (f\"--- {(tt.time() - start_time)/60} minutes ---\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now let's check the training loss and validation loss."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1200x500 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "fig = plt.figure(figsize=(12, 5))\n",
    "fig.add_subplot(1, 2, 1)\n",
    "plt.plot(np.asarray(hist_train), 'b', label=\"Training loss\")\n",
    "plt.plot(np.asarray(hist_valid), 'r', label=\"Validation loss\")\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Loss')\n",
    "plt.title(\"MSE\")\n",
    "plt.legend()\n",
    "\n",
    "fig.add_subplot(1, 2, 2)\n",
    "plt.semilogy(np.asarray(hist_train), 'b', label=\"Training loss\")\n",
    "plt.semilogy(np.asarray(hist_valid), 'r', label=\"Validation loss\")\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Logarithmic loss')\n",
    "plt.title(\"Logarithmic MSE\")\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# save the checkpoint model training if necessary\n",
    "output_path = \"./\"\n",
    "\n",
    "torch.save({\n",
    "            'epoch': epoch,\n",
    "            'model_state_dict': model.state_dict(),\n",
    "            'optimizer_state_dict': optimizer.state_dict(),\n",
    "            'loss': loss.item()\n",
    "            }, Path(output_path,'lstm_train_checkpoint.pt'))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Evaluate model\n",
    "Now we can evaluate our model with testing set and compare the predictions with the ground truth."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# switch model into evaluation mode\n",
    "model.eval()\n",
    "hist_test = []\n",
    "predictions = []\n",
    "hist_test_step = 0\n",
    "for batch_idx, (X_batch, y_batch) in enumerate(test_loader):\n",
    "    var_X_batch = Variable(X_batch).to(device)\n",
    "    var_y_batch = Variable(y_batch).to(device)\n",
    "    optimizer.zero_grad()\n",
    "    with torch.no_grad():\n",
    "        output = model(var_X_batch)\n",
    "        loss = criterion(output[:,-1,:].squeeze(), var_y_batch)\n",
    "    wandb.log({'testing_loss': loss.item()})\n",
    "    predictions.append(output.squeeze().cpu().detach().numpy()[:,-1])\n",
    "    hist_test_step += loss.item()\n",
    "\n",
    "hist_test.append(hist_test_step / len(test_loader.dataset))\n",
    "# call wandb finish to stop logging\n",
    "wandb.finish()"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot the predictions versus ground truth."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The MSE loss is 0.366\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAj8AAAGwCAYAAABGogSnAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8g+/7EAAAACXBIWXMAAA9hAAAPYQGoP6dpAABQG0lEQVR4nO3deVxU5f4H8M8wyj4Mos4MKCLhSrhvIaV0I0RNpbqapSnpTTMwyUy0X6amhZqWrVh5A82svNcwtMIQhXIBFyIljJRwucpSLjOKgjpzfn+MjI4sgs5wZjif9+s1rzrPeeaZ74zCfH1WmSAIAoiIiIgkwkHsAIiIiIgaE5MfIiIikhQmP0RERCQpTH6IiIhIUpj8EBERkaQw+SEiIiJJYfJDREREktJM7ABskcFgwOnTp6FQKCCTycQOh4iIiOpBEARcuHABPj4+cHCovX+HyU8NTp8+DV9fX7HDICIiojtw8uRJtG3bttb7TH5qoFAoABg/PA8PD5GjISIiovrQ6XTw9fU1fY/XhslPDaqGujw8PJj8EBER2ZnbTVnhhGciIiKSFCY/REREJClMfoiIiEhSOOeHiMgO6PV6XL16VewwiETVvHlzyOXyu26HyQ8RkQ0TBAElJSU4f/682KEQ2QRPT09oNJq72oePyQ8RkQ2rSnxUKhVcXV258SpJliAIuHTpEsrKygAA3t7ed9wWkx8iIhul1+tNiU/Lli3FDodIdC4uLgCAsrIyqFSqOx4C44RnIiIbVTXHx9XVVeRIiGxH1c/D3cyBY/JDRGTjONRFdIMlfh447NVYDHrg+G7gYingrgb8BgIOdz9jnYiIiBpG1J6f+Ph49OvXDwqFAiqVCpGRkSgoKDCrExoaCplMZvZ47rnn6mxXEAS89tpr8Pb2houLC8LCwnDkyBFrvpW65acAK4OANY8AGycb/7syyFhOREREjUrU5CczMxPR0dHIyspCWloarl69ivDwcJSXl5vVe/bZZ1FcXGx6LFu2rM52ly1bhvfeew+rVq1CdnY23NzcMGTIEFRUVFjz7dQsPwXYMAHQnTYv1xUby5kAERHdsaioKERGRpquQ0NDERsbe1dtWqINsm2iDnulpqaaXSclJUGlUuHAgQMYNGiQqdzV1RUajaZebQqCgJUrV+LVV1/FqFGjAABr166FWq3Gpk2bMHbsWMu9gdsx6IHUOABCTZECkAGpc4AuwzkERkRWozcI2Ft0FmUXKqBSOKO/vxfkDtadRxQVFYU1a9YAMG5M165dO0yYMAGvvPIKmjWz3lfPN998g+bNm9erbkZGBh588EGcO3cOnp6ed9QG2SebmvOj1WoBAF5eXmblX3zxBdatWweNRoMRI0Zg3rx5ta5+KCoqQklJCcLCwkxlSqUSAwYMwJ49e2pMfiorK1FZWWm61ul0lng7xjk+t/b4mBEA3SljPf8HLPOaREQ3Sc0rxsLN+SjW3uj59lY6Y/6IQEQE3fk+KfURERGBxMREVFZW4vvvv0d0dDSaN2+OuXPnmtW7cuUKHB0dLfKat35/iNUG2TabWe1lMBgQGxuLkJAQBAUFmcqfeuoprFu3Djt27MDcuXPx+eefY/z48bW2U1JSAgBQq9Vm5Wq12nTvVvHx8VAqlaaHr6+vBd4RjJObLVmPiKgBUvOKMW1djlniAwAl2gpMW5eD1Lxiq76+k5MTNBoN/Pz8MG3aNISFhSElJcU0VPXGG2/Ax8cHnTt3BgCcPHkSY8aMgaenJ7y8vDBq1CgcO3bM1J5er8fMmTPh6emJli1bYvbs2RAE8571W4esKisrERcXB19fXzg5OaFDhw7497//jWPHjuHBBx8EALRo0QIymQxRUVE1tnHu3DlMmDABLVq0gKurK4YOHWo2jzQpKQmenp7YunUrunbtCnd3d0RERKC4+Mbnm5GRgf79+8PNzQ2enp4ICQnB8ePHLfRJU0PZTPITHR2NvLw8fPXVV2blU6ZMwZAhQ9CtWzeMGzcOa9euRXJyMgoLCy322nPnzoVWqzU9Tp48aZmG3dW3r9OQekRE9aQ3CFi4Ob/WQXcAWLg5H3pDTTWsw8XFBVeuXAEApKeno6CgAGlpadiyZQuuXr2KIUOGQKFQ4Oeff8auXbtMSUTVc1asWIGkpCR89tln2LlzJ86ePYvk5OQ6X3PChAn48ssv8d577+Hw4cP4+OOP4e7uDl9fX2zcuBEAUFBQgOLiYrz77rs1thEVFYX9+/cjJSUFe/bsgSAIGDZsmNk+M5cuXcLy5cvx+eef46effsKJEycwa9YsAMC1a9cQGRmJwYMH4+DBg9izZw+mTJnCLQxEZBPDXjExMdiyZQt++ukntG3bts66AwYMAAAcPXoUAQEB1e5XzQ0qLS012/q6tLQUPXv2rLFNJycnODk53WH0dfAbCHj4GCc31/grSGa87zfQ8q9NRJK2t+hstR6fmwkAirUV2Ft0FsEB1t09WhAEpKenY+vWrZg+fTr++usvuLm5YfXq1abhrnXr1sFgMGD16tWmpCAxMRGenp7IyMhAeHg4Vq5ciblz5+Kxxx4DAKxatQpbt26t9XX/+OMPbNiwAWlpaaapEPfcc4/pftXwlkqlMpvzc7MjR44gJSUFu3btwsCBxt/VX3zxBXx9fbFp0yaMHj0agHHDvVWrVpm+l2JiYvD6668DME6l0Gq1eOSRR0z3u3bt2vAPkixG1J4fQRAQExOD5ORkbN++Hf7+/rd9Tm5uLoDaz/Tw9/eHRqNBenq6qUyn0yE7OxvBwcEWibveHORAxNLrF7dm+NevI5ZwsjMRWVzZhfqtbq1vvTuxZcsWuLu7w9nZGUOHDsUTTzyBBQsWAAC6detmNs/n119/xdGjR6FQKODu7g53d3d4eXmhoqIChYWF0Gq1KC4uNv0DGACaNWuGvn371vr6ubm5kMvlGDx48B2/h8OHD6NZs2Zmr9uyZUt07twZhw8fNpW5urqa/YPc29vbdAaVl5cXoqKiMGTIEIwYMQLvvvuu2ZAYNT5Rk5/o6GisW7cO69evh0KhQElJCUpKSnD58mUAQGFhIRYtWoQDBw7g2LFjSElJwYQJEzBo0CB0797d1E6XLl1MXZ8ymQyxsbFYvHgxUlJScOjQIUyYMAE+Pj5myyEbTeBIYMxawOOWZM3Dx1geOLLxYyKiJk+lcLZovTvx4IMPIjc3F0eOHMHly5exZs0auLm5AYDpv1UuXryIPn36IDc31+zxxx9/4Kmnnrqj1686B6ox3Lo6TCaTmc1HSkxMxJ49ezBw4EB8/fXX6NSpE7KyshotPjIn6rBXQkICAOPkspslJiYiKioKjo6O2LZtG1auXIny8nL4+vri8ccfx6uvvmpWv6CgwLRSDABmz56N8vJyTJkyBefPn8f999+P1NRUODtb74e8ToEjjcvZucMzETWS/v5e8FY6o0RbUdugOzRK47J3a3Fzc0OHDh3qVbd37974+uuvoVKp4OHhUWMdb29vZGdnm7ZCuXbtGg4cOIDevXvXWL9bt24wGAzIzMw0WwFcparnSa/X1xpX165dce3aNWRnZ5uGvc6cOYOCggIEBgbW671V6dWrF3r16oW5c+ciODgY69evx3333degNsgyRE1+bp2lfytfX19kZmY2uB2ZTIbXX3/dNN5qExzkXM5ORI1G7iDD/BGBmLYuBzKYzzqsGoSfPyLQ6vv91Ne4cePw1ltvYdSoUXj99dfRtm1bHD9+HN988w1mz56Ntm3bYsaMGViyZAk6duyILl264O2338b58+drbbN9+/aYOHEiJk2ahPfeew89evTA8ePHUVZWhjFjxsDPzw8ymQxbtmzBsGHD4OLiAnd3d7M2OnbsiFGjRuHZZ5/Fxx9/DIVCgTlz5qBNmzamveRup6ioCJ988glGjhwJHx8fFBQU4MiRI5gwYcLdfGR0F2xmtRcREVlWRJA3Esb3hkZp3uutUTojYXxvq+/z0xCurq746aef0K5dOzz22GPo2rUrJk+ejIqKClNP0EsvvYSnn34aEydORHBwMBQKBR599NE6201ISMA///lPPP/88+jSpQueffZZ0ykCbdq0wcKFCzFnzhyo1WrExMTU2EZiYiL69OmDRx55BMHBwRAEAd9//329N0J0dXXF77//jscffxydOnXClClTEB0djalTpzbgEyJLkgm3636RIJ1OB6VSCa1WW2v3KxGRtVVUVKCoqAj+/v53NWwvxg7PRNZS189Ffb+/bWKpOxERWY/cQWb15exE9oTDXkRERCQpTH6IiIhIUpj8EBERkaQw+SEiIiJJYfJDREREksLkh4iIiCSFyQ8RERFJCpMfIiKi6xYsWICePXuKHQZCQ0MRGxsrdhhNFpMfIiKyuJKSEsyYMQMdOnSAs7Mz1Go1QkJCkJCQgEuXLokd3h3LyMiATCar80wxMduj+uEOz0RETZ1BDxzfDVwsBdzVgN9A42HLVvLnn38iJCQEnp6eePPNN9GtWzc4OTnh0KFD+OSTT9CmTRuMHDmyxudevXq13mdm2bIrV66YTo0n28OeHyKipiw/BVgZBKx5BNg42fjflUHGcit5/vnn0axZM+zfvx9jxoxB165dcc8992DUqFH47rvvMGLECFNdmUyGhIQEjBw5Em5ubnjjjTcAGA8kDQgIgKOjIzp37ozPP//c9Jxjx45BJpMhNzfXVHb+/HnIZDJkZGQAuNGjkp6ejr59+8LV1RUDBw5EQUGBWaxLliyBWq2GQqEwHaRam2PHjuHBBx8EALRo0QIymQxRUVEAjMNUMTExiI2NRatWrTBkyJDbxllXewBgMBgwe/ZseHl5QaPRYMGCBfX9I6DbYPJDRNRU5acAGyYAutPm5bpiY7kVEqAzZ87gxx9/RHR0NNzc3GqsI5OZH6q6YMECPProozh06BAmTZqE5ORkzJgxAy+99BLy8vIwdepUPPPMM9ixY0eD4/m///s/rFixAvv370ezZs0wadIk070NGzZgwYIFePPNN7F//354e3vjo48+qrUtX19fbNy4EQBQUFCA4uJivPvuu6b7a9asgaOjI3bt2oVVq1bdNrb6tOfm5obs7GwsW7YMr7/+OtLS0hr8GVB1HPYiImqKDHogNQ6AUMNNAYAMSJ0DdBlu0SGwo0ePQhAEdO7c2ay8VatWpl6V6OhoLF261HTvqaeewjPPPGO6fvLJJxEVFYXnn38eADBz5kxkZWVh+fLlpp6S+nrjjTcwePBgAMCcOXMwfPhwVFRUwNnZGStXrsTkyZMxefJkAMDixYuxbdu2Wnt/5HI5vLy8AAAqlQqenp5m9zt27Ihly5aZro8dO1ZnbLdrr3v37pg/f76p7Q8++ADp6el4+OGH6/XeqXbs+SEiaoqO767e42NGAHSnjPUawd69e5Gbm4t7770XlZWVZvf69u1rdn348GGEhISYlYWEhODw4cMNft3u3bub/t/b2xsAUFZWZnqdAQMGmNUPDg5u8GtU6dOnzx0/tyY3xw4Y46+Kne4Oe36IiJqii6WWrVdPHTp0gEwmqza35p577gEAuLi4VHtObcNjtXFwMP67XRBu9GpdvXq1xro3T56uGm4zGAwNer36uvV9NCTOmtw68Vsmk1ktdqlhzw8RUVPkrrZsvXpq2bIlHn74YXzwwQcoLy+/oza6du2KXbt2mZXt2rULgYGBAIDWrVsDAIqLi033b55U3JDXyc7ONivLysqq8zlVK7j0ev1t269PnA1pjyyHPT9ERE2R30DAw8c4ubnGeT8y432/gRZ/6Y8++gghISHo27cvFixYgO7du8PBwQH79u3D77//ftvhoZdffhljxoxBr169EBYWhs2bN+Obb77Btm3bABh7j+677z4sWbIE/v7+KCsrw6uvvtrgOGfMmIGoqCj07dsXISEh+OKLL/Dbb7+Zeqlq4ufnB5lMhi1btmDYsGFwcXGBu7t7jXXrE2dD2iPLYc8PEVFT5CAHIqomFctuuXn9OmKJVfb7CQgIwC+//IKwsDDMnTsXPXr0QN++ffH+++9j1qxZWLRoUZ3Pj4yMxLvvvovly5fj3nvvxccff4zExESEhoaa6nz22We4du0a+vTpg9jYWCxevLjBcT7xxBOYN28eZs+ejT59+uD48eOYNm1anc9p06YNFi5ciDlz5kCtViMmJqbO+reLs6HtkWXIhJsHIwkAoNPpoFQqodVq4eHhIXY4RCRRFRUVKCoqgr+/P5ydne+skfwU46qvmyc/e7QxJj6BNW80SGTL6vq5qO/3N4e9iIiassCRxuXsjbjDM5GtY/JDRNTUOcgB/wfEjoLIZnDODxEREUkKkx8iIiKSFCY/REREJClMfoiIiEhSmPwQERGRpDD5ISIiIklh8kNERESSwuSHiIioFiUlJXj44Yfh5uYGT09PscOxmIyMDMhkMpw/f17UOBYsWICePXs2+usy+SEiIouKioqCTCbDc889V+1edHQ0ZDIZoqKiGj+wO/DOO++guLgYubm5+OOPP8QOp1HJZDJs2rTJZtu7G0x+iIiaOL1Bj30l+/D9n99jX8k+6A16q7+mr68vvvrqK1y+fNlUVlFRgfXr16Ndu3ZWf31LKSwsRJ8+fdCxY0eoVCqLtXvlyhWLtSUme30fTH6IiJqwbce3YcjGIZi0dRLifo7DpK2TMGTjEGw7vs2qr9u7d2/4+vrim2++MZV98803aNeuHXr16mVW12AwID4+Hv7+/nBxcUGPHj3w3//+13Rfr9dj8uTJpvudO3fGu+++a9ZGVFQUIiMjsXz5cnh7e6Nly5aIjo7G1atX64wzISEBAQEBcHR0ROfOnfH555+b7rVv3x4bN27E2rVr6+ytunbtGl544QV4enqiZcuWiIuLw8SJExEZGWmqExoaipiYGMTGxqJVq1YYMmQIACAzMxP9+/eHk5MTvL29MWfOHFy7ds0shpUrV5q9Xs+ePbFgwQLTtUwmw+rVq/Hoo4/C1dUVHTt2REpKitlzvv/+e3Tq1AkuLi548MEHcezYsTo/l/bt2wMAHn30UchkMtN11TDV6tWrzQ4WvV2ctbVX5fPPP0f79u2hVCoxduxYXLhwoc747haTHyKiJmrb8W2YmTETpZdKzcrLLpVhZsZMqydAkyZNQmJioun6s88+wzPPPFOtXnx8PNauXYtVq1bht99+w4svvojx48cjMzMTgDE5atu2Lf7zn/8gPz8fr732Gl555RVs2LDBrJ0dO3agsLAQO3bswJo1a5CUlISkpKRa40tOTsaMGTPw0ksvIS8vD1OnTsUzzzyDHTt2AAD27duHiIgIjBkzBsXFxdUSripLly7FF198gcTEROzatQs6na7G4Z01a9bA0dERu3btwqpVq3Dq1CkMGzYM/fr1w6+//oqEhAT8+9//xuLFi2/30VazcOFCjBkzBgcPHsSwYcMwbtw4nD17FgBw8uRJPPbYYxgxYgRyc3Pxr3/9C3PmzKmzvX379gEAEhMTUVxcbLoGgKNHj2Ljxo345ptvkJubW6/46mqvsLAQmzZtwpYtW7BlyxZkZmZiyZIlDXn7DSdQNVqtVgAgaLVasUMhIgm7fPmykJ+fL1y+fLnBz72mvyY8tOEhISgpqMZHt6RuQtiGMOGa/prF4544caIwatQooaysTHBychKOHTsmHDt2THB2dhb++usvYdSoUcLEiRMFQRCEiooKwdXVVdi9e7dZG5MnTxaefPLJWl8jOjpaePzxx81e08/PT7h27cb7GT16tPDEE0/U2sbAgQOFZ5991qxs9OjRwrBhw0zXN8daG7VaLbz11lum62vXrgnt2rUTRo0aZSobPHiw0KtXL7PnvfLKK0Lnzp0Fg8FgKvvwww8Fd3d3Qa/XC4IgCH5+fsI777xj9rwePXoI8+fPN10DEF599VXT9cWLFwUAwg8//CAIgiDMnTtXCAwMNGsjLi5OACCcO3eu1vcFQEhOTjYrmz9/vtC8eXOhrKzMrLy+cdbUnqurq6DT6UxlL7/8sjBgwIBa46rr56K+39881Z2IqAnKKcup1uNzMwECSi6VIKcsB/00/awSQ+vWrTF8+HAkJSVBEAQMHz4crVq1Mqtz9OhRXLp0CQ8//LBZ+ZUrV8yGxz788EN89tlnOHHiBC5fvowrV65UWyV07733Qi6Xm669vb1x6NChWuM7fPgwpkyZYlYWEhJSaw9PTbRaLUpLS9G/f39TmVwuR58+fWAwGMzq9unTp9rrBwcHQyaTmb3+xYsX8b///a9Bc6O6d+9u+n83Nzd4eHigrKzM9DoDBgwwqx8cHFzvtm/l5+eH1q1b3/Hzb9W+fXsoFArTtbe3tyl2a2HyQ0TUBP116S+L1rtTkyZNQkxMDABjAnOrixcvAgC+++47tGnTxuyek5MTAOCrr77CrFmzsGLFCgQHB0OhUOCtt95Cdna2Wf3mzZubXctksmoJiJjc3Nwa/BwHBwcYO01uqGkeU2O+95reR33jrIkYf26c80NE1AS1dq3fv8zrW+9ORURE4MqVK7h69appku/NAgMD4eTkhBMnTqBDhw5mD19fXwDArl27MHDgQDz//PPo1asXOnTogMLCwruOrWvXrti1a5dZ2a5duxAYGFjvNpRKJdRqtdkcFr1ej5ycnHq9/p49e8yShl27dkGhUKBt27YAjL1nxcXFpvs6nQ5FRUX1jq/qdfbu3WtWlpWVddvnNW/eHHp9/VYG1ifOhrRnbUx+6AaDHij6GTj0X+N/G2E5LBFZR29Vb6hd1ZBBVuN9GWTQuGrQW9XbqnHI5XIcPnwY+fn5ZkNSVRQKBWbNmoUXX3wRa9asQWFhIXJycvD+++9jzZo1AICOHTti//792Lp1K/744w/MmzfPLNm4Uy+//DKSkpKQkJCAI0eO4O2338Y333yDWbNmNaid6dOnIz4+Ht9++y0KCgowY8YMnDt3zmw4qybPP/88Tp48ienTp+P333/Ht99+i/nz52PmzJlwcDB+Pf/jH//A559/jp9//hmHDh3CxIkTa/wc6/Lcc8/hyJEjePnll1FQUID169fXORG8Svv27ZGeno6SkhKcO3euzrr1ibMh7Vkbkx8yyk8BVgYBax4BNk42/ndlkLGciOyO3EGOOf2NK3puTYCqruP6x0Hu0LAv0jvh4eEBDw+PWu8vWrQI8+bNQ3x8PLp27YqIiAh899138Pf3BwBMnToVjz32GJ544gkMGDAAZ86cwfPPP3/XcUVGRuLdd9/F8uXLce+99+Ljjz9GYmIiQkNDG9ROXFwcnnzySUyYMAHBwcFwd3fHkCFDTMvAa9OmTRt8//332Lt3L3r06IHnnnsOkydPxquvvmqqM3fuXAwePBiPPPIIhg8fjsjISAQEBDQovnbt2mHjxo3YtGkTevTogVWrVuHNN9+87fNWrFiBtLQ0+Pr6Vtue4Fb1ibMh7VmbTLh1kI6g0+mgVCqh1Wrr/IFtMvJTgA0TANz6V+H6L8wxa4HAkY0dFZHkVVRUoKioyGw/lYbadnwbluxdYjb5WeOqQVz/OIT5hVkqVLqJwWBA165dMWbMGCxatEjscJqcun4u6vv9zQnPUmfQA6lxqJ744HqZDEidA3QZDjTCvxCJyLLC/MLwoO+DyCnLwV+X/kJr19borerdKD0+UnH8+HH8+OOPGDx4MCorK/HBBx+gqKgITz31lNihUS2Y/Ejd8d2A7nQdFQRAd8pYz/+BRguLiCxH7iC32nJ2Mq50SkpKwqxZsyAIAoKCgrBt2zZ07dpV7NCoFqLO+YmPj0e/fv2gUCigUqkQGRmJgoKCGusKgoChQ4fW62C0qkP1bn5ERERY4R00ARdr3wfkjuoREUmMr68vdu3aBa1WC51Oh927d2PQoEFih0V1EDX5yczMRHR0NLKyspCWloarV68iPDwc5eXl1equXLnytjPnbxYREYHi4mLT48svv7Rk6E2Hu9qy9YiIiGycqMNeqampZtdJSUlQqVQ4cOCAWdacm5uLFStWYP/+/fD29q5X205OTtBoNPWqW1lZicrKStO1Tqer1/OaBL+BgIcPoCtGzfN+ZMb7fgMbOzIiuo7rUohusMTPg00tdddqtQAALy8vU9mlS5fw1FNP4cMPP6x3MgMAGRkZUKlU6Ny5M6ZNm4YzZ87UWjc+Ph5KpdL0qNpYSxIc5EDE0usXt/asXb+OWMLJzkQiqNr59tKlSyJHQmQ7qn4ebt0ZuiFsZqm7wWDAyJEjcf78eezcudNUPnXqVOj1eqxevRqAcdvr5ORkREZG1trWV199BVdXV/j7+6OwsBCvvPIK3N3dsWfPnho3h6qp58fX11c6S90B43L31Djzyc8ebYyJD5e5E4mmuLgY58+fh0qlgqura4OG/4lsjiAAVy8B+quAvDnQ3BWo599pQRBw6dIllJWVwdPTs8aRILtb6h4dHY28vDyzxCclJQXbt2/HL7/80qC2xo4da/r/bt26oXv37ggICEBGRgYeeuihavWdnJxMZ8hIVuBI43L247uNk5vd1cahLvb4EImqqsfb2gc9Elnd1UvA5fOA4dqNModmgIunMQmqJ09PzwaNBNXEJpKfmJgYbNmyBT/99JPpPBMA2L59OwoLC+Hp6WlW//HHH8cDDzyAjIyMerV/zz33oFWrVjh69GiNyQ9d5yDncnYiGyOTyeDt7Q2VSlXvgyKJbM7R7cCPNe0pVzW9YinQ4R+3baZ58+YNPt6jJqImP4IgYPr06UhOTkZGRoZpK/Mqc+bMwb/+9S+zsm7duuGdd97BiBEj6v06//vf/3DmzJl6T5YmIrI1crncIr/0iRqdQQ/8+BJwsbY95WTAj7OAwEONNtog6oTn6OhorFu3DuvXr4dCoUBJSQlKSkpw+fJlAMbu3qCgILMHYDyn5OZEqUuXLkhOTgYAXLx4ES+//DKysrJw7NgxpKenY9SoUejQoUONJwoTERGRFTVkM91GImryk5CQAK1Wi9DQUHh7e5seX3/9dYPaKSgoMK0Uk8vlOHjwIEaOHIlOnTph8uTJ6NOnD37++WfO6yEiImpsNriZrujDXpZ4zs1lLi4u2Lp1613FRURERBZig5vp2tQ+P0RERNTEVG2mW20vuSoy49YqjbiZLpMfIiIish4b3EyXyQ8RERFZV+BIYMxawOOWVdcePsbyRt5M1yb2+SEiIqImzoY202XyQ0RERI3DRjbT5bAXERERSQqTHyIiIpIUJj9EREQkKZzzQ0R0pwx6m5i8SUQNw+SHiOhO5KcAqXHmZxZ5+Bj3M2nkZbtE1DAc9iIiaqj8FGDDhOqHNeqKjeX5KeLERUT1wuSHiKghDHpjjw9qOpvwelnqHGM9IrJJTH6IiBri+O7qPT5mBEB3yliPiGwSkx8iooa4WGrZekTU6Jj8EBE1hLvasvWIqNEx+SEiagi/gcZVXdVOp64iAzzaGOsRkU1i8kNE1BAOcuNydgDVE6Dr1xFLuN8PkQ1j8kNE1FCBI4ExawEPb/NyDx9jOff5IbJp3OSQiOhOBI4EugznDs9EdojJDxHRnXKQA/4PiB0FETUQh72IiIhIUpj8EBERkaQw+SEiIiJJ4ZwfIiIiahR6g4C9RWdRdqECKoUz+vt7Qe5Q255Z1sPkh4iIiKwuNa8YCzfno1hbYSrzVjpj/ohARAR51/FMy+OwFxEREVlVal4xpq3LMUt8AKBEW4Fp63KQmlfcqPEw+SEiIiKr0RsELNycD6GGe1VlCzfnQ2+oqYZ1MPkhIiIiq9lbdLZaj8/NBADF2grsLTrbaDEx+SEiIiKrKbtQe+JzJ/UsgckPERERWY1K4WzRepbA5IeIiIispr+/F7yVzqhtQbsMxlVf/f29Gi0mJj9ERERkNXIHGeaPCASAaglQ1fX8EYGNut8Pkx8iIiKyqoggbySM7w2N0nxoS6N0RsL43o2+zw83OSQiImoCbGX35NpEBHnj4UCNTcTI5IeIiMjO2dLuyXWRO8gQHNBS7DA47EVERGTPbG33ZHvA5IeIiMhO2eLuyfaAyQ8REZGdssXdk+0Bkx8iIiI7ZYu7J9sDTngmkiBbXxVCZGts9WfGFndPtgdMfogkxl5WhRDZClv+manaPblEW1HjvB8ZjHvpNObuyfaAw15EEsJVIUQNY+s/M7a4e7I9YPJDJBFcFULUMPbyM2NruyfbAw57EUlEQ1aF2MImZERis6efGVvaPdkeiNrzEx8fj379+kGhUEClUiEyMhIFBQU11hUEAUOHDoVMJsOmTZvqbFcQBLz22mvw9vaGi4sLwsLCcOTIESu8AyL7wVUhRA1jbz8zVbsnj+rZBsEBLZn41EHU5CczMxPR0dHIyspCWloarl69ivDwcJSXl1eru3LlSshk9fuDXLZsGd577z2sWrUK2dnZcHNzw5AhQ1BRYRt/QYnEwFUhRA3Dn5mmS9Rhr9TUVLPrpKQkqFQqHDhwAIMGDTKV5+bmYsWKFdi/fz+8veseuxQEAStXrsSrr76KUaNGAQDWrl0LtVqNTZs2YezYsdWeU1lZicrKStO1Tqe7m7dFZJO4KoSoYfgz03TZ1IRnrVYLAPDyuvEX6dKlS3jqqafw4YcfQqPR3LaNoqIilJSUICwszFSmVCoxYMAA7Nmzp8bnxMfHQ6lUmh6+vr53+U6IbA9XhRA1DH9mmi6bSX4MBgNiY2MREhKCoKAgU/mLL76IgQMHmnpxbqekpAQAoFarzcrVarXp3q3mzp0LrVZrepw8efIO3wWRbeOqEKKG4c9M02Qzq72io6ORl5eHnTt3mspSUlKwfft2/PLLL1Z9bScnJzg5OVn1NYhsBVeFEDUMf2aaHptIfmJiYrBlyxb89NNPaNu2ral8+/btKCwshKenp1n9xx9/HA888AAyMjKqtVU1NFZaWmo2P6i0tBQ9e/a0RvhEdqdqVQgR1Q9/ZpoWUYe9BEFATEwMkpOTsX37dvj7+5vdnzNnDg4ePIjc3FzTAwDeeecdJCYm1timv78/NBoN0tPTTWU6nQ7Z2dkIDg622nshIsvSGwTsKTyDb3NPYU/hGdE3kiOipkPUnp/o6GisX78e3377LRQKhWlOjlKphIuLCzQaTY2TnNu1a2eWKHXp0gXx8fF49NFHIZPJEBsbi8WLF6Njx47w9/fHvHnz4OPjg8jIyMZ6a0R0F2z5LCUisn+iJj8JCQkAgNDQULPyxMREREVF1budgoIC00oxAJg9ezbKy8sxZcoUnD9/Hvfffz9SU1Ph7My9GIhsXdVZSrf281SdpcRJpkR0t2SCILAv+RY6nQ5KpRJarRYeHh5ih0MkGXqDgPuXbq/1SIGqfVV2xv2Dk02JqJr6fn/bzFJ3IqKGnKVERHSnmPwQkc2wt7OUiMg+MfkhIpvBs5SIqDEw+SEim1F1llJts3lkMK764llKRHQ3mPwQkc3gWUpE1BiY/BCRTeFZSkRkbTZxvAUR0c14lhIRWROTHyKySTxLiYishcNeREREJCns+SG7ojcIHAohIqK7wuSH7AYPuyQiIkvgsBfZharDLm89+qDqsMvUvGKRIiMiInvD5Idsnt4gYOHm/GqnfAMwlS3cnA+9gWf0EhHR7TH5IZvHwy6JiMiSmPyQzeNhl0REZElMfsjm8bBLIiKyJCY/ZPN42CUREVkSkx+yeTzskoiILInJD9kFHnZJRESWwk0OyW7wsEsiIrIEJj9kV3jYJVHTw2NrqLEx+SEiItHw2BoSA+f8EBGRKHhsDYmFyQ8RETU6HltDYmLyQ0REjY7H1pCYmPwQEVGj47E1JCYmP0RE1Oh4bA2JickPERE1Oh5bQ2Ji8kNERI2Ox9aQmJj8EBGRKHhsDYmFmxwSEZFoeGwNiYHJDxERiYrH1lBj47AXERERSQp7foiISFwGPXB8N3CxFHBXA34DAQe52FFRE8bkh4iIxJOfAqTGAbrTN8o8fICIpUDgSPHioiaNw15ERCSO/BRgwwTzxAcAdMXG8vwUceKyVwY9UPQzcOi/xv8a9GJHZLPY80NERI3PoDf2+NR6tKkMSJ0DdBnOIbD6YA9ag7Dnh4iIGt/x3dV7fMwIgO6UsR7VjT1oDcbkh4iIGt/FUsvWk6rb9qDB2IPGITAzTH6IiKjxuastW0+q2IN2Rzjnp5HoDXrklOXgr0t/obVra/RW9Ybcxsax7SLGa1eQc+hz/KU7gdYe7dC729OQN3MUOywzdvE5MkaLYIx3wW+gcU6Krhh6CMhxdsJfcjla6/XoXVEJOWTG+34DxY4UgA1/jjf1jOmBGj7H6vXEZCufI5OfRrDt+DYs2bsEpZdu/OVTu6oxp/8chPmFiRjZDXYR4854LPnjC5TKb2x7r/7lHczpNA5h988VMbIb7OJzZIwWwRjvkoMciFiKbVumYklLT5Q2u/F1pL52DXPOnEdYxBKbmOxs05/j9Z6xba4uWNKyRQ2f4zmEXbpsEz1otvQ5ygRBqGmgUNJ0Oh2USiW0Wi08PDzuqq1tx7dhZsZMCLeMx8qun1v8dujbov/w2EWMO+Mx8+gXxghlN5If2fW/vm93ED8BsovPkTFaBGO0DGOML0IQhOo/1zIZ3g59x0ZitOHP0aDHto+6Yaa7Q+2/Hy8aEPb8IVETycb6HOv7/X1Xc34qKiqwZs0afPTRRzhy5MjdNNUk6Q16LNm7pNofNgBT2dK9S6EXcSKaXcR47QqW/FE98QEA4fr10j++gP7alcYP7jq7+BwZo0UwRsu4ESNq+bmW2VCMNvw5Alji1aLu348tvSDmdGdb/BzrnfzMnDkT06dPN11fuXIFwcHBePbZZ/HKK6+gV69e2LNnT4NePD4+Hv369YNCoYBKpUJkZCQKCgrM6kydOhUBAQFwcXFB69atMWrUKPz+++91thsVFQWZTGb2iIiIaFBslpBTlmPWvXcrAQJKLpUgpyynEaMyZxcxHvrcONQlq/mUZ0EmQ4lchpxDnzdyZDfYxefIGC2CMVoGY7SMnLIclF7V1f378YpW/Bht7HOsd/Lz448/4uGHHzZdf/HFFzh+/DiOHDmCc+fOYfTo0Vi8eHGDXjwzMxPR0dHIyspCWloarl69ivDwcJSXl5vq9OnTB4mJiTh8+DC2bt0KQRAQHh4Ovb7uDDEiIgLFxcWmx5dfftmg2Czhr0t/WbSeNdhFjLoTFq1nDXbxOTJGi2CMlsEYLYMx3pl6T3g+ceIEAgMDTdc//vgj/vnPf8LPzw8AMGPGDAwbNqxBL56ammp2nZSUBJVKhQMHDmDQoEEAgClTppjut2/fHosXL0aPHj1w7NgxBAQE1Nq2k5MTNBpNg+KxtNaurS1azxrsIkaPdhatZw128TkyRotgjJbBGC2DMd6Zevf8ODg44Oa50VlZWbjvvvtM156enjh37txdBaPVagEAXl5eNd4vLy9HYmIi/P394evrW2dbGRkZUKlU6Ny5M6ZNm4YzZ87UWreyshI6nc7sYQm9Vb2hdlWbJnTdSgYZNK4a9Fb1tsjr3Qm7iLHb01DrBdPkvVvJBAEavYDe3Z5u5MhusIvPkTFaBGO0DMZoGYzxztQ7+enatSs2b94MAPjtt99w4sQJPPjgg6b7x48fh1p950vpDAYDYmNjERISgqCgILN7H330Edzd3eHu7o4ffvgBaWlpcHSsfW+XiIgIrF27Funp6Vi6dCkyMzMxdOjQWofK4uPjoVQqTY/bJVb1JXeQY07/OQBQ7Q+96jquf5yoe0XYRYzNHDGn0zhjTLckQFXXcZ3Gibrfj118jozRIhijZTBGy2CMd6beS92Tk5MxduxY3H///fjtt9/Qr18/UzIEAHFxcSgqKsKGDRvuKJBp06bhhx9+wM6dO9G2bVuze1qtFmVlZSguLsby5ctx6tQp7Nq1C87OzvVq+88//0RAQAC2bduGhx56qNr9yspKVFZWmq51Oh18fX0tstQdqHlvA42rBnH940RfxlnFLmKsYZ8fjV5AnI3v82NznyNjtAjGaBmM0TIYo1F9l7o3aJ+f9PR0bNmyBRqNBtOnT4erq6vp3sKFCzF48GCEhoY2ONiYmBh8++23+Omnn+Dv719n3StXrqBFixZYvXo1nnzyyXq/RuvWrbF48WJMnTr1tnUtuc9PFVvZ1bIudhEjd3i2CMZoGYzRMhijZTBGKyQ/r7/+OmbNmmWW8NwtQRAwffp0JCcnIyMjAx07drztcyorK9GiRQt89NFHiIqKqtfr/O9//0O7du2wadMmjBw58rb1rZH8EBERkXVZfJPDhQsX4uLFixYJrkp0dDTWrVuH9evXQ6FQoKSkBCUlJbh8+TIA43BVfHw8Dhw4gBMnTmD37t0YPXo0XFxczFaWdenSBcnJyQCAixcv4uWXX0ZWVhaOHTuG9PR0jBo1Ch06dMCQIUMsGj8RSZveIGBP4Rl8m3sKewrPQG/ghvlE9qDeS92tcQpGQkICAFQbKktMTERUVBScnZ3x888/Y+XKlTh37hzUajUGDRqE3bt3Q6VSmeoXFBSYVorJ5XIcPHgQa9aswfnz5+Hj44Pw8HAsWrQITk5OFn8PRCRNqXnFWLg5H8XaClOZt9IZ80cEIiLIW8TIiOh26j3s5eDggNLSUrRuLd5eAY2Fw15EVJfUvGJMW5dTbbP+qqn4CeN7MwFqagx64Phu4+no7mrjafM2Np+G6v/93aBT3Tt16gRZLVtoVzl79mxDmiQisit6g4CFm/NrOKUIEGBMgBZuzsfDgRrIHer+fUl2Ij8FSI0DdKdvlHn4ABFLgcDbzyMl29Og5GfhwoVQKpXWioWIyObtLTprNtR1KwFAsbYCe4vOIjigZeMFRtaRnwJsmADcmu7qio3lY9YyAbJDDUp+xo4dazbXhohIasou1J743Ek9smEGvbHHp65+vtQ5QJfhHAKzM/Ve7XW74S4iIilQKeq3uWp965ENO77bfKirGgHQnTLWI7tS7+THGqu9iIjsTX9/L3grnWs5pcg458db6Yz+/jWfUUh25GLp7es0pB7ZjHonPwaDgUNeRCR5cgcZ5o8IBIBqCVDV9fwRgZzs3BS41/O8yvrWI5tR7+SHiIiMIoK8kTC+NzRK86EtjdKZy9ybEr+BxlVddfXzebQx1iO70qAJz0REZBQR5I2HAzXYW3QWZRcqoFIYh7rY49OEOMiNy9k3TIAxAbp5+sf1P+eIJZzsbIeY/BAR3SG5g4zL2Zu6wJHG5ew17vOzhMvc7RSTHyIioroEjjQuZ+cOz00Gkx8iIqLbcZAD/g+IHQVZCCc8ExERkaQw+SEiIiJJYfJDREREksLkh4iIiCSFyQ8RERFJCpMfIiIikhQmP0RERCQpTH6IiIhIUpj8EBERkaQw+SEiIiJJYfJDREREksLkh4iIiCSFyQ8RERFJCpMfIiIikhQmP0RERCQpTH6IiIhIUpj8EBERkaQw+SEiIiJJYfJDREREksLkh4iIiCSFyQ8RERFJCpMfIiIikhQmP0RERCQpTH6IiIhIUpj8EBERkaQw+SEiIiJJYfJDREREksLkh4iIiCSFyQ8RERFJCpMfIiIikhQmP0RERCQpTH6IiIhIUpj8EBERkaSImvzEx8ejX79+UCgUUKlUiIyMREFBgVmdqVOnIiAgAC4uLmjdujVGjRqF33//vc52BUHAa6+9Bm9vb7i4uCAsLAxHjhyx5lshIiIiOyFq8pOZmYno6GhkZWUhLS0NV69eRXh4OMrLy011+vTpg8TERBw+fBhbt26FIAgIDw+HXq+vtd1ly5bhvffew6pVq5CdnQ03NzcMGTIEFRUVjfG2iIiIyIbJBEEQxA6iyl9//QWVSoXMzEwMGjSoxjoHDx5Ejx49cPToUQQEBFS7LwgCfHx88NJLL2HWrFkAAK1WC7VajaSkJIwdO/a2ceh0OiiVSmi1Wnh4eNzdmyIiIqJGUd/vb5ua86PVagEAXl5eNd4vLy9HYmIi/P394evrW2OdoqIilJSUICwszFSmVCoxYMAA7Nmzp8bnVFZWQqfTmT2IiIioabKZ5MdgMCA2NhYhISEICgoyu/fRRx/B3d0d7u7u+OGHH5CWlgZHR8ca2ykpKQEAqNVqs3K1Wm26d6v4+HgolUrTo7bEioiIiOyfzSQ/0dHRyMvLw1dffVXt3rhx4/DLL78gMzMTnTp1wpgxYyw6f2fu3LnQarWmx8mTJy3WNhEREdmWZmIHAAAxMTHYsmULfvrpJ7Rt27ba/aoemY4dO+K+++5DixYtkJycjCeffLJaXY1GAwAoLS2Ft7e3qby0tBQ9e/as8fWdnJzg5ORkmTdDRERENk3Unh9BEBATE4Pk5GRs374d/v7+9XqOIAiorKys8b6/vz80Gg3S09NNZTqdDtnZ2QgODrZY7ERERGSfRE1+oqOjsW7dOqxfvx4KhQIlJSUoKSnB5cuXAQB//vkn4uPjceDAAZw4cQK7d+/G6NGj4eLigmHDhpna6dKlC5KTkwEAMpkMsbGxWLx4MVJSUnDo0CFMmDABPj4+iIyMFONtEhERkQ0RddgrISEBABAaGmpWnpiYiKioKDg7O+Pnn3/GypUrce7cOajVagwaNAi7d++GSqUy1S8oKDCtFAOA2bNno7y8HFOmTMH58+dx//33IzU1Fc7Ozo3yvoiIiMh22dQ+P7aC+/wQERHZH7vc54eIiIjI2pj8EBERkaQw+SEiIiJJYfJDREREksLkh4iIiCSFyQ8RERFJCpMfIiIikhQmP0RERCQpNnGwKVFTojcI2Ft0FmUXKqBSOKO/vxfkDjKxwyIiouuY/BBZUGpeMRZuzkextsJU5q10xvwRgYgI8hYxMiIiqsJhLyILSc0rxrR1OWaJDwCUaCswbV0OUvOKRYqMiIhuxuSHyAL0BgELN+dDAOAAA+5zyMdIh924zyEfMhgAAAs350Nv4FF6RERi47AXkQXsLTqLYm0Fhjjsxfzma+EjO2u6d1rwwsKrE7BV2x97i84iOKCliJESERF7fogsoOyCMfFJaL4SGpw1u6fBWSQ0X4khDntRdqGilhaIiKixMPkhsgCVW3PMb74WAHDrwq6q6/nNP4fKrXkjR0ZERLfisBeRBfSX/w657Gyt9x1kgA/OQC3/HYCq8QIjIqJq2PNDZAHy8jKL1iMiIuth8kNkCe5qy9YjIiKrYfJDZAl+AwEPHwC17eQsAzzaGOsREZGomPwQWYKDHIhYev3i1gTo+nXEEmM9IiISFZMfIksJHAmMWQt43HKMhYePsTxwpDhxERGRGa72IrKkwJFAl+HA8d3AxVLjHB+/gezxISKyIUx+iCzNQQ74PyB2FEREVAsOexEREZGksOeHiGyTQc/hQyKyCiY/RGR78lOA1DhAd/pGmYePcUUdJ44T0V3isBcR2Zb8FGDDBPPEBwB0xcby/BRx4iKiJoPJDxHZDoPe2OMDoYab18tS5xjrERHdISY/RGQ7ju+u3uNjRgB0p4z1iIjuEJMfIrIdF0stW4+IqAZMfojIdvCAWCJqBEx+iMh28IBYImoETH6IyHbwgFgiagRMfojItvCAWCKyMm5ySES2hwfEEpEVMfkhItvEA2KJyEo47EVERESSwuSHiIiIJIXJDxEREUkKkx8iIiKSFCY/REREJClMfoiIiEhSuNSdSIoMeu6hIxF6g4C9RWdRdqECKoUz+vt7Qe5Q2/EhRNLA5IdIavJTgNQ4QHf6RpmHj/FYCe6e3KSk5hVj4eZ8FGsrTGXeSmfMHxGIiCDvOp5J1LSJOuwVHx+Pfv36QaFQQKVSITIyEgUFBab7Z8+exfTp09G5c2e4uLigXbt2eOGFF6DVautsNyoqCjKZzOwRERFh7bdDZPvyU4ANE8wTHwDQFRvL81PEiYssLjWvGNPW5ZglPgBQoq3AtHU5SM0rFikyIvGJmvxkZmYiOjoaWVlZSEtLw9WrVxEeHo7y8nIAwOnTp3H69GksX74ceXl5SEpKQmpqKiZPnnzbtiMiIlBcXGx6fPnll9Z+O0S2zaA39vhAqOHm9bLUOcZ6ZNf0BgELN+fX9SeNhZvzoTfUVIOo6RN12Cs1NdXsOikpCSqVCgcOHMCgQYMQFBSEjRs3mu4HBATgjTfewPjx43Ht2jU0a1Z7+E5OTtBoNFaLncjuHN9dvcfHjADoThnr8VgJu7a36Gy1Hp+bCQCKtRXYW3QWwQEtGy8wIhthU6u9qoazvLy86qzj4eFRZ+IDABkZGVCpVOjcuTOmTZuGM2fO1Fq3srISOp3O7EHU5FwstWw9slllF2pPfO6kHlFTYzPJj8FgQGxsLEJCQhAUFFRjnb///huLFi3ClClT6mwrIiICa9euRXp6OpYuXYrMzEwMHToUen3N3fnx8fFQKpWmh6+v712/HyKb4662bD2yWSqFs0XrETU1MkEQbGLQd9q0afjhhx+wc+dOtG3bttp9nU6Hhx9+GF5eXkhJSUHz5s3r3faff/6JgIAAbNu2DQ899FC1+5WVlaisrDR7LV9fX1MvE1GTYNADK4OMk5trnA0iM676ij3EZe92Tm8QcP/S7SjRVtT2Jw2N0hk74/7BZe/UpOh0OiiVytt+f9tEz09MTAy2bNmCHTt21Jj4XLhwAREREVAoFEhOTm5Q4gMA99xzD1q1aoWjR4/WeN/JyQkeHh5mD6Imx0FuXM4OwPj1d7Pr1xFLmPg0AXIHGeaPCARQ65805o8IZOJDkiVq8iMIAmJiYpCcnIzt27fD39+/Wh2dTofw8HA4OjoiJSUFzs4N76b93//+hzNnzsDbm/takMQFjgTGrAU8bvlZ8PAxlnOfnyYjIsgbCeN7Q6M0/52pUTojYXxv7vNDkibqsNfzzz+P9evX49tvv0Xnzp1N5UqlEi4uLqbE59KlS0hOToabm5upTuvWrSGXG/+F2qVLF8THx+PRRx/FxYsXsXDhQjz++OPQaDQoLCzE7NmzceHCBRw6dAhOTk63jau+3WYNwV1WyaZwh2fJ4O8ekpL6fn+LutQ9ISEBABAaGmpWnpiYiKioKOTk5CA7OxsA0KFDB7M6RUVFaN++PQCgoKDAtFJMLpfj4MGDWLNmDc6fPw8fHx+Eh4dj0aJF9Up8rIG7rJLNcZBzObtEyB1kXM5OdAubmfBsSyzZ81O1y+qtH3LVv7vY/UxERGQZdjXhuaniLqtERES2h8mPFTVkl1UiIiJqHEx+rIi7rBIREdkeJj9WxF1WiYiIbA+THyvq7+8Fb6VztU3GqshgXPXV37/2s8yIiIjIspj8WBF3WSUiIrI9TH6sjLusEhER2RZRNzmUioggbzwcqOEuq0RERDaAyU8j4S6rREREtoHDXkRERCQpTH6IiIhIUpj8EBERkaQw+SEiIiJJYfJDREREksLVXmSiNwhcjk9ERE0ekx8CAKTmFWPh5nyzU+i9lc6YPyKQGzESEVGTwmEvQmpeMaatyzFLfACgRFuBaetykJpXLFJkRERElsfkR+L0BgELN+dDqOFeVdnCzfnQG2qqQUREZH+Y/Ejc3qKz1Xp8biYAKNZWYG/R2cYLioiIyIqY/Ehc2YXaE587qUdERGTrmPxInErhfPtKDahHRERk65j8SFx/fy94K51R24J2GYyrvvr7ezVmWERERFbD5Efi5A4yzB8RCADVEqCq6/kjArnfDxERNRlMfggRQd5IGN8bGqX50JZG6YyE8b25zw8RETUp3OSQABgToIcDNdzhmYiImjwmP2Qid5AhOKCl2GEQERFZFYe9iIiISFKY/BAREZGkMPkhIiIiSWHyQ0RERJLC5IeIiIgkhckPERERSQqTHyIiIpIUJj9EREQkKUx+iIiISFKY/BAREZGkMPkhIiIiSWHyQ0RERJLC5IeIiIgkhckPERERSQqTHyIiIpIUJj9EREQkKUx+iIiISFKY/BAREZGkMPkhIiIiSRE1+YmPj0e/fv2gUCigUqkQGRmJgoIC0/2zZ89i+vTp6Ny5M1xcXNCuXTu88MIL0Gq1dbYrCAJee+01eHt7w8XFBWFhYThy5Ii13w4RERHZAVGTn8zMTERHRyMrKwtpaWm4evUqwsPDUV5eDgA4ffo0Tp8+jeXLlyMvLw9JSUlITU3F5MmT62x32bJleO+997Bq1SpkZ2fDzc0NQ4YMQUVFRWO8LSIiIrJhMkEQBLGDqPLXX39BpVIhMzMTgwYNqrHOf/7zH4wfPx7l5eVo1qxZtfuCIMDHxwcvvfQSZs2aBQDQarVQq9VISkrC2LFjqz2nsrISlZWVpmudTgdfX19otVp4eHhY6N0RERGRNel0OiiVytt+f9vUnJ+q4SwvL68663h4eNSY+ABAUVERSkpKEBYWZipTKpUYMGAA9uzZU+Nz4uPjoVQqTQ9fX9+7eBdERDbEoAeKfgYO/df4X4Ne7IiIRFdzBiECg8GA2NhYhISEICgoqMY6f//9NxYtWoQpU6bU2k5JSQkAQK1Wm5Wr1WrTvVvNnTsXM2fONF1X9fwQEdm1/BQgNQ7Qnb5R5uEDRCwFAkeKFxeRyGwm+YmOjkZeXh527txZ432dTofhw4cjMDAQCxYssOhrOzk5wcnJyaJtEhGJKj8F2DABwC0zG3TFxvIxa5kAkWTZxLBXTEwMtmzZgh07dqBt27bV7l+4cAERERFQKBRITk5G8+bNa21Lo9EAAEpLS83KS0tLTfeIiJo0g97Y43Nr4gPcKEudwyEwkixRkx9BEBATE4Pk5GRs374d/v7+1erodDqEh4fD0dERKSkpcHZ2rrNNf39/aDQapKenm7WRnZ2N4OBgi78HIiKbc3y3+VBXNQKgO2WsRyRBoiY/0dHRWLduHdavXw+FQoGSkhKUlJTg8uXLAG4kPuXl5fj3v/8NnU5nqqPX3/gXS5cuXZCcnAwAkMlkiI2NxeLFi5GSkoJDhw5hwoQJ8PHxQWRkpBhvk4iocV0svX2dhtQjamJEnfOTkJAAAAgNDTUrT0xMRFRUFHJycpCdnQ0A6NChg1mdoqIitG/fHgBQUFBgtvHh7NmzUV5ejilTpuD8+fO4//77kZqaetteIyKiJsFdffs6DalH1MTY1D4/tqK++wQQEdkkgx5YGWSc3FzjvB+ZcdVX7CHAQd7Y0RFZjV3u80NERBbgIDcuZwcAyG65ef06YgkTH5IsJj9ERE1R4EjjcnYPb/NyDx8ucyfJs5l9foiIyMICRwJdhhtXdV0sNc7x8RvIHh+SPCY/RERNmYMc8H9A7CiIbAqHvYiIiEhSmPwQERGRpDD5ISIiIklh8kNERESSwuSHiIiIJIXJDxEREUkKkx8iIiKSFCY/REREJClMfoiIiEhSuMNzDaoOutfpdCJHQkRERPVV9b1d9T1eGyY/Nbhw4QIAwNfXV+RIiIiIqKEuXLgApVJZ632ZcLv0SIIMBgNOnz4NhUIBmUxmsXZ1Oh18fX1x8uRJeHh4WKxdqeHnaBn8HC2Dn6Nl8HO0DKl/joIg4MKFC/Dx8YGDQ+0ze9jzUwMHBwe0bdvWau17eHhI8i+lpfFztAx+jpbBz9Ey+DlahpQ/x7p6fKpwwjMRERFJCpMfIiIikhQmP43IyckJ8+fPh5OTk9ih2DV+jpbBz9Ey+DlaBj9Hy+DnWD+c8ExERESSwp4fIiIikhQmP0RERCQpTH6IiIhIUpj8EBERkaQw+WlEH374Idq3bw9nZ2cMGDAAe/fuFTskuxIfH49+/fpBoVBApVIhMjISBQUFYodl95YsWQKZTIbY2FixQ7E7p06dwvjx49GyZUu4uLigW7du2L9/v9hh2RW9Xo958+bB398fLi4uCAgIwKJFi257NpPU/fTTTxgxYgR8fHwgk8mwadMms/uCIOC1116Dt7c3XFxcEBYWhiNHjogTrA1i8tNIvv76a8ycORPz589HTk4OevTogSFDhqCsrEzs0OxGZmYmoqOjkZWVhbS0NFy9ehXh4eEoLy8XOzS7tW/fPnz88cfo3r272KHYnXPnziEkJATNmzfHDz/8gPz8fKxYsQItWrQQOzS7snTpUiQkJOCDDz7A4cOHsXTpUixbtgzvv/++2KHZtPLycvTo0QMffvhhjfeXLVuG9957D6tWrUJ2djbc3NwwZMgQVFRUNHKkNkqgRtG/f38hOjradK3X6wUfHx8hPj5exKjsW1lZmQBAyMzMFDsUu3ThwgWhY8eOQlpamjB48GBhxowZYodkV+Li4oT7779f7DDs3vDhw4VJkyaZlT322GPCuHHjRIrI/gAQkpOTTdcGg0HQaDTCW2+9ZSo7f/684OTkJHz55ZciRGh72PPTCK5cuYIDBw4gLCzMVObg4ICwsDDs2bNHxMjsm1arBQB4eXmJHIl9io6OxvDhw83+XlL9paSkoG/fvhg9ejRUKhV69eqFTz/9VOyw7M7AgQORnp6OP/74AwDw66+/YufOnRg6dKjIkdmvoqIilJSUmP1sK5VKDBgwgN851/Fg00bw999/Q6/XQ61Wm5Wr1Wr8/vvvIkVl3wwGA2JjYxESEoKgoCCxw7E7X331FXJycrBv3z6xQ7Fbf/75JxISEjBz5ky88sor2LdvH1544QU4Ojpi4sSJYodnN+bMmQOdTocuXbpALpdDr9fjjTfewLhx48QOzW6VlJQAQI3fOVX3pI7JD9ml6Oho5OXlYefOnWKHYndOnjyJGTNmIC0tDc7OzmKHY7cMBgP69u2LN998EwDQq1cv5OXlYdWqVUx+GmDDhg344osvsH79etx7773Izc1FbGwsfHx8+DmS1XDYqxG0atUKcrkcpaWlZuWlpaXQaDQiRWW/YmJisGXLFuzYsQNt27YVOxy7c+DAAZSVlaF3795o1qwZmjVrhszMTLz33nto1qwZ9Hq92CHaBW9vbwQGBpqVde3aFSdOnBApIvv08ssvY86cORg7diy6deuGp59+Gi+++CLi4+PFDs1uVX2v8Dundkx+GoGjoyP69OmD9PR0U5nBYEB6ejqCg4NFjMy+CIKAmJgYJCcnY/v27fD39xc7JLv00EMP4dChQ8jNzTU9+vbti3HjxiE3NxdyuVzsEO1CSEhIta0W/vjjD/j5+YkUkX26dOkSHBzMv4rkcjkMBoNIEdk/f39/aDQas+8cnU6H7Oxsfudcx2GvRjJz5kxMnDgRffv2Rf/+/bFy5UqUl5fjmWeeETs0uxEdHY3169fj22+/hUKhMI1dK5VKuLi4iByd/VAoFNXmSbm5uaFly5acP9UAL774IgYOHIg333wTY8aMwd69e/HJJ5/gk08+ETs0uzJixAi88cYbaNeuHe6991788ssvePvttzFp0iSxQ7NpFy9exNGjR03XRUVFyM3NhZeXF9q1a4fY2FgsXrwYHTt2hL+/P+bNmwcfHx9ERkaKF7QtEXu5mZS8//77Qrt27QRHR0ehf//+QlZWltgh2RUANT4SExPFDs3ucan7ndm8ebMQFBQkODk5CV26dBE++eQTsUOyOzqdTpgxY4bQrl07wdnZWbjnnnuE//u//xMqKyvFDs2m7dixo8bfhxMnThQEwbjcfd68eYJarRacnJyEhx56SCgoKBA3aBsiEwRuo0lERETSwTk/REREJClMfoiIiEhSmPwQERGRpDD5ISIiIklh8kNERESSwuSHiIiIJIXJDxEREUkKkx8iIiKSFCY/RCQ5UVFR3OafSMKY/BCRVURFRUEmk1V7REREiB0a3n33XSQlJYkdBgBAJpNh06ZNYodBJCk82JSIrCYiIgKJiYlmZU5OTiJFA+j1eshkMiiVStFiICLxseeHiKzGyckJGo3G7NGiRQtkZGTA0dERP//8s6nusmXLoFKpUFpaCgAIDQ1FTEwMYmJioFQq0apVK8ybNw83H0dYWVmJWbNmoU2bNnBzc8OAAQOQkZFhup+UlARPT0+kpKQgMDAQTk5OOHHiRLVhr9DQUEyfPh2xsbFo0aIF1Go1Pv30U5SXl+OZZ56BQqFAhw4d8MMPP5i9v7y8PAwdOhTu7u5Qq9V4+umn8ffff5u1+8ILL2D27Nnw8vKCRqPBggULTPfbt28PAHj00Uchk8lM10RkXUx+iKjRhYaGIjY2Fk8//TS0Wi1++eUXzJs3D6tXr4ZarTbVW7NmDZo1a4a9e/fi3Xffxdtvv43Vq1eb7sfExGDPnj346quvcPDgQYwePRoRERE4cuSIqc6lS5ewdOlSrF69Gr/99htUKlWNMa1ZswatWrXC3r17MX36dEybNg2jR4/GwIEDkZOTg/DwcDz99NO4dOkSAOD8+fP4xz/+gV69emH//v1ITU1FaWkpxowZU61dNzc3ZGdnY9myZXj99deRlpYGANi3bx8AIDExEcXFxaZrIrIykU+VJ6ImauLEiYJcLhfc3NzMHm+88YYgCIJQWVkp9OzZUxgzZowQGBgoPPvss2bPHzx4sNC1a1fBYDCYyuLi4oSuXbsKgiAIx48fF+RyuXDq1Cmz5z300EPC3LlzBUEQhMTERAGAkJubWy22UaNGmb3W/fffb7q+du2a4ObmJjz99NOmsuLiYgGAsGfPHkEQBGHRokVCeHi4WbsnT54UAAgFBQU1tisIgtCvXz8hLi7OdA1ASE5OruVTJCJr4JwfIrKaBx98EAkJCWZlXl5eAABHR0d88cUX6N69O/z8/PDOO+9Ue/59990HmUxmug4ODsaKFSug1+tx6NAh6PV6dOrUyew5lZWVaNmypena0dER3bt3v22sN9eRy+Vo2bIlunXrZiqr6pEqKysDAPz666/YsWMH3N3dq7VVWFhoiuvW1/b29ja1QUTiYPJDRFbj5uaGDh061Hp/9+7dAICzZ8/i7NmzcHNzq3fbFy9ehFwux4EDByCXy83u3ZyQuLi4mCVQtWnevLnZtUwmMyurasNgMJhef8SIEVi6dGm1try9vetst6oNIhIHkx8iEkVhYSFefPFFfPrpp/j6668xceJEbNu2DQ4ON6YiZmdnmz0nKysLHTt2hFwuR69evaDX61FWVoYHHnigscNH7969sXHjRrRv3x7Nmt35r9LmzZtDr9dbMDIiuh1OeCYiq6msrERJSYnZ4++//4Zer8f48eMxZMgQPPPMM0hMTMTBgwexYsUKs+efOHECM2fOREFBAb788ku8//77mDFjBgCgU6dOGDduHCZMmIBvvvkGRUVF2Lt3L+Lj4/Hdd99Z/b1FR0fj7NmzePLJJ7Fv3z4UFhZi69ateOaZZxqUzLRv3x7p6ekoKSnBuXPnrBgxEVVhzw8RWU1qaqrZEBAAdO7cGU899RSOHz+OLVu2ADAOE33yySd48sknER4ejh49egAAJkyYgMuXL6N///6Qy+WYMWMGpkyZYmorMTERixcvxksvvYRTp06hVatWuO+++/DII49Y/b35+Phg165diIuLQ3h4OCorK+Hn54eIiAiz3qvbWbFiBWbOnIlPP/0Ubdq0wbFjx6wXNBEBAGSCcNOmGURENiI0NBQ9e/bEypUrxQ6FiJoYDnsRERGRpDD5ISIiIknhsBcRERFJCnt+iIiISFKY/BAREZGkMPkhIiIiSWHyQ0RERJLC5IeIiIgkhckPERERSQqTHyIiIpIUJj9EREQkKf8P0Kq/AA+dOkkAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(f\"The MSE loss is {hist_test[0]:.3f}\")\n",
    "\n",
    "fig = plt.figure()\n",
    "instances = np.arange(len(np.concatenate(predictions)))\n",
    "plt.scatter(instances, np.concatenate(predictions), label=\"Predictions\")\n",
    "plt.scatter(instances, test_y_torch.squeeze().numpy(), label=\"Ground truth\")\n",
    "plt.scatter(instances, [test_y_torch.squeeze().numpy().mean()] * len(instances), label=\"Mean of ground truth\")\n",
    "plt.xlabel(\"Experiment\")\n",
    "plt.ylabel(\"TS\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "s2spy",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.6"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}