BCI_binary.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X.shape (160, 31, 32, 3)\n",
      "Y.shape (160, 1)\n",
      "Original data type: float64\n",
      "Normalized data type: float16\n",
      "test input shape (80, 31, 32, 3) Nomralized MEAN: 331.5 min 0.0013247 max 17072.0\n",
      "train input shape (80, 31, 32, 3) Nomralized MEAN: 304.5 min 0.00088501 max 10232.0\n",
      "test labels shape (80, 2) Nomralized MEAN: 0.5 min 0.0 max 1.0\n",
      "train labels shape (80, 2) Nomralized MEAN: 0.5 min 0.0 max 1.0\n"
     ]
    }
   ],
   "source": [
    "# get the Dataset:\n",
    "\n",
    "import numpy as np\n",
    "import time\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn import preprocessing\n",
    "import scipy.io as sio\n",
    "\n",
    "t0 = time.time()\n",
    "\n",
    "mat_contents = sio.loadmat('/home/amplifier/home/DATASETS/BCI.mat')\n",
    "\n",
    "X = mat_contents['X']\n",
    "Y = mat_contents['Z']\n",
    "print('X.shape', X.shape)\n",
    "print('Y.shape', Y.shape)\n",
    "\n",
    "# verify that the model REALLY finds a mapping between the input and the labels. If we get\n",
    "# our accuracy by chance, then we should get the same accuracy on a permuted dataset:\n",
    "# Y = np.random.permutation(Y)\n",
    "\n",
    "x_train, x_test, y_train, y_test = train_test_split(X, Y, test_size=0.50)\n",
    "print('Original data type:', x_train.dtype)\n",
    "\n",
    "# convert to float64 for numerical stability:\n",
    "x_train = x_train.astype('float64')\n",
    "y_train = y_train.astype('float64')\n",
    "x_test = x_test.astype('float64')\n",
    "y_test = y_test.astype('float64')\n",
    "\n",
    "# normalize to unit variance and zero mean: (by the way, it kills the learning for some reason)\n",
    "# for j in range(3):\n",
    "#     for i in range(x_train.shape[0]):\n",
    "#         x_train[i,:,:,j] = preprocessing.scale(x_train[i,:,:,j], axis=1)\n",
    "#         x_train[i,:,:,j] = preprocessing.minmax_scale(x_train[i,:,:,j], axis=1)\n",
    "\n",
    "#     for i in range(x_test.shape[0]):\n",
    "#         x_test[i,:,:,j] = preprocessing.scale(x_test[i,:,:,j], axis=1)\n",
    "#         x_test[i,:,:,j] = preprocessing.minmax_scale(x_test[i,:,:,j], axis=1)\n",
    "\n",
    "# # ensure the tensors are sized right:\n",
    "# x_train = np.transpose(x_train,(0,2,1))\n",
    "# x_test = np.transpose(x_test,(0,2,1))\n",
    "\n",
    "# one hot encode the labels:\n",
    "onehot_encoder = preprocessing.OneHotEncoder(sparse=False)\n",
    "y_train = onehot_encoder.fit_transform(y_train)\n",
    "y_test = onehot_encoder.fit_transform(y_test)\n",
    "\n",
    "# convert to float16 to save space:\n",
    "x_train = x_train.astype('float16')\n",
    "y_train = y_train.astype('float16')\n",
    "x_test = x_test.astype('float16')\n",
    "y_test = y_test.astype('float16')\n",
    "print('Normalized data type:', x_train.dtype)\n",
    "\n",
    "print('test input shape', x_test.shape, \"Nomralized MEAN:\", np.mean(x_test), \"min\", np.min(x_test),\"max\", np.max(x_test))\n",
    "print('train input shape', x_train.shape, \"Nomralized MEAN:\", np.mean(x_train), \"min\", np.min(x_train),\"max\", np.max(x_train))\n",
    "\n",
    "print('test labels shape', y_test.shape, \"Nomralized MEAN:\", np.mean(y_test), \"min\", np.min(y_test),\"max\", np.max(y_test))\n",
    "print('train labels shape', y_train.shape, \"Nomralized MEAN:\", np.mean(y_train), \"min\", np.min(y_train),\"max\", np.max(y_train))\n",
    "\n",
    "# print('Window length', winsize)\n",
    "# print('Step size:', stepsize)\n",
    "# print('Length of trial during experiment', trial_len)\n",
    "# print('Low cutoff freq.', low_cutoff)\n",
    "# print('High cutoff freq.', high_cutoff)\n",
    "# print('Source data file:', source)\n",
    "\n",
    "leng = x_test.shape[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_17 (InputLayer)        (None, 31, 32, 3)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_95 (Batc (None, 31, 32, 3)         124       \n",
      "_________________________________________________________________\n",
      "conv2d_81 (Conv2D)           (None, 31, 32, 6)         114       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_65 (MaxPooling (None, 16, 16, 6)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_96 (Batc (None, 16, 16, 6)         64        \n",
      "_________________________________________________________________\n",
      "dropout_80 (Dropout)         (None, 16, 16, 6)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_82 (Conv2D)           (None, 16, 16, 12)        444       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_66 (MaxPooling (None, 8, 8, 12)          0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_97 (Batc (None, 8, 8, 12)          32        \n",
      "_________________________________________________________________\n",
      "dropout_81 (Dropout)         (None, 8, 8, 12)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_83 (Conv2D)           (None, 8, 8, 24)          1752      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_67 (MaxPooling (None, 4, 4, 24)          0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_98 (Batc (None, 4, 4, 24)          16        \n",
      "_________________________________________________________________\n",
      "dropout_82 (Dropout)         (None, 4, 4, 24)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_84 (Conv2D)           (None, 4, 4, 48)          6960      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_68 (MaxPooling (None, 2, 2, 48)          0         \n",
      "_________________________________________________________________\n",
      "dropout_83 (Dropout)         (None, 2, 2, 48)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_85 (Conv2D)           (None, 2, 2, 96)          27744     \n",
      "_________________________________________________________________\n",
      "flatten_17 (Flatten)         (None, 384)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_99 (Batc (None, 384)               1536      \n",
      "_________________________________________________________________\n",
      "dense_31 (Dense)             (None, 10)                3850      \n",
      "_________________________________________________________________\n",
      "dropout_84 (Dropout)         (None, 10)                0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_100 (Bat (None, 10)                40        \n",
      "_________________________________________________________________\n",
      "dense_32 (Dense)             (None, 2)                 22        \n",
      "=================================================================\n",
      "Total params: 42,698\n",
      "Trainable params: 41,792\n",
      "Non-trainable params: 906\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "######################################################################\n",
    "\n",
    "# http://cs231n.github.io/convolutional-networks/#case\n",
    "# we first define the autoencoder's architecture:\n",
    "\n",
    "# make sure you don't hog all the video memory\n",
    "import tensorflow as tf\n",
    "config = tf.ConfigProto()\n",
    "config.gpu_options.allow_growth = True\n",
    "sess = tf.Session(config=config)\n",
    "from keras import backend as K\n",
    "K.set_session(sess)\n",
    "###################################\n",
    "\n",
    "from keras.layers import Input, Dense, Conv2D, MaxPooling2D, UpSampling2D, Dropout, BatchNormalization, Flatten, Conv1D, MaxPooling1D\n",
    "from keras.layers.advanced_activations import LeakyReLU\n",
    "# act = LeakyReLU(alpha=0.01)\n",
    "act = 'relu'\n",
    "from keras.models import Model\n",
    "from keras.models import load_model\n",
    "from keras.callbacks import TensorBoard, EarlyStopping, Callback\n",
    "from keras import optimizers\n",
    "from keras.initializers import Orthogonal as orth\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "init_num_kernels = 6\n",
    "kernel_size = (3,2) # this means we're convolving more agressively in frequency domain than in time\n",
    "dense_n = 10\n",
    "\n",
    "def Encoder():\n",
    "    input_img = Input(shape=(31, 32, 3))  # adapt this if using `channels_first` image\n",
    "    input_norm = BatchNormalization(axis=1)(input_img)\n",
    "    e1 = Conv2D(init_num_kernels, kernel_size, activation='relu', padding='same',\n",
    "                                    kernel_initializer = orth(gain=1.0, seed=None))(input_norm)\n",
    "    e2 = MaxPooling2D((2, 2), padding='same')(e1)\n",
    "    e2b = BatchNormalization(axis=1)(e2)\n",
    "    e2a = Dropout(0.1)(e2b)\n",
    "    e3 = Conv2D(init_num_kernels*2, kernel_size, activation='relu', padding='same',\n",
    "                                    kernel_initializer = orth(gain=1.0, seed=None))(e2a)\n",
    "    e4 = MaxPooling2D((2, 2), padding='same')(e3)\n",
    "    e4b = BatchNormalization(axis=1)(e4)\n",
    "    e4a = Dropout(0.2)(e4b)\n",
    "    e5 = Conv2D(init_num_kernels*4, kernel_size, activation='relu', padding='same',\n",
    "                                    kernel_initializer = orth(gain=1.0, seed=None))(e4a)\n",
    "    e6 = MaxPooling2D((2, 2), padding='same')(e5)\n",
    "    e6b = BatchNormalization(axis=1)(e6)\n",
    "    e6a = Dropout(0.3)(e6b)\n",
    "    e7 = Conv2D(init_num_kernels*8, kernel_size, activation='relu', padding='same',\n",
    "                                    kernel_initializer = orth(gain=1.0, seed=None))(e6a)\n",
    "    e8 = MaxPooling2D((2, 2), padding='same')(e7)\n",
    "    e9 = Dropout(0.4)(e8)\n",
    "    e9b = Conv2D(init_num_kernels*16, kernel_size, activation='relu', padding='same',\n",
    "                                    kernel_initializer = orth(gain=1.0, seed=None))(e9)\n",
    "    e10 = Flatten()(e9b)\n",
    "    e10b = BatchNormalization(axis=1)(e10)\n",
    "    e11 = Dense(dense_n, activation='sigmoid',\n",
    "                                     kernel_initializer = orth(gain=1.0, seed=None))(e10b)\n",
    "    e12 = Dropout(0.5)(e11)\n",
    "    e12b = BatchNormalization(axis=1)(e12)\n",
    "    e13 = Dense(2, activation='softmax')(e12b)\n",
    "\n",
    "    return Model(input_img, e13)\n",
    "\n",
    "enc = Encoder()\n",
    "enc.compile(loss=\"binary_crossentropy\", optimizer=\"adadelta\", metrics=[\"accuracy\"])\n",
    "   \n",
    "enc.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 80 samples, validate on 80 samples\n",
      "Epoch 1/1000\n",
      "80/80 [==============================] - 1s 11ms/step - loss: 1.1692 - acc: 0.4750 - val_loss: 0.6310 - val_acc: 0.6875\n",
      "Epoch 2/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.8701 - acc: 0.5625 - val_loss: 0.5477 - val_acc: 0.7625\n",
      "Epoch 3/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.8438 - acc: 0.6000 - val_loss: 0.5011 - val_acc: 0.7500\n",
      "Epoch 4/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.8601 - acc: 0.6125 - val_loss: 0.4803 - val_acc: 0.7500\n",
      "Epoch 5/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.7420 - acc: 0.6375 - val_loss: 0.4816 - val_acc: 0.7625\n",
      "Epoch 6/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.5217 - acc: 0.7000 - val_loss: 0.4576 - val_acc: 0.7625\n",
      "Epoch 7/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.5835 - acc: 0.7500 - val_loss: 0.4231 - val_acc: 0.7750\n",
      "Epoch 8/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.6394 - acc: 0.6750 - val_loss: 0.3723 - val_acc: 0.8250\n",
      "Epoch 9/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.5985 - acc: 0.7375 - val_loss: 0.3393 - val_acc: 0.8375\n",
      "Epoch 10/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.6359 - acc: 0.7000 - val_loss: 0.3575 - val_acc: 0.8375\n",
      "Epoch 11/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4966 - acc: 0.7000 - val_loss: 0.3416 - val_acc: 0.8500\n",
      "Epoch 12/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.5681 - acc: 0.7500 - val_loss: 0.3171 - val_acc: 0.8625\n",
      "Epoch 13/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.5242 - acc: 0.7625 - val_loss: 0.2883 - val_acc: 0.8750\n",
      "Epoch 14/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3491 - acc: 0.8375 - val_loss: 0.2981 - val_acc: 0.8625\n",
      "Epoch 15/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.5528 - acc: 0.7500 - val_loss: 0.3158 - val_acc: 0.8750\n",
      "Epoch 16/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4413 - acc: 0.7500 - val_loss: 0.3341 - val_acc: 0.8750\n",
      "Epoch 17/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.5551 - acc: 0.7500 - val_loss: 0.3606 - val_acc: 0.8500\n",
      "Epoch 18/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4533 - acc: 0.8000 - val_loss: 0.3224 - val_acc: 0.8625\n",
      "Epoch 19/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3629 - acc: 0.8375 - val_loss: 0.3073 - val_acc: 0.8750\n",
      "Epoch 20/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4154 - acc: 0.8000 - val_loss: 0.2707 - val_acc: 0.9000\n",
      "Epoch 21/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4412 - acc: 0.8125 - val_loss: 0.2888 - val_acc: 0.8750\n",
      "Epoch 22/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4313 - acc: 0.8500 - val_loss: 0.3586 - val_acc: 0.8500\n",
      "Epoch 23/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4895 - acc: 0.7875 - val_loss: 0.3359 - val_acc: 0.8625\n",
      "Epoch 24/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4002 - acc: 0.8500 - val_loss: 0.3289 - val_acc: 0.8625\n",
      "Epoch 25/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.6102 - acc: 0.7000 - val_loss: 0.3375 - val_acc: 0.8500\n",
      "Epoch 26/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3202 - acc: 0.8250 - val_loss: 0.3121 - val_acc: 0.8500\n",
      "Epoch 27/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3144 - acc: 0.9000 - val_loss: 0.2944 - val_acc: 0.8875\n",
      "Epoch 28/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.5079 - acc: 0.8250 - val_loss: 0.2700 - val_acc: 0.9125\n",
      "Epoch 29/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3913 - acc: 0.8625 - val_loss: 0.2621 - val_acc: 0.9000\n",
      "Epoch 30/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3963 - acc: 0.8500 - val_loss: 0.3111 - val_acc: 0.8625\n",
      "Epoch 31/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3504 - acc: 0.8875 - val_loss: 0.2615 - val_acc: 0.9000\n",
      "Epoch 32/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4196 - acc: 0.8250 - val_loss: 0.2759 - val_acc: 0.9125\n",
      "Epoch 33/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4330 - acc: 0.7875 - val_loss: 0.3660 - val_acc: 0.8375\n",
      "Epoch 34/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4303 - acc: 0.7875 - val_loss: 0.3614 - val_acc: 0.8250\n",
      "Epoch 35/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3421 - acc: 0.8750 - val_loss: 0.3940 - val_acc: 0.8375\n",
      "Epoch 36/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3084 - acc: 0.8875 - val_loss: 0.3958 - val_acc: 0.8375\n",
      "Epoch 37/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3804 - acc: 0.8500 - val_loss: 0.4093 - val_acc: 0.8375\n",
      "Epoch 38/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3901 - acc: 0.8125 - val_loss: 0.3445 - val_acc: 0.8500\n",
      "Epoch 39/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3962 - acc: 0.7875 - val_loss: 0.3866 - val_acc: 0.8500\n",
      "Epoch 40/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3934 - acc: 0.8750 - val_loss: 0.3592 - val_acc: 0.8500\n",
      "Epoch 41/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4799 - acc: 0.8125 - val_loss: 0.4019 - val_acc: 0.8375\n",
      "Epoch 42/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.4605 - acc: 0.8250 - val_loss: 0.4513 - val_acc: 0.8000\n",
      "Epoch 43/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3419 - acc: 0.8875 - val_loss: 0.4400 - val_acc: 0.8000\n",
      "Epoch 44/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2997 - acc: 0.8750 - val_loss: 0.4658 - val_acc: 0.8000\n",
      "Epoch 45/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3501 - acc: 0.8625 - val_loss: 0.4110 - val_acc: 0.8000\n",
      "Epoch 46/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3525 - acc: 0.8750 - val_loss: 0.3999 - val_acc: 0.8500\n",
      "Epoch 47/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2504 - acc: 0.9250 - val_loss: 0.3435 - val_acc: 0.8625\n",
      "Epoch 48/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2654 - acc: 0.9000 - val_loss: 0.3522 - val_acc: 0.8625\n",
      "Epoch 49/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2580 - acc: 0.9125 - val_loss: 0.2613 - val_acc: 0.8875\n",
      "Epoch 50/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3967 - acc: 0.8000 - val_loss: 0.3139 - val_acc: 0.8500\n",
      "Epoch 51/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4318 - acc: 0.8375 - val_loss: 0.2492 - val_acc: 0.9125\n",
      "Epoch 52/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2894 - acc: 0.8750 - val_loss: 0.2941 - val_acc: 0.8750\n",
      "Epoch 53/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3195 - acc: 0.8875 - val_loss: 0.2465 - val_acc: 0.8875\n",
      "Epoch 54/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3197 - acc: 0.9000 - val_loss: 0.2882 - val_acc: 0.8750\n",
      "Epoch 55/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2814 - acc: 0.9125 - val_loss: 0.2391 - val_acc: 0.9000\n",
      "Epoch 56/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2977 - acc: 0.9000 - val_loss: 0.2359 - val_acc: 0.9000\n",
      "Epoch 57/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3396 - acc: 0.8750 - val_loss: 0.2923 - val_acc: 0.8500\n",
      "Epoch 58/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2322 - acc: 0.9250 - val_loss: 0.3204 - val_acc: 0.8750\n",
      "Epoch 59/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3564 - acc: 0.8500 - val_loss: 0.3181 - val_acc: 0.8750\n",
      "Epoch 60/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2374 - acc: 0.9125 - val_loss: 0.3666 - val_acc: 0.8625\n",
      "Epoch 61/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4146 - acc: 0.8000 - val_loss: 0.3724 - val_acc: 0.8625\n",
      "Epoch 62/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2500 - acc: 0.9250 - val_loss: 0.3895 - val_acc: 0.8625\n",
      "Epoch 63/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3715 - acc: 0.8250 - val_loss: 0.2945 - val_acc: 0.8750\n",
      "Epoch 64/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2041 - acc: 0.9250 - val_loss: 0.2732 - val_acc: 0.8750\n",
      "Epoch 65/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3658 - acc: 0.8125 - val_loss: 0.2490 - val_acc: 0.9000\n",
      "Epoch 66/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2864 - acc: 0.9000 - val_loss: 0.2411 - val_acc: 0.9125\n",
      "Epoch 67/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2394 - acc: 0.8750 - val_loss: 0.2687 - val_acc: 0.9000\n",
      "Epoch 68/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3636 - acc: 0.8500 - val_loss: 0.2820 - val_acc: 0.8750\n",
      "Epoch 69/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3434 - acc: 0.8625 - val_loss: 0.2719 - val_acc: 0.8875\n",
      "Epoch 70/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1966 - acc: 0.9500 - val_loss: 0.2660 - val_acc: 0.9000\n",
      "Epoch 71/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2740 - acc: 0.8875 - val_loss: 0.3128 - val_acc: 0.8500\n",
      "Epoch 72/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3911 - acc: 0.8500 - val_loss: 0.2680 - val_acc: 0.8625\n",
      "Epoch 73/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3362 - acc: 0.9000 - val_loss: 0.3784 - val_acc: 0.8500\n",
      "Epoch 74/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3711 - acc: 0.8750 - val_loss: 0.4205 - val_acc: 0.8250\n",
      "Epoch 75/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1839 - acc: 0.9375 - val_loss: 0.3431 - val_acc: 0.8625\n",
      "Epoch 76/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2501 - acc: 0.9250 - val_loss: 0.3309 - val_acc: 0.8625\n",
      "Epoch 77/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2662 - acc: 0.8875 - val_loss: 0.3017 - val_acc: 0.8750\n",
      "Epoch 78/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3251 - acc: 0.8625 - val_loss: 0.3212 - val_acc: 0.8750\n",
      "Epoch 79/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1763 - acc: 0.9375 - val_loss: 0.2986 - val_acc: 0.8625\n",
      "Epoch 80/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2952 - acc: 0.8875 - val_loss: 0.2849 - val_acc: 0.8875\n",
      "Epoch 81/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3045 - acc: 0.9000 - val_loss: 0.2347 - val_acc: 0.9250\n",
      "Epoch 82/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2586 - acc: 0.8625 - val_loss: 0.3608 - val_acc: 0.8500\n",
      "Epoch 83/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2156 - acc: 0.9000 - val_loss: 0.3000 - val_acc: 0.8750\n",
      "Epoch 84/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3529 - acc: 0.8375 - val_loss: 0.4826 - val_acc: 0.8125\n",
      "Epoch 85/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2379 - acc: 0.9125 - val_loss: 0.2855 - val_acc: 0.8625\n",
      "Epoch 86/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2447 - acc: 0.9000 - val_loss: 0.4357 - val_acc: 0.8250\n",
      "Epoch 87/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2528 - acc: 0.9250 - val_loss: 0.4699 - val_acc: 0.8125\n",
      "Epoch 88/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2088 - acc: 0.9375 - val_loss: 0.4006 - val_acc: 0.8375\n",
      "Epoch 89/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3375 - acc: 0.8875 - val_loss: 0.4543 - val_acc: 0.8125\n",
      "Epoch 90/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2192 - acc: 0.9250 - val_loss: 0.4379 - val_acc: 0.8375\n",
      "Epoch 91/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3358 - acc: 0.8750 - val_loss: 0.3776 - val_acc: 0.8500\n",
      "Epoch 92/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3272 - acc: 0.8625 - val_loss: 0.3291 - val_acc: 0.8500\n",
      "Epoch 93/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2855 - acc: 0.8500 - val_loss: 0.3344 - val_acc: 0.8375\n",
      "Epoch 94/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2557 - acc: 0.9000 - val_loss: 0.3640 - val_acc: 0.8375\n",
      "Epoch 95/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3739 - acc: 0.8750 - val_loss: 0.4598 - val_acc: 0.8250\n",
      "Epoch 96/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3302 - acc: 0.8875 - val_loss: 0.4821 - val_acc: 0.8250\n",
      "Epoch 97/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2580 - acc: 0.9375 - val_loss: 0.4237 - val_acc: 0.8375\n",
      "Epoch 98/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2282 - acc: 0.8875 - val_loss: 0.3392 - val_acc: 0.8250\n",
      "Epoch 99/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2151 - acc: 0.9250 - val_loss: 0.2964 - val_acc: 0.8750\n",
      "Epoch 100/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3571 - acc: 0.8375 - val_loss: 0.3614 - val_acc: 0.8375\n",
      "Epoch 101/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2857 - acc: 0.9125 - val_loss: 0.4009 - val_acc: 0.8375\n",
      "Epoch 102/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2820 - acc: 0.8875 - val_loss: 0.3168 - val_acc: 0.8375\n",
      "Epoch 103/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3143 - acc: 0.8875 - val_loss: 0.2544 - val_acc: 0.9125\n",
      "Epoch 104/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2011 - acc: 0.9250 - val_loss: 0.2449 - val_acc: 0.9125\n",
      "Epoch 105/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1508 - acc: 0.9250 - val_loss: 0.2294 - val_acc: 0.9125\n",
      "Epoch 106/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2883 - acc: 0.8875 - val_loss: 0.3167 - val_acc: 0.8375\n",
      "Epoch 107/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4374 - acc: 0.8625 - val_loss: 0.3316 - val_acc: 0.8625\n",
      "Epoch 108/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2530 - acc: 0.9375 - val_loss: 0.2685 - val_acc: 0.8750\n",
      "Epoch 109/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2178 - acc: 0.9250 - val_loss: 0.3308 - val_acc: 0.8625\n",
      "Epoch 110/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3242 - acc: 0.9000 - val_loss: 0.4384 - val_acc: 0.8125\n",
      "Epoch 111/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2716 - acc: 0.9000 - val_loss: 0.3651 - val_acc: 0.8375\n",
      "Epoch 112/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2061 - acc: 0.9125 - val_loss: 0.3695 - val_acc: 0.8375\n",
      "Epoch 113/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2760 - acc: 0.9125 - val_loss: 0.3713 - val_acc: 0.8375\n",
      "Epoch 114/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3439 - acc: 0.9000 - val_loss: 0.4251 - val_acc: 0.8250\n",
      "Epoch 115/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2670 - acc: 0.9375 - val_loss: 0.4331 - val_acc: 0.8125\n",
      "Epoch 116/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3132 - acc: 0.8750 - val_loss: 0.3553 - val_acc: 0.8500\n",
      "Epoch 117/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3043 - acc: 0.8500 - val_loss: 0.3253 - val_acc: 0.8625\n",
      "Epoch 118/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1989 - acc: 0.9375 - val_loss: 0.3384 - val_acc: 0.8500\n",
      "Epoch 119/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3708 - acc: 0.8625 - val_loss: 0.4401 - val_acc: 0.8125\n",
      "Epoch 120/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2694 - acc: 0.9125 - val_loss: 0.3640 - val_acc: 0.8375\n",
      "Epoch 121/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.4259 - acc: 0.8625 - val_loss: 0.2982 - val_acc: 0.9125\n",
      "Epoch 122/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2530 - acc: 0.9250 - val_loss: 0.4050 - val_acc: 0.8125\n",
      "Epoch 123/1000\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "80/80 [==============================] - 0s 3ms/step - loss: 0.3314 - acc: 0.8750 - val_loss: 0.4153 - val_acc: 0.8250\n",
      "Epoch 124/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1856 - acc: 0.9000 - val_loss: 0.3206 - val_acc: 0.8500\n",
      "Epoch 125/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1631 - acc: 0.9500 - val_loss: 0.3259 - val_acc: 0.8625\n",
      "Epoch 126/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2973 - acc: 0.8625 - val_loss: 0.2612 - val_acc: 0.9125\n",
      "Epoch 127/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2065 - acc: 0.9500 - val_loss: 0.3164 - val_acc: 0.8625\n",
      "Epoch 128/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1941 - acc: 0.9375 - val_loss: 0.2768 - val_acc: 0.9000\n",
      "Epoch 129/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1831 - acc: 0.9250 - val_loss: 0.3253 - val_acc: 0.8625\n",
      "Epoch 130/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2708 - acc: 0.8625 - val_loss: 0.3716 - val_acc: 0.8625\n",
      "Epoch 131/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1222 - acc: 0.9750 - val_loss: 0.3411 - val_acc: 0.8750\n",
      "Epoch 132/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2219 - acc: 0.9500 - val_loss: 0.2549 - val_acc: 0.9125\n",
      "Epoch 133/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2330 - acc: 0.9125 - val_loss: 0.3390 - val_acc: 0.8750\n",
      "Epoch 134/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1963 - acc: 0.9375 - val_loss: 0.3764 - val_acc: 0.8625\n",
      "Epoch 135/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1589 - acc: 0.9375 - val_loss: 0.3384 - val_acc: 0.8750\n",
      "Epoch 136/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2196 - acc: 0.9125 - val_loss: 0.3215 - val_acc: 0.8875\n",
      "Epoch 137/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1766 - acc: 0.9375 - val_loss: 0.3302 - val_acc: 0.8750\n",
      "Epoch 138/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.0793 - acc: 0.9875 - val_loss: 0.3855 - val_acc: 0.8500\n",
      "Epoch 139/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1604 - acc: 0.9625 - val_loss: 0.2671 - val_acc: 0.9125\n",
      "Epoch 140/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1948 - acc: 0.9500 - val_loss: 0.2515 - val_acc: 0.9250\n",
      "Epoch 141/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.2326 - acc: 0.9250 - val_loss: 0.3437 - val_acc: 0.8750\n",
      "Epoch 142/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.1398 - acc: 0.9625 - val_loss: 0.2945 - val_acc: 0.8875\n",
      "Epoch 143/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.1554 - acc: 0.9375 - val_loss: 0.3876 - val_acc: 0.8625\n",
      "Epoch 144/1000\n",
      "80/80 [==============================] - 0s 2ms/step - loss: 0.3702 - acc: 0.8625 - val_loss: 0.3718 - val_acc: 0.8750\n",
      "Epoch 145/1000\n",
      "80/80 [==============================] - 0s 3ms/step - loss: 0.2436 - acc: 0.9000 - val_loss: 0.3599 - val_acc: 0.8750\n"
     ]
    }
   ],
   "source": [
    "##########################################################\n",
    "\n",
    "# Training time!\n",
    "# Try 20 epochs using batch size = 3, then switch to 4\n",
    "from keras.callbacks import TensorBoard, EarlyStopping\n",
    "early_stopping = EarlyStopping(monitor='val_loss', patience=40, mode='min')\n",
    "\n",
    "train_history = enc.fit(x_train, y_train,\n",
    "                epochs=1000,\n",
    "                batch_size=10,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_test, y_test),\n",
    "                callbacks=[TensorBoard(log_dir='/home/amplifier/home/CNN_stim'), early_stopping])\n",
    "\n",
    "# save the entire model and the weights after the training:\n",
    "enc.save('/home/amplifier/home/NEW_DL/models/EEGnet.h5')\n",
    "enc.save_weights('/home/amplifier/home/NEW_DL/weights/EEGnet_weights.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(80, 2)\n",
      "(80, 2)\n"
     ]
    }
   ],
   "source": [
    "###########################################################\n",
    "\n",
    "# let's predict a TRAINING sammple:\n",
    "decoded_train = enc.predict(x_train)\n",
    "decoded_test = enc.predict(x_test)\n",
    "print(decoded_train.shape)\n",
    "print(decoded_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEWCAYAAACJ0YulAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4xLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvAOZPmwAAIABJREFUeJzsnXd8m9W9/99Hw5JsSd527CTODiGDMMKGAKVltBQ6KKPstrRcuged91Jayu29pZf+2kLpZBUoUFZpoewREhJIAtk7ju043lse2uf3x3meR7ItObJjY0Dn/Xr5JVt6nkdHsnQ+5zuPkFKi0Wg0Gg2AbbIHoNFoNJr3DloUNBqNRmOhRUGj0Wg0FloUNBqNRmOhRUGj0Wg0FloUNBqNRmOhRUGjSUIIcbUQYuW7+Hw1QogPv1vPp9EcDC0KmqxBCPGqEKJTCOEap+vdJISICCECxs8uIcTtQoiKQ7je/eMxNo1mrGhR0GQFQoiZwKmABM4fx0s/LKX0AUXAJ4EpwPqxCoNGM9loUdBkC1cCa4B7gKvMO4UQxUKIp4QQPUKIt4A5yScJIX4thNhvPL5eCHFqqotLKSNSyq3AxUAr8O2ka5wnhNgghOgSQrwhhDhi6PlCiHOAHwIXCyF6hRAbjfuvEUJsNyyRaiHElw71jdBoRkKLgiZbuBJ4wPg5WwhRbtx/BxAEKoDPGT/JrAWORFkCDwJ/F0K40z2JlDIG/ANllSCEOBq4C/gSUAz8AXhqqAtLSvks8N8oy8MrpVxqPNQCnAf4gWuAXxnX1GgmBC0Kmg88QohTgBnAI1LK9cBe4LNCCDvwaeBGKWWflHILcG/yuVLK+6WU7VLKqJTy/wAXcNhBnrIBJSIA1wJ/kFK+KaWMSSnvBULACZmMXUr5tJRyr1S8BjyPITgazUSgRUGTDVwFPC+lbDP+ftC4rxRwAPuTjq1NPlEI8W3DfdMthOgC8oGSgzzfVKDD+H0G8G3DddRlXGM6UJnJwIUQ5woh1gghOoxzP5rB82s0Y8Yx2QPQaCYSIYQHuAiwCyGajLtdQAFQDkRRk/QO47GqpHNPBb4HnAlslVLGhRCdgBjh+WzAx4EXjbv2A7dIKW/JYLiDWhYbLqbHUK6vf0gpI0KIJ0d6fo3mUNGWguaDzieAGLAQFRs4EjgceB012T4O3CSEyBVCLCQpCA34UKLRCjiEEDeifPvDEEI4hRCHA39DZSDdZjz0J+A6IcTxQpEnhPiYEMKX4jLNwExDWAByUALWCkSFEOcCZ43pXdBoMkSLguaDzlXA3VLKOillk/kD3A5cBnwF8AJNqMyku5POfQ74N7AL5VYKMtjVBEa2ENAFPAW0A8dIKRsApJTrUHGF24FOYA9wdZqx/t24bRdCvC2lDABfAx4xzv2s8RwazYQh9CY7Go1GozHRloJGo9FoLLQoaDQajcZCi4JGo9FoLLQoaDQajcbifVenUFJSImfOnDnZw9BoNJr3FevXr2+TUpYe7Lj3nSjMnDmTdevWTfYwNBqN5n2FEKL24EdNoPtICHGXEKJFCLElzeOXCSE2GT9vCCGWpjpOo9FoNO8eExlTuAc4Z4TH9wGnSSmPAG4G/jiBY9FoNBpNBkyY+0hKucLY2CTd428k/bkGmDZRY9FoNBpNZrxXYgqfR7UTSIkQ4ovAFwGqqqrSHabRaD7ARCIR6uvrCQaDkz2U9zRut5tp06bhdDrHdP6ki4IQ4gyUKJyS7hgp5R8x3EvLli3TfTk0miykvr4en8/HzJkzEUI3ik2FlJL29nbq6+uZNWvWmK4xqXUKxraEfwYukFK2T+ZYNBrNe5tgMEhxcbEWhBEQQlBcXHxI1tSkiYIQogrVtvgKKeWuyRqHRqN5/6AF4eAc6ns0Ye4jIcTfgNOBEiFEPfBjwAkgpfw9cCNqz9rfGS8iKqVcNlHj2dkU4F+bGrjm5FkU5eVM1NNoNBrN+5qJzD669CCPfwH4wkQ9/1CqW3v57ct7+NgRFVoUNBrNmPB6vfT29k72MCaUrOl95M6xAzAQjk3ySDQajea9S9aIgsdpiEJEi4JGozk0pJTccMMNLF68mCVLlvDwww8D0NjYyPLlyznyyCNZvHgxr7/+OrFYjKuvvto69le/+tUkj35kJj0l9d3CFIWgFgWN5n3PT/65lW0NPeN6zYWVfn788UUZHfv444+zYcMGNm7cSFtbG8ceeyzLly/nwQcf5Oyzz+ZHP/oRsViM/v5+NmzYwIEDB9iyRXX86erqGtdxjzfZYylY7qP4JI9Eo9G831m5ciWXXnopdrud8vJyTjvtNNauXcuxxx7L3XffzU033cTmzZvx+XzMnj2b6upqvvrVr/Lss8/i9/sne/gjknWWgnYfaTTvfzJd0U8U6fa2X758OStWrODpp5/miiuu4IYbbuDKK69k48aNPPfcc9xxxx088sgj3HXXXe/yiDMnaywFtxYFjUYzTixfvpyHH36YWCxGa2srK1as4LjjjqO2tpaysjKuvfZaPv/5z/P222/T1tZGPB7n05/+NDfffDNvv/32ZA9/RLLHUjDcR0GdfaTRaA6RT37yk6xevZqlS5cihOAXv/gFU6ZM4d577+XWW2/F6XTi9Xq57777OHDgANdccw3xuHJd//znP5/k0Y9M1oiC26GMIm0paDSasWLWKAghuPXWW7n11lsHPX7VVVdx1VVXDTvvvW4dJJM17iOH3UaO3aZFQaPRaEYga0QBwO206eI1jUajGYGsEgVPjl3XKWg0Gs0IZJcoOO3afaTRaDQjkFWi4HbatftIo9FoRiCrRMGToy0FjUajGYnsEgWnjiloNBrNSGSdKGhLQaPRvBt4vd60j9XU1LB48eJ3cTSZk1Wi4M7RMQWNRqMZiaypaAbTfaS7pGo073v+/X1o2jy+15yyBM79n7QPf+9732PGjBlcf/31ANx0000IIVixYgWdnZ1EIhF+9rOfccEFF4zqaYPBIP/xH//BunXrcDgc3HbbbZxxxhls3bqVa665hnA4TDwe57HHHqOyspKLLrqI+vp6YrEY//Vf/8XFF198SC97KFknCtp9pNFoxsIll1zCN77xDUsUHnnkEZ599lm++c1v4vf7aWtr44QTTuD888/H2Hc+I+644w4ANm/ezI4dOzjrrLPYtWsXv//97/n617/OZZddRjgcJhaL8cwzz1BZWcnTTz8NQHd397i/zqwSBV3RrNF8QBhhRT9RHHXUUbS0tNDQ0EBrayuFhYVUVFTwzW9+kxUrVmCz2Thw4ADNzc1MmTIl4+uuXLmSr371qwAsWLCAGTNmsGvXLk488URuueUW6uvr+dSnPsW8efNYsmQJ3/nOd/je977Heeedx6mnnjrurzOrYgqmpZCuF7pGo9GMxIUXXsijjz7Kww8/zCWXXMIDDzxAa2sr69evZ8OGDZSXlxMMBkd1zXTz0Wc/+1meeuopPB4PZ599Ni+//DLz589n/fr1LFmyhB/84Af89Kc/HY+XNYjsshSM9tmhaNzaX0Gj0Wgy5ZJLLuHaa6+lra2N1157jUceeYSysjKcTievvPIKtbW1o77m8uXLeeCBB/jQhz7Erl27qKur47DDDqO6uprZs2fzta99jerqajZt2sSCBQsoKiri8ssvx+v1cs8994z7a8wqUbB2XwvHtChoNJpRs2jRIgKBAFOnTqWiooLLLruMj3/84yxbtowjjzySBQsWjPqa119/Pddddx1LlizB4XBwzz334HK5ePjhh7n//vtxOp1MmTKFG2+8kbVr13LDDTdgs9lwOp3ceeed4/4axfvNlbJs2TK5bt26MZ370Ft1fP/xzbzx/Q9RWeAZ55FpNJqJZPv27Rx++OGTPYz3BaneKyHEeinlsoOdm10xhRy9JadGo9GMRFa5j9xJ7iONRqOZaDZv3swVV1wx6D6Xy8Wbb745SSM6OFklCmZMQfc/0mjen0gpR1UDMNksWbKEDRs2vKvPeaghgQlzHwkh7hJCtAghtqR5XAghfiOE2COE2CSEOHqixmKi3UcazfsXt9tNe3u7TikfASkl7e3tuN3uMV9jIi2Fe4DbgfvSPH4uMM/4OR6407idMDzafaTRvG+ZNm0a9fX1tLa2TvZQ3tO43W6mTZs25vMnTBSklCuEEDNHOOQC4D6pZH+NEKJACFEhpWycqDFZMQVtKWg07zucTiezZs2a7GF84JnM7KOpwP6kv+uN+4YhhPiiEGKdEGLdoawSTPeRjiloNBpNaiZTFFJFi1I6C6WUf5RSLpNSListLR3zE2r3kUaj0YzMZIpCPTA96e9pQMNEPqElCrp9tkaj0aRkMkXhKeBKIwvpBKB7IuMJAC6Herk6pqDRaDSpmbBAsxDib8DpQIkQoh74MeAEkFL+HngG+CiwB+gHrpmosZjYbAK306ZjChqNRpOGicw+uvQgj0vgyxP1/OnwOPWWnBqNRpOOrOp9BOaWnFoUNBqNJhVZJwruHL0lp0aj0aQj60RBWwoajUaTnqwUBW0paDQaTWqyTxRydKBZo9Fo0pF1ouB22nXxmkaj0aQh60RBxxQ0Go0mPVkpCtp9pNFoNKnJPlHQKakajUaTlqwTBbfOPtJoNJq0ZJ0oeJx2wtE4sbje0k+j0WiGkn2ikKNesg42azQazXCyThT0lpwajUaTnuwVBZ2BpNFoNMPIOlEwd1/T7iONRqMZTtaKgnYfaTQazXCyTxRytPtIo9Fo0pF1opDvcQJQ29E/ySPRaDSa9x5ZJwoLK/zMKc3jvtU1qB1B4ZUdLbT0BCd3YBqNRvMeIOtEwWYTfO6UWWw50MPamk6e39rENfes5d7VNZM9NI1Go5l0sk4UAD511DQKcp3c9sJOvvfYJgAau7WloNFoNFkpCp4cO5cfP4M11R0MRGJMLfDQGghN9rA0Go1m0slKUQC48sQZzCzO5eYLFrN4qp9mHVPQaDQaHJM9gMmizO/mle+cjhCCTfXdvLmvY7KHpNFoNJNO1loKAEIIAMp8Lrr6I4SiunZBo9FkN1ktCiZlfheAjitoNJqsR4sCypUE0NyjRUGj0WQ3EyoKQohzhBA7hRB7hBDfT/F4lRDiFSHEO0KITUKIj07keNJR5jMtBR1s1mg02c2EiYIQwg7cAZwLLAQuFUIsHHLYfwKPSCmPAi4BfjdR4xmJMp+yFFq0+0ij0WQ5E2kpHAfskVJWSynDwEPABUOOkYDf+D0faJjA8aSlOC8Hu03otFSNRpP1TKQoTAX2J/1db9yXzE3A5UKIeuAZ4KupLiSE+KIQYp0QYl1ra+u4D9RmE5R6XbTomIJGo8lyJlIURIr75JC/LwXukVJOAz4K/FUIMWxMUso/SimXSSmXlZaWTsBQVQaSdh9pNJpsZyJFoR6YnvT3NIa7hz4PPAIgpVwNuIGSCRxTWsp8WhQ0Go1mIkVhLTBPCDFLCJGDCiQ/NeSYOuBMACHE4ShRGH//UAaU+ty6fbZGo8l6JkwUpJRR4CvAc8B2VJbRViHET4UQ5xuHfRu4VgixEfgbcLU0Nzl4lyn3u2jvCxOJxSfj6TUajeY9wYT2PpJSPoMKICffd2PS79uAkydyDJlipqW29YaoyPdM8mg0Go1mctAVzQZmAZvOQNJoNNmMFgUDs/+RrlXQaDTZjBYFA13VrNFoNFoULEq8OQihRUGj0WQ3WhQMHHYbxXku3RRPo9FkNVoUksj3OOgZiE72MDQajWbS0KKQhNftpDekRUGj0WQvWhSS8LrsWhQ0Gk1Wo0UhCa/LQW9Qi4JGo8letCgkkedyaEtBo9FkNVoUkvBpUdBoNFmOFoUk8lwO+kJRJqknn0aj0Uw6WhSS8LodROOSUFR3StVoNNmJFoUkvC7VNFa7kDQaTbaiRSEJSxR0BpJGo8lStCgkkactBY1Gk+VoUUjCp0VBo9FkOVoUkvC6tftIo9FkN1oUkjDdR31hLQoajSY70aKQhOk+CmhLQaPRZCkZiYIQ4utCCL9Q/EUI8bYQ4qyJHty7jWUp6JiCRqPJUjK1FD4npewBzgJKgWuA/5mwUU0SuTl2hNCBZo1Gk71kKgrCuP0ocLeUcmPSfR8YhBB4c4b3P2oNhFhb0zFJo9JoNJp3j0xFYb0Q4nmUKDwnhPABH8heEF738PbZv315N5/90xp6ghEA6jv7uf6B9dqi0Gg0HzgyFYXPA98HjpVS9gNOlAvp/cOOp+EXs6GzZsTD8lyOYdlH79R1EYlJVu1uA+Dhtft5ZnMT2xp6Jmq0Go1GMylkKgonAjullF1CiMuB/wS6J25YE4DLB/3t0LFvxMO8Lseg7KNgJMb2RjX5v7qzFYDntzYD0NEXmqDBajQazeSQqSjcCfQLIZYC3wVqgfsmbFQTQdFsddtRPeJhXqN9tsn2xh6icYnf7eC1Xa3UtvexszkAQHtfeMKGq9FoNJNBpqIQlWqTgQuAX0spfw34DnaSEOIcIcROIcQeIcT30xxzkRBimxBiqxDiwcyHPkp8lWB3ZSQKybGCjfu7APj8KbNp6gnym5f2WI+1906OKMTjkmjsAxnS0Wg0k0ymohAQQvwAuAJ4WghhR8UV0mIccwdwLrAQuFQIsXDIMfOAHwAnSykXAd8Y5fgzx2aDolkHjSkMDTRvqu+m1OfikuOmA/DY2/UcXuHH73bQMUmWwq9f2s2n73xjUp5bo9F8sMlUFC4GQqh6hSZgKnDrQc45DtgjpayWUoaBh1CWRjLXAndIKTsBpJQtGY98LBTNHrWlsKG+i6XT8in3uzm8wg/ARxaWU+x10dY7OTGFva291LT3T8pzazSaDzYZiYIhBA8A+UKI84CglPJgMYWpwP6kv+uN+5KZD8wXQqwSQqwRQpyT6kJCiC8KIdYJIda1trZmMuTUFM1WgeZ4eteLKQpSSnqCEapb+1g6rQCA0w8rBeCsheUU5+VMmqXQG4oyEIlNynNrNJoPNpm2ubgIeAv4DHAR8KYQ4sKDnZbivqGbHzuAecDpwKXAn4UQBcNOkvKPUsplUsplpaWlmQw5NYUzIToAvU1pD8lzOYhLCEbibKlXCVZHTFdDuvbU2dx20VIWVfopmkxRCEYJR+PE4novaY1GM744MjzuR6gahRYAIUQp8CLw6Ajn1APTk/6eBjSkOGaNlDIC7BNC7ESJxNoMxzU6kjOQ/JUpDzHbZwdCETbUqyDz0mn56vS8HD519DQAir0u3q7rmpBhHgzTvTUQiVm7xWk0Gs14kGlMwTbE39+ewblrgXlCiFlCiBzgEuCpIcc8CZwBIIQoQbmTRnb6HwqWKKSvVfC67AD0hWJsru+mqiiXgtycYccV5+XQ2R8mPgmrdbOOYiCsXUgajWZ8yXSZ+awQ4jngb8bfFwPPjHSClDIqhPgK8BxgB+6SUm4VQvwUWCelfMp47CwhxDYgBtwgpWwfywvJiPzpYHOMGGz2ulRSVW8wyrbGHhZP9ac8rigvh1hc0j0QoTBvuGhMJAGj3UZQxxU0Gs04k5EoSClvEEJ8GjgZFSv4o5TyiQzOe4Yh4iGlvDHpdwl8y/iZeOwOKJgxoijkGZZCY/cAte39fOaYaSmPK/YqIWjvC1OYl8PTmxo5ZV4J+Z4RM3UPGSml5T7q15aCRqMZZzLeZEdK+ZiU8ltSym9mIgjvWQ6SluozLIV1tZ0ALKrMT3lccZ4LgI6+MPs7+vnyg2/zl5Ujt9AYDwYiMUyPlc5A0mg0482IoiCECAghelL8BIQQ789ucGYBm0wdCzAthbf2qVbZCytTu48sS6E3xO4W1fZi1Z62gz59PC4Pye2TXFinYwoajWa8GVEUpJQ+KaU/xY9PSpl6tnyvUzQbQj2qOV4KzOyjLQe6Kc7LocznSnlccV7CfbS7uReADfu7LH+/yT83NvDkOwesv+9bXcPpt7465gB1T7IoRHTrbo1GM75k3x7NZgZS+96UD5vuo2hcsrDSjxCp9xIyg8vtvWF2tyhRiMUlb1YP3oznnjdq+OOKhLtqe2OApp7gmJvpJVdbD4R1/yONRjO+ZJ8olC9WtwfWp3zY7bRhM3QgnesIwGm3ke9x0tEXYndLL8tmFOJ22lg5xIUUCEZo7glafzcH1O8NXQNjGv4g95GOKWg0mnEm+0QhfyoUzoKalSkfFkJYBWELK0b2kBXn5dDWF2ZvSy+LKv0cO7NoWFyhNxilvS9MKKom8OYe1S+psXuMohBKuKe0KGg0mvEm+0QBYOYpULsqbQ8kUxQWjWApgAo2b2vooTcUZW65j1PmlrC7pXeQZWAWmrUGlBiYjzV0BYdfMAMGxRTCOqag0WjGlywVhVMh2AXNW1I+7HU7cDttzCrxjniZorwc9rX1ATCvzMvJc0uARBZSPC7pNSbu5p4goWjM6pc0ZkshqGMKGo1m4shSUThZ3aZxIRXk5rCwwo/dljrIbFKUl8hMmlvmZWGFH7fTZu3d3BeOWpmvTd0hy1qAsVsKZqDZYRPafaTRaMad7Oymlj8tEVc48fphD//sE4uxpck6SqbEqFUozHVSnJeDEILC3By6B5TfP3mv56aeIFN6lIg4bIKGMccUoridNlwOu3YfaTSacSc7LQWAWaemjSvML/cxt2xk1xEo9xHAvDKflbqa73GmFIWWnqAVZF5Q4aNxjJZCIBjF63KSm2MfZimEojH+tKKaiN6qU6PRjJHsFYWDxBUyodirVv5zyxMC4k8SheRMoaaeoBVkPnJ6Ac2B4Jgm70Awgt/twOO0MxAZfP7qve3c8sx21lRPXE9BjUbzwSa7RUHYYOPfDn5sGootSyEhCsmWgpkplOOw0dStLAWnXbCoMh8pGZSllCm9oagRCB/uPjKf17RINBqNZrRkryj4K+Coy+GtP6Wtbj4Y88t9VBXlctKcEuu+VO6jOaVemg1LocznprLAA0Bjd2ai0NkXps8IMPcGo3hdDjwp3Efm841FbDQajQayWRQAzvgR2HPgxZvGdHqpz8WK757BYVN81n3JomCmj84t89LcE6KpO8iUfDeV+W4g86rmy//yJj9+aqu6ZkiJQm6OfVhDvB6j71JylpNGo9GMhuwWBd8UOPnrsP0pePhyuPuj8M4Dh3TJfI+T/nCMSCxuNcebW+plIBJjT2sv5X4XFYalkElaaiAYYWtDDzuaeoy/k9xHQ2IK77alsL62g031k7Mladbx0GXKqp1INjwID16StoOwJjvIblEAOOkrMOUIaNyk9ll4/kcQ6h3z5cxNdroHIgSCUew2wcySXECt4Mt8brwuB363I6MCtq1GzcP+DnWsCjQ7VaB5SEyhx7BQWt4lS+GHj2/hmrvX0t0fOfjBmkOj+lXY/+bEXT/YA8/9CHb9GwKNE/c8mvc8WhRy8uC61+Ebm+Civ8JAJ6y7a8yXSxYF09VTke+xHi/3K9dRZYEnI0thy4Fu63rd/ZHB7qNJjClIKant6KO9L8z/vbBzwp8vq4lFINyrJu6h1L0Jb9x+6M+x5k4YMDr8Nmw49Otp3rdoUUhm+rEw+3R447cQGVtxWbIo9AQjeF0OphhCAFDuV2msShQO/hybDVEA2NUSIC5Jyj4aKgqGpdATQk6wC6A1ECIYiVPqc3H/mlpLvADY+W/480cgqmMb44IpBqEUovDyzfDCf0HkEBYC/R2w+naYc6bKyGvUopDNaFEYyvLvQl8LrL93TKf7h7iPfG4HZf5EOwxTICry3Rm5jzYf6GaqEYMw22ekyz4yU2DDsbgV7J4oajv6Afiv8xZSlJfD/z67Qz0gJbz0U6h/C5o2T+gYsoagEbcZain0NKqqfBmHtl1jv/6aOyEUgLN+BiWHaUshy9GiMJSZJ8P042HdX8YUcDMthZ6BCL2GKLiddgpy1f1lSe6jzv7IiFtqBoIRqlv7+NgRFUBCFHxG8VokJgcVwAWCEWsviImuVahrV6KwuNLPx+e66WisUQ/sfQlatqnf0+xZoRklpigMtRS2/QMwPqMt29Vt606493zo2p/59evfgsqjoHwhVB6pLYUsR4tCKo64WK28zMltFAwKNIci+Nzqb9NCMN1H5ur/QFd/2muZQeYT5xST73GyrTEhCrk5ai/p5P2eewaiVBWpoHZLYGLjCnUd/QgBUws9fLblVv4e+SqxPa8q15t3CnjLoX7dhI4hawgarrmhlsLWx9XK3uaEVkMUtj8F+16DJ66DeIYNE7v2Q0GV+r3iSOhtVlaIJivRopCKw89XvtUtj4/6VEsU+hOWAqgAc26O3dqrYVqhEgUzq8hESskL25pp6g5afvolU/OZXuRhZ3MAAK/LidupRCHZhRQIRqyeTRNuKXT0U5nvwWWDGT1vkytC2P52kcqSOeE6mHasthTGC1MUQj2JXl3d9Sob6YiLoHgutBjuu/p1YHdB7UpY9euDXzseV9cyRaHySHX7frUWuusPKXtQo0UhNd5SmLUctj4xahdSjsOGx2m3YgqmCBw/u4iT5pRYjfOmGyv6/Z2DLYX1tZ1ce986zr99JU9vbqQi302J10VVUS7hqJoQvC7lPgIs91M0FqcvHGOOIQrvhqUwvcgDLdvJiQa4OXI5wYK54MqHY66GqUdDx14VxNQcGgNmLYiEsFoYsPUJdbv4U1C2QFkKUipRWHIhLPwEvHILtO0e+dp9rRALJURhyhK1IBrnuEJLIMj1D6y3kiEGcf+Fh1wfZPHnj8Ar/z0+18pStCikY9Gn1KTWtGnUp5pVzSrQrCyH60+fy5+vWmYdU+p1keOwsb9jsCj86fVq8j1OnHYb79R1sWRqPgDTC3OtY3xuFWiGhKVg7rNQ7nPjcztoGcFSiMbi/Mf96znulhe5/oH1PLtl9K6C2vZ+ZhTlQd1qAJ6LH8ubH3oIvvIWeAphqvFaG94e1XWf2tjAna+Ore3IB5ZgUmaX6UKqfUO5jopmQ+nh0Fmr4gr9bTBtGZx1M8SjynIbia46dZs/Xd3m5EHJ/HG3FNbu6+SZzU2WS9QiEoQ9L6j6iJEI98Mz34XfHJUkkkMIBSDQMKbvrCaBFoV0HP5xsDkSK7JRkO9x0tobIhyLW+6jodhsgmmFnkHuo31tfTy/rZkrTpjBP75yMh9fWsnFx6ovq2lZwBBRMCwFs0bB53ZQ5nOltRQCyWc+AAAgAElEQVSklNz0z638e0sTiyr9vLWvg28/snFUKaz94ShtvSGqinOhbjUxbwX1soSmPlSVOKjAJQIOZC4KUkp++dxO7nljX8bnZAXJomAGm/vaVP8uUJYCEjYYq+2py9Qk7ymCxo0jX7vbEIWC6Yn7KpaOu6XQ0a92HOwZmhUXaFC3rSNkT3VUwx9Pg7f+oH5v3pr6uO4D6jbJOnpxWzOPra8f67CzEi0K6cgtglmnwfZ/jvrUfI+TA51qsk8nCqBW/8nuo7+srMZps3HlSTMo8br47aVHcebh5erYJFHIS+E+MlNQfW4n5X532pjCXatquH9NHdedNoe7rzmOr5wxl75wjHZjm9BMqDOsm6pCD9SuhqoTADH4Od1+KD1sVHGFrQ091HX009EXnvA6iwkjHleVweOZjhtMWhmblkJ/u5r0QVkKABsfAmculC0EIdTkfrBVs5mllD9EFHqboLd1fMaPauoIg/cYV3cYotCxVxXppWLNnWqc5/9W/d2+J/VxPcbk39tkvU93v7GP372a5nhNSrQojMTs09QHcJRfDr/HSX0molDksdxHXf1hHl1fzyePmkqZzz3sWDOryO204bTbEqIQGWwp+D0jWwp3r9rHSXOK+e7ZhxljUNet60ifBTUUMx11Tk4nBBqwzzyZwlzn8OeceozycWc4wT+9WbmxIjFJIDR+u8rtag5wz6p3yfpo2qQKwTY9PH7XTGUpDHRAbrH6vWi2auzY36YsNLvxmas4Apq3QXQEwe/eD+58JeImxXPVbUf1uL0Ec2/yYfUzpijEo9BZk/rk+nXKJXbkZSqI3p4mTmJaCmAJR2dfhO6BpM/S1ifgif9IubmWRjGhoiCEOEcIsVMIsUcI8f0RjrtQCCGFEMvSHTMpTD9B3Y6y50y+x2lN1l6XM/3lC3PpCUbpHojw5r4OgpE4Fy6blvLYygI3QiSuNzSmYHZI9SdZCkNX232hKPWdA5w0pxibUdBgis3Q2AY9jfD0t1XbjyGYAjKjz3BNVJ1Amc89vOfS1GPURLX7hbTvgYmUkmc2N1r7Ynf0Zm65HIxH19dz0z+3vTs70u19Sd22jePqNNitAvjm77Go8qubomB3qDgAqMnTpGIpxCPQuiP9tZPTUU2KZqvbzvET0s507iNTFEDVWAwlElRW17RlYLND8Zz0re67k9xEpij0h+kZiCDjcZWN9ferYeODKu1Wk5IJEwUhhB24AzgXWAhcKoRYmOI4H/A1YAK7fY2RyiPVCmz/mlGdZqalwsEshcSE/HZdJ067sALLQ3E57FT43fiN65mWQv+QmILf7aTM7yYcjdMzMHi1vbtFperNK0+0+p5WmEIUQgF48DOw9s+w56VhY6nr6Fe1Ek1rweWHsoWU+V3DRWHxp6F8CTx8Gex4Ou37AMp1VNvez4cPLwMYlTvrYJgN+ya6yhuAPS+r23QujrEw0JXw+Qe7DXeSTIgCQOkCdTs1SRSmLFW3I8UVuvdD/hBRKKgCBHSMIAqh3szrIEhYCj1Ds496GsBhWMZtKUShaZMSNvN1Fc8ZwX10AHJLVPaUEVfo7A8TjsWJvPkXeOHGxPtkBtg1w5hIS+E4YI+UslpKGQYeAi5IcdzNwC+A997OMA6XMsf3vzWq0zIWBWNCru/s553aLhZV5lv1B6mYUZxHfu5gSyFouY/MmIJyHwE0dA+wrqbDijvsMuoc5ieJgifHTpnPpVb/Uqov6aOfU24HYbeCelJKNtV3sa+tT2UeFecialep6m+bnTKfm9ahjfg8BXD1P1UX2oevgKb0W5+aVsIlx6kJqmM8RcEQg66J7uYaCqgFhD1HrbLT+chHS7A7sZoP9ah4Aqi4l8mUxYBQ9SEmRbMhx5s+riClmhyTg8ygPvf509O7j6IhlQX0xm8zfgkJS2FoTOEAFM4EX2Xq9Nn6terWtICK5yqxiqVwL3bXQ9EsKJgB7bsJRmIEjfbysX0r1P0XGs0uu0dR8Z1lTKQoTAWS3/l64z4LIcRRwHQp5b9GupAQ4otCiHVCiHWtreMX/MqI6cdDwzujajiW70kIgW8k91GRKmCrbutj04EujqoqGPG6P7lgET/7xGKAYYFm88vmczusTqwX/2E1F/5+NX9Yoczt3c0BXA6b5TIyqSrKpbBhBfzPDLjtcNj9PHzs/6DscGjeytqaDi7+wxrOv30VZ/zyVV7b1cpSX0BVfc85A4Ayv4vW3hSN+DyFcMkDIGMjpkeurm7n6KoC5paqOouOvvErvjNFoXsgPLJ/fQzE45IVu1rV665ZqXzjiz+tbsdrNRrsBm+ZqlwOJotCkqVw7LVwzb8TGUkANpuqO2hMIwoDnar7av704Y8VzUzvPqpdpfqDNbyT8Uvo7DO3qB2afdQI/koomZfafVS/To3PzGornqcsh67a4cd210P+NHWttj2WEAHI7gYlrAUz1B2pztcAEysKIsV91owhhLABvwK+fbALSSn/KKVcJqVcVlpaOo5DzICqEyAWHlXetrmah5EthXyPE5/LwQvbmglG4hxdVTjideeX+1hUqdxLQyuaA8EIuTl2HHYbs0vz8DjtHDZFbRe6ak8bALuae5lT6rX89tZLLMrlnM4HVbDxo7+EL7wEy65Bli0kUPsOn/n9amra+3jwqG38+qx8LjxmGleX16iT53wIgDKfi0hM0plqNe6bAv5pI9YstPWGmFrgodir9r0eV/eRmZn1zp/gF7PS57mPgTf2tnPlXW+xck+bcrU581RAFA5eOJYJUip3kbtA/X9CaUTB5YUZJw4/f8oRyiefytVjrpaHWgoAhbPSu492PaduRxFzGDHQ7K9UmWptu4cnJdSvU7EpEzMIPjSuIKWyOvxTlXC076GzN7GwsPc2qMdcXpW1NZreUONJxz5VU3IwQgF4/Itj3ir4UJhIUagHkj9t04CkqBI+YDHwqhCiBjgBeOo9F2yedpy6HUWwucCTY/3uHUEUhBBMK8rlnTo1SR09Y2RRSMZuE7gctoSlEIxYAlTidbH1J2fz9+tO4twlU9iwv4uBcIzdzYFBW4eaLMzr4Si5jeiRV8Jx18K0ZURjcZ5sLMQXbuGKpX5WfGEWJ23/GRe03MkvP7OUeb1vga/C8tGaGVNpK6mnHjVizUJHb5iiPBe5OQ7cTtu4Bpq7ByLME/XM2XSrWhl3jN8Xzexd9XZtlwoyzzwFyhepB8cjrhANqkWJO1/Fb4I9iSrxZPdROiqWQqQvtSvInBiHBppBuZ7624b3W5ISdj2rfu/Yl1Fm2UA4lkiISBaFWAQCTWqyLpmvqrWTA8+9LaqOItklZonCEMHt71DvVf40KJkL0QEGWtXkayOOs79FiY/5eicrpvDoNfDPrx/8uPp1KoPtkSvH3MZ/rEykKKwF5gkhZgkhcoBLgKfMB6WU3VLKEinlTCnlTGANcL6U8r3VRc1bCkVz1GYmGWK2zzbTR0diutEDqdzvsvZuzpTk9tmBYBS/O2GhmNlFJ84uJhKTvLqzhYbuIPPKvcOuc3yvCiY3Vp1n3bdqbztPNCh31k9PAHfdK+qBnc+oyaD6VZh9hsqHB6s9eNqeS5VHq5VlirYXwUiMvnDMshKK81zjGlPoDw5wm/N3CTt1HCcEs3K8rmaXmnjnfEhN1p6i9KmTo8FMR/UUKGEIdicsBU8monCEuk1VjGZYCgF3BS1D40FFs9TtUGugbbdKHS2eZ1gtB29jYrpxhEgkRABGBpBUiwszeyq5BbjZUDE5oyqvWLkkhwquafXkT1NjA6JGQVwx3dhkNEkUpk9MTGGgC576qhKzVEQGlCsvk8+feUzzFlX38i4yYaIgpYwCXwGeA7YDj0gptwohfiqEOH+inndCqDpBBRAzzLYwA80+d/p4gomZgXR0VaHVFylTPEkb7SRbCskcO7MIh01w32q1appfNsRSkJI5Tc+wLj6f6ljCNbe7OcD2uFpBipZtKqsmtwQQaqUz0Gm5jgAruD1scjGZerS6TeGHNl1FxXlKFIrycsbNfRSrfZPbY7ewxFbDi7NVVnRP0z6uvvutcdlGtNmwjLob1CQVKzmMG/+xhX7frPFJSzVdXWYtgek+cuZCTu7I54IqbHPmpbZ0u/aDM5ebX2rioj+sHhwPKjREYagLabfhOjr+S8bjB69lMAW+Mt8z2FIwO7H6pyr3EajEhu3/UhPhK7eorgIVSwdfsHjucFHoUTUKT1bDhgH1ObZ1qGMqREfieUDFFbr2j741/lt/GnmC3voEvH0fbPxb6scbN6nYWibpsF11IOzIE66HdX+hZ/U9oxvrITChdQpSymeklPOllHOklLcY990opXwqxbGnv+esBJN5Z6kvYs3rGR1uiYIrvevIxLQUDhZPSIXHaac/2VLwDBehPJeDI6bls7parS6tzKO1f4a/XQqr7yC3axdPxk4elJa6u7kXmVem/NYNG2DfCjj8PFjwMdWaGdQudQYJ91EaS6HC6L6ZIq7Qbvh+i71KWIrycsbHUlj1a+x3n8UCWx0/jlzFKv854MqnrX43r+5sZWujWoXH4pJvP7Jx8O5xBrXtfexpCaR9CtNSyA2qL/rmnlzuW11LDZXj4z4yLYWh7qPkeMJI2B1QdbzqlTSUrlrIn05NxwA17f3UtCelJZuWwtBJf9dzULZINYyEjOIKpqUwsySXQChKLG5MxsZEjr9StVp3+dUe6Q9fpiZgKeHUb4PTM/iCxXOHC65RuHbrmn7u2zQAOT5c3cpNmBCFJPdRdEC1ChkNb/4eVt+RPh5hpl3veCb142Z1f7j34J1cu+ogfyotx/+QlbFFeJ/7Jmz6++jGO0Z0RXMmzD8bcnyw2finSKl6taRZaSQshRSisP1fKj3TOHdBhaokPXFOii/5zn+PuBLz5NgJJtUppLNMzGt7nHarZTcbH1KuoOd/hLQ5eEGcOEgU9rT2MqfMp/zj255U/t45ZyZWiFOOUK61pLH43A5a04mCp0B9mQ+ktxSK8kz30aGJgpRSrXo3PUJoyjGcEvo198bOVkHwguk4AqrIybQUWgMhHnu73qqoTubGf2zl4j+sSVvj0BIIUeLNYYox8Txdo+5vck4d1G5hVIT74J7zVH2BJQqFShhMSyGTeIJJ1Ulqb5BkV08sqtwzpfMt627l7qTMPpcP8koHT/qhALJuNXLeWUYWjxiVpTCjOA9IpE9b8QN/pfItHfsF1d31s4/ADxvg+jfgjB8Ov2DxXNUzKXli7d6PtLs4EMnjQHcQSufjC1STm2NnVo7xHpqWgplt1T0KN2KgyRB5mdoSCPaoxVKOT1llqQQnueXLwayFrjoomEEgAl+IfIf9vqXwxBfH1M5/tGhRyASnBxaeD9ueUqmpK2+DO46F9fekPNxsnz1sko7H4cWb1EYohhvlhNnFrPzeGSweWrQWDasg0+NfSis+Hqd9UAAvXabTibNLAJhb5rViDXTWwJGXw2f/jrj4AbyF5ValspSS3c0BtTdD+WKI9KuahdmnwYyT1X4Ty64Z9jyp2mtEYnG2NnTz3NYm4pVHJyyF5q1q8gPajaByiTfZfTT2lNRvPbKR7z/wOjRvpXPq6QygrJiu/jAUVOHuU6vKLrN+YUA9/77WvmHXauweoL0vzG9fSh0faOkJcvLcEqbauwjZcnlim7Iq9tuMCWgsQe32Pcoq3f4vq+/Rlg5JL7mJlNRMLQWAGScBcrALaffzSrSWXmpZdyv3DJnIimYPch/11m1ExKOsCs8Bp1tNsiMVuBmYfY9mFit3l1Wr0HNAFa55DCv5wz+Gi+5VizD7CFa2GWxOjj/0HCDmqwAEjd1BKFtIaf8eCj1Opju6iApn4j0zA+ujiS3VrlK3vkp45/7hbTL2vKgSAj70n0BSMD6ZA+sTrzVwkM7EXXVQUEUgGCWIi99V/rcS90jm7WjGihaFTFlyoVqlrfkdvPq/qkDp2e8ntkEcQr7Hae2lYLHnxUTwcWeiVfC0whS+4Y5q9SGrf0udlwIz0CylHBZoTuaYGYXk2G0J11G4T/XRL5oF88+Cw86hqijXEoXW3hA9wSjzyryJTJppx6qVqhBw8V9h2eeGPU+Zzz2oZfeG/V0s/cnzfOw3K/nSX9ez1zlPfRke/xLceRK8cbt6qYYAmJZCkTeHYCQ+4lalI7GtoQf7gTcBSUuhSmfMy1F7XJA/HX+oEZDW6t/Moa9pHy4KrYEQdpvgnjdq2Ns62OSPxyWtvSEqCzzM8wRoiBXQZrjCaqRRLzCWuIIZqGzaZFkK3/xHLasPRJTF1tc6OlGYeoz6vJoTG8Db94K3nN6qD9EfjuGwCd7Y255w7cCwtNS+OlUZvbrPqBkompWZpdAfQYhESxWrVsGsURhlLI0ZJw3vYNx9gHCecg81dg8QL1uIN9bNLE8flbYOOmzFqm4DEim4o0lLrX1DFQKeeaNyu9WuHPz4jqdVzO3YLyix3DmkFXh/h7K65p2t/g40pX+uaEi9NwVVVkv8hn47XPVPOOryzMc8RrQoZMrM5ZBXBi/9RK2SvvCSMrH/fk3KlLFrl8/mwmOG9DFa8zuVaTHt2OEfmqGY2ys681TALYW14DYCzaFofMQ23Z4cO3+88hi+dqaxwjJXSIUzrWOqinKpa+9HSskeox3GXNN9BDD3zJHHC8NaXby4rZlQNM7NRsHdHoeRYbLpITVJGW0N2nvD5DhsloiaAeexWgstgSBz+zeBPYcDeaqzSlVxnqpoLqjCHe/HT59V4dxtWgptfcSTJsVwNE5nf4TLj6/C47Tz82cG9xDq7A8TiUnKfC6mO7poiBfidtpYVOlnb6wcENA8hm6ppmuhcZNlKdT0OWiJqJgL3fWjEwWnWwlDrdr7gp4GZSkceRnNfUp4T5tfSiAYZXNyXKVollrNR9X/QTZvoUfmsq4jL/F4JjGFvjAFHicFuer/agWbexoSLp0ReGNvG9946J1EINxbBvPPUW4cs2q8u55+txKrSEzS7VeftcWOeqbQQast6f1y56uf0VgKNUb1/qJPqD5U79yfeCwaVu/nYecqC+ewc2Hvy4PnBTMde8HH1O1Q91HjJrVYikWMHk4SCqroM0ShNRBKiNoEo0UhU+wOtcsVwEduVql+n/y9mrzf+uOwwz9/yiw+vLA8cUfzNqh+RdUBHH6+mixG+lC27FA9XM76qXI1bTdi81KqgNOa35NrWApWM7wUgWaT0w8rs3y6VvGMWd2J6ocUCEXZ09KbJApe1T/n7P9WK6CDML0wl4auAWt1s662g0WVfi47rgqnXbBZzoFln4dLH4aqE61xtPeFKc7LsbKvivLU5JdJXCEai/Ph217jgTfVtcyJ/Ei5jVjFUXSGVZHfjKJcw31k7E8h2iwxMMUhFI3TmJQ9ZYrSggo/FxxVyZv72gc9tymA5X43xfF2mijkQwvKmOJ30x4SSkhX3wE7U7gSRsK0FAIN0L4X6fAQwUFTyKh/kbG06agdfWF+/u/tg5r/barvIjz1eFWAGe5Tu5zJOBx9hWXZXXCUmpxXJbuQimYD0rIGctq2s11Wscd0sxXNVlZLKH0gHtReCoV5OZYla1kKPQfUIukg/HNjA09uaBic7nz0leq5dz2rVuGBRgKuxPftQI5q6jefOkpkG01yiIjmV2WeltrXrr7nM09WruRFFyjLwBSpujeUF8Gc8A/7qHLzmEV+YMQThErOsLuGWwo7nlaLpQNvJ+YFw30EWBbou4EWhdFwyrfg479WH0iAuR9W/+TVdxy8Dcbr/wcODxxzjfrQwMiTRet2tZI/+mq1w9ajn1PpcE98CR7/Ajz7PWbE6hgIxywfrX+EQrlBmCX+hQlROGtROTahvoB7WnrxuRyU+11qdXLilzMKbJ40p5hoXLJmbzuRWJwN+7s4ZkYhNpug3O/mQCAK590Gh52jntsYR3tvyKpRgIQbKZO01I313exp6bUKANv7QngIskTso7f8OMtFNKNYdaSN+ZUoTBWtlhgkV2HXtCVcSGbQvNTrotTrJhCMDppsmw0BKfM68YRa6bAV85ll0/F7nGriu/Bu1WbikStV9lamJOe573udaI5KRmgIJt6jdP+Pl7Y384fXqq0dzpp7glxwxyp+9LZftd54/IsqJjbrNCiabcWADp/iY2GFn9eTg81mJXHNSpASb/cudsSn094XVoKdLm11CJ19YWa6+5m68de4CKvPazyuUlLNjKAR2NmkRKe6Lcl9N+dMJShv/QkevBhsDvaVJFKk9wc9tFLArNg+CqJt1MeUL7+xe4BH19eProDNdLvNOFndli9RGUR9xnvVvE3dmoWuM09VtRLP/TCRUnxgvSr0dPtVptVQS8EcS93qQaJgLrDa+8JE340uv2hRGB2+crX/cLIP9JRvqX+wuetVKna/AFsehZO+qr7MJXPVh2ZnmtQ1UH1gShcoC+Waf6vWCavvUBlQp3wTHB7O6HiYgXDMyuZIF1MYRmetynPPS2QPlfncnDinmH9uamR3cy9zyryjrps4ZmYhHqedFbtb2drQQzASZ9kMNXlV5nto7EoSzoIZ6ksV7qOjL2xZB5BwH2VS1bxil/piNnWra7cGQhxp24tTxGguPJrugQhOu2CKURjY7lSryWmizRIFM9AMqg+VibmKLvW5KMpT721yPx3TUqhw9iHiUT5/7smccVgZfrdDTXxuP1z+uLJOnv6OtbJsCQTZVD9Cq42+FqMmBOipJ+xUomBZCpDWfdRmvGdmRlF95wBSwot9MwlLO3Lnv9WK9oLbrfcLoMzv5qQ5xbxd25WYfIpmq/9T9avQVUdOrI8dUgVp97T0JlpsHySu0NU7wHcD/0P+m7/kJNtWJdSBBtXDqGA6zT1Bq7HjUFTSgxKD6uREALtDfSf2vaaa5n36z9TkzLMeru8cYHt8OnP63sEpw9RGCpBS8tfVtXzn7xsJeaeqmELzVtVSOzrCSrx2lVrQVRq1Nqbb1RTDzn0qndYUakcOfOoPyhp4+ttqz+jdzytLA9Q8MtRSMBdqpigIO/gq6TUsBSnHt0nkSGhROFRmLVdtfVf9v9SdG0MB+Oc31AS//DuJ+w87V63AUgW7omGVgWK2+c0rhvN/A196TcUyPnwTHH0FR3Q8hy/SSoMx2Y7UZ2kQXbVqpTRk0v/4EZXsa+tjXW2HCjKPEpfDzolzinl9dxvralT647KZaoVWWeCmoTvJx2p+sbrqaOsNU5KXZCkYVkNHX5ieYISH3qrj3jdq+Pu6/YN8/gArjJVtU08Qtj6BZ+0dXGh/jZgUVHsW0T0QId/jpNDwZ+/uyaFfupgq2hLZR30RSn0uPE77YEuhNyEKhcb4zKA0JCbUkrhyK9ny1arX73ESCEbUWHOL4ORvqPiJkf3zxKMP8vpdabcXUZZCyTwrSyZoVwkCAZmUkJBWFNSYmo2xtRqWwO1Xn8aXcn7O96f/FT79Z+vazT1BXA4bfreDBRV+wrF4YsMlIVTDw30rrPbbpijsbgmkr3oewmcDd7EgqM5fZKtVVpTRMVeWL+Zjv1nJr15MvR1nU0/Q2nCpemh22NFXgncKfPRWWHi+ZRW6HDZ2NAXYEa/CH1Yr8gPxQvrCMfYZ/98+T6UK2t95kmqpnSpbyKRmFUw/Vk32kPS6a9RtR7W6L/n7NPUYOO27ajH42v/C0kvhIz9Vj/mmjGAprFHXzZ8GdodlKUDi8zjRaFE4VIRQBTZddan3c37xJ8p3ev7tqiWxybLPKf/k368avkrp2KtM/bLDB99fsTRRGXzil7HJOFfanuWrf1NBrEwqqAH1oUsKMpucs3gKTrsgEpMqnjAGls8rYV9bH09uOMD0Io/VsbWiwENTdzCR3WJ1q6yjoy/MHHuz9SXzuRw47YL2vjB3vLKH7z++mR8/tZUbHt1kiQCoOoON+7tw2ATB7hZ4/EvM2/gLPm1fyRY5i4aBHHqCEfwep9WkcHdLL/WyhJmOdivg2TUQpig3hxnFudakAYlJv9ibk+TSMv5XUtLT0Yzf7cA1YHzBfYYouJ3EJfSFjS/04k+p/PX190B/BxfX/Zgvx/9GbG8al1JvswqmTlEtKvpsRn4/BxcFc8xmG3PTDz+/3EflwpP4V42dcDThhmgJhCjzuxBCWP9zM6YEqKr1UI+q1AWChYeRm2NXq3ezlqF9L9saetjaMLz4T1a/xuXxp1hf9mkoqGKJY796343tSnv882nrDbFmb/uwcyHhOrLbxCD30b62PmRBFXx7h4rToXpc+dwOphZ62NbYw454ovVasyyieyBi/X+bS45X7+8ZP1KxO3Pf53gcHv18oqPvQKdqNTHjlMSgzP0mTDHs2JdwpSVz6rdV99pP/gE+eSfkGDE975TBKamxiJoj8qtUYsHely3RTt6BMG0N0DijRWE8mH+O6o+09s+D769dDWv/BMdfp1YayRTNgk/8TvkanxtSoGOmuZqWQioKZyIWfYJrPa/wqXnK/19RMKR3Uiyq/K3JLaulVO6jpCCzSUFuDsvnKZfSWEXh1Pnq/C0HeizXEUBlvptoXCYCZsaHPtxWzUAkxkU1N8KT1wOqUWBRXg4tPUEeW1/PmQvKePOHZ+Jx2nlxe2KFtXJPG3EJZx5exseiL0EsxGNH/InLwj/gG9Gv0RII0WNYCgVGEH5XSy8HZAmzHO0q8IyKKeTnOpldmjdMFApynbgcdksULEth6+N8Z/MFLPYmNXEz2lb7jdbp1n7EOXlwxEWw9QmiT38XXzxAh/QSf/lnqWtQeltUppvR3iGA+l+MxlIwXVvNPUHsNkFxXg7L55fSF47xdl1iN72WnpBVjT6nVE1ae5JTb2ctV5PmnhdotleQ581nbpk3IRylC5At27nu/vX815PD98sIV68iLgUbDv8WlC9hgahT70vTJiiaTVNQ/V+Uu3G4C8l0HZ0wu8j636zc3cYZv3yV6+5fT0/SpGlahZX5HnY3B9gpE43+GmURnavigMYAACAASURBVH1h6xoNrjlw3etqNV88NyEKHXvV6v5NI3mkdjUgE64fUIs7s0YjFlWWt+lKS8buhI/9EpZeMvh+X7lKNTazk7rrVeD/iIvU3wMd1vezNxjFYdQWtY1jk8iR0KIwHthsKtawf00i6BQJquZYBVVGQUsKDv+4ijOs/TPsX5u4v3Wn+iKWzEt9nskZP8Iuo/zScw9v/uDM4TGFpo3KLN74UOK+gU5lNhcOFwWAi4+djsthG15MlyGzS/KYWqCqpo9J6vpaka/ua+gyvgjeMnB4CLXuI48Bint3qbQ8oyioMDeH57c109Yb5rPHV1Hud3PqvBJe2t5ipSau2NWKz+3g7IWlXG5/kf7KE9loX8gW19GEfFW0BIJ0D0TwuxPpkLuaAhyQJUyRrfSFY4Sjcbr7IxTmOplZnMf+jn4rmNwaCFFqtt4wzu8wYwq7nsdJhNOdW9SqT9jURE4itjOoz88xV0M0iGPr3/lr7CPcFv0MzgNvDtrZrrq1l1v+sUGtFr3llqXQLT3kOGwESGr3kCbQPFQUWgIhynwubDbBSXOKcdgEr+1KWFstgaBKKEBZmlP8bvY0J4mCp9Dype8RMyjKyxksClOWIJu2UN/Rq4rGhhBtq6aJQvy+fJiyhOmygWBfj7IUpiyh0XApRuNycDqswa7mACVeF8dUFbK/o59QNMZru1pw2AQvbm/h/N+utOIn3QMRCnKdVBaoBcgeWYkUdqSw00oBO5sChAwraVASQ/kiZQ1Aouq4+hX1Ha5dpbKFkne0AyMdtwZ66pVVX5TCUkiH16jzMF1Iputo1vLEY8aiqTcUteo7tKXwfuPIy1Tu/fq71d+v3KIK1T7+a9XDPR3Lb1Cbp2x7MnGfmXk0tOfLUErmKvN35zOw+dHhj5t56XWrE/eZftAUlgLAWYumsPHHZ1lun9EihGC5YS2Y8QTAsmKsiUMIKKgi1lHDEbZqbMSVWBmtB4q9OfSGopT5XJxmXO/DC8tp7A6ytaEHKSUrdrdyytwSFvWvZbqtldrZl6qJ3Oei1O+mNRCyVo+mpbC7pZd6SsmN9ZDHAN0DETr7wxR4cphVkkc0LqnvVBNVa6+6FmDFFDp6w2p1X6OKl5bFNqksGm+5VYVrpgYPEoWKI2DqMQw4C7kteiEPx84glDcVXv4pxKK8vKOZC25fxdOrja0zvWVWh9OOWC6zivOQdjdRkaOKqJJdkUmYq8lmy30UtJoV+txOjplRyGs7k0QhyVIAmFfuHWwpgNX4cGtsOsVeJQpNPUEVG5iyBFssyEzRREsgNCzmIztr2C/LlKU1ZTE2JFN6NivXy5QlVoIAwNu1w/cD39UcYH65l9mlXuIS6tr7WVPdwTEzCrnnmmOpae/npR0qW6urP0y+x2ktQELkEC2YTTS3jDg23tmfuH7nUFHorGHtzlpi5uIs0q8K1GpXqS6tziHfh8IZ6jWYQfZUlkI6zDTcwBBRKJyhmm9CQhSCUUp9LvJy7FoU3nfkFau+LRsfgn99C974DRx91aBOoilx56u01h3/SrgSWnao7paZcOKXVTHcv29QVkAyphh01iQ6UqZIRx02pBG2BM2Ez508k+tOmzOoI6tpPViWgjEGe3cdR4mkFhKGGX8SG7nZcRcXHjMNh9F+/EMLyhACXtzezMNr99PYHeQjC8up2vsgzbKALf5TrdV9mc9FS09CFMyJunsgQrdTrcZmiUa6B8J0GSvM2Yb7ZJ/huzYFBsBpt+FzO1T2kbFCDEon8wfeUavFpHz7RD7+kMSDi+/n/838HT14ieBg++JvQeNGWh79Fp+/dx3Ti3KZk2sEeb1l6pofuZlnbcsp9uZQ5ncxYMtLayVEYnErO6rFCjSHKEsS+OXzS9nW2ENLIKgy10JR6zUCzClVVsCgyX3eWQCsDStLYZ7xf93T0kukVBU3HmGvIxaXw9KInT111MXLlKiWqyLGk/peUA9OOYLG7iBCqM+HmVZsEo9Ldrf0Mr/cZ/1v3tnfxdaGbk6YXcyJs4uxicRCo3sgQoEnh8okN2p84QWEZn0YUBX2JoMyeYxx/fyex+nevUZZBQ6P6jPUuDGRippM4Sy10jc9A6liCunwGfUUvUYGUletsjT9U42WJFj1NIFQFJ/bQanP9a7VKmhRGE+WfU4F5db9BU78Cpz3q8zOO/w8NdE0b1HZSh17oWyEeEIyNrt6noFOlbNtIqUSBVNc9q9RtykK18abeeU+vn/ugkSfJVTbD4/TbmVKmWNw9R3gaNtuor5pqApgJQofCTzJFY4XufQw4xqBJkpevoGTprl57O16fvLPbZwyt4RPLMzHXfsKj8WW0xiIWav7Mp+L5kDQiinYbcKq46jxHU3M7uFLjn/R1B0iHI1TkJvDzGJTFFRld7L7CJIa9RlWwl9jH8Eb6VCuv6R8eyumMLSJnr+S11q9LDA2OtpWdBac+BXKtt/L5bbnufdzx3FKhXJvyLwyZU2d/DW2hMoo9roo97tV/6M08YSOvjBSGr2jekNEY/FBlgJgWV2v72qzahSSH59b5qU/HBtUxMf0Ywlc9TIvxI6iKM9lZabtaAzwakchYWnnExUqUNycfF5kANdAM7WyXLnfCmYwYMvjlIjRsbX8/7d35vFxluXe/16zz2QmM9n3pOm+b7TshQqUtlCgHgFR5KCouAseOaCHo++R13NexQ3xoOIRBc+LoChyEF8VRAQBgZYW6N5Cl6Rt0qZNmn3P/f5xP88zSyZpUttk2t7fz6efZp55ZnLnSea+nmv7XbOpb+6iIOxn8YQc1tU0Jcl37zvSSUdPP1OLIlTn69/NL9fUMqC0yKPH7aIwEqDOutFo7tRKwaXWDYjHJfgu+VeU9TncWtdKyOemKDtlXofVtT/XtZNoy1adP6i+QHdMq4HkfIKNHS7a+ZzWbxpBE56DHSJK9BSyy3UOYs41sOQ2p+ehvbuPsN9DfthvPIWTksqz4exPwuofwvJ/1xv2SJh2OSBaAO3Zu/TcBrs7ciQUz4GpK7WMhq0ceWi7Fk4762bdk1BjG4XdOk4cyB7NT/Z3IyKUxAJODBmAWCW+3hbOdm1BVV+gP2gHNkJ/L5M7dBilosVSVX3j57DuIW4ofIfaxk4CXhffunYergMbEDXANt8s6pq7nLv7wkiAIx29DKi4aq0dAvJEizk05yNc4X6Flp26TDQW8pKb5SMn5GVrXQvt1rSwRV1/gzceAaXIyfLpO/HdL9IXyOWhfkvHprc92Sikdu5adPX2s+NgmxNea2zvhmV3sSO2hH/zPER+/wHm5+jNqqYnHnI83KY7vouy/dSRN6RBtzeNWaXZDCh9B93U0ZsUCpxZkk1BxM/Tm+sdbyLRk5iSrgIJOBieBuiEdUVuiIjfw51PbOC2x7eySypY4NPKs0lGwboBqVGWp+By0RCaTJBu3ZGdXUpdSxcl0QALq3I42NrNvgRPcoclWT61KEwk4KUw4mftniZ8HhfzK/QAqOJogPqWLpRSNHcmh49iIS8iQtjvwe0S+gYU1flZg4c4RSvo9Ub4B/dfcas+7SlMXa4Ngssbb0pLxK7e2/2S/no0EhShPK3d5HgKNXGRvlAuXPwlp/y1rbuPsPEUTmJEYMX/gfnvG93rwgVa9mHtT/Td/lkfS55LOxKWfF57C+se0o9t/fzqC3VM1A4lHUlfeTQWlEaD7E9MRlohrLB04ak8y0r4bYJ963D1WlVAdjeplZA917udvCwf37h6nt7sLE2ZQ9kz2dnQRkdPvzYK2fG7X9so2HmFwkiA7rM+zSGVzawN3wAUOdYGsnhCLq/uaqShtZulrvUs33gbPPFx+P3t5AXdHG7thj0v0VRwJntVAV0R61om3Cna/SKtKeGjLXUt9A8oFlbGCPs9NLb3gsvNE7kfwi0K2f0SU7N0+Oilev3R7OrVIZ68LB+FkQC39H0Wrrgn7fW1N42Zpdrg213NRQnXwuUSVs8v5dktB9mc5nm76mzHgWTpisYEeXO3S3ji0+dxy8VTqMoLIcVzCDfpirkkKQorf1XnKnK8tKaI1iQaKJ4DItQ3d1IcDTjzRNYlhJC21WvDNMUScrS9hYWVMSfEWRINUNfcRWdvP739ykk0A05xgYg4fwPV+Vl6XkdCEyIiNIanMNdllZiWneGEzChbmH6YkR0u6uscXT4BtAEJJzSwNe0ZMpzb1tVH2O/VnoIxCqcZM67QnazRcrjoS6N/fcVi3V7/8ve0t1DzN10NkztRG5z6DVqLZdcLULrg+K9/BJRE464+kGScpGKxlg84/A5st8QCK87Sd2JdLU74K/vAGl7/0rK4rtT+dZBdTjCnxBmSY+cUbOx8QtTaJAoifqLRPL7b9w9Uta1nqetNotZc7bMm5lHT2MGejS9zn/de2nJnwFmfgNd+xO2H7mR220vQXEtttq7I6a206tcTPAWP20WWzz0ofLTR2oRnl0XJyfI68f8NPaW0SxbUvkJsoIkWwry0W2/K9mZsh49qu0O0uwbP2YZ4knmmNaPDvh6JiWTQFWZ9A4r/+uvOQc/nhf3khLyDFGFteXO7NHdSQZhbL5nKk58+n6nzzsXdcZACadZNhDaWUWgLljvd8W0xHRbtztNChXXNXZREg0wvjhDwupKSzTsOtFKcHXA29IkF2mDZUvCgq9rqjnQ63enRoJeQz0Ms5CUnFK/Gc97DNgopuY/6wCQAGsjRv8tYhc4JJqgBv1F7hL9ssyRIgjlaGA9Gl0+wsY1CgiJqKt19/Y7QZUHEz5GO3qQekxOFMQqZwqzVegO/8nvDVysNx4V36D+0+5foBpiqc7T3Unm2doUfvV67usu+clyXPlJKYkEa2rrp7uvnwZd28doRvXl1SFD3ZBTNAhS8/pAOic24QudXNjymy/4qz9X17YkCbPvWQdkCiqMB2i2pbTt8ZGPH+OOegp9IwMOjAxdxQAr4jOc35IT0OWdV5yIMMP3FW2giQv3lD8HKr8Gq71DZtYW7+74GwBvu2XhcQnCapR4bTVbEjQS8g8JH62uayAl5KYsFyQ3FN6YDrb3sDMyCmleQ9oN0+nJ5dedhlFIJRsHn3NEPNd3O9hRmWZ6CXeKZ6DWBVr9dVJXD3qZOvG5J2jwBphRGBoWPEtcxiOI5AJwZ3J88jrVpN10SwJUwjKmr0Oq9yJ1DW3cfrV19FEcDeNwuZpdGHe8FYPvB1qSZ4nYfxdkT44n2Euv3bleM2Zv/jOJsx7OA+I1BdUF6o7Dbozf29f0T6ey1Nt4r703qMfjW09u467dWYlkEcifor0dTjmoTKdZ/2017sBVRU7ElLuycAhy7cvBoMEYhU8guhc+u17ICx0r1Eq25PtCndYUqrUqG8sW6usEbhOse0RVP40BZLIBS8B+/28K//XYz1/5sCy0qxE7/DJ1/sWW6Oxt12MuuxHj+bl2Gef6t2rjttcoGbY360oUUJ8TFhwwfWZtfgVW3HwoE+UHfFZzh2kHhYf2eM0qyuSywkeK+fXyt933kFlkf1kU38fNzfse9favpnXMda9oKqcwL4Zm9Gq79WfxaW2QHPfFhMsBruxp5Yv0+VswuRkTi+Ql0r0B9dB40bIWGbUikiENtPew42OZs9Plhn5MbODDEHOxDrd0EvW4qc/VmOJSnANpbAO1VpWpcTSoMs+NgW1LSN3XmRRLFunpnUaA2Jaewm3pXMbkJyXqK53J195epK1vhlKOWWLpU00sibKnX5cb9A1rzaFpR3Cu6cl4pt1w8Jan/xda02lavjYlt+B/44CLuumq2c17cUwiTm+Wjtasv6a57u9Je6xsDk9jblH6QTW1jR3J1lZ1XOBajMPs92pN66lb9OE1I15a4yPJ7nAqxsUg2G6NwqlG9BD7xMqy6BxbeoI/5I7Dybnjfo1AwddyWZicAH/rbHpbPKuKTSyfxtf7rWVNuTXGLVenNH7RRKJ6n5SHa6nU1SNW52rjZ/RfW9DrKFlIcjfd0FET8lhS3fjw4p6A/YLGQl0d6L+CgihFdey+g5RQ+HnyWAyrGnzjT0UwCCMfy+XbftdS/6zvsPNzJxPwsbcxmXjUo0Zid4Ck0d/Ry66PrqcwNceflOmxiewq21HdLgdUc1bCVrLy4jHU8bON3PIXEjbe5o5c91nAgu/LK53FZ0+t6nG7mVC6fW0LY76EgTT/KtKIwRzp6k2Y2H27vIez34PekKZ4I5kC0gnnsSOo7oGkXNarIucsFfce+Vk2npWfAOdc26NOKs2nt6qOuuctqVBuID4ZCJ8Q/t2yqU6IMcYOyxZLDsD2CkM+TVFpt/+4n5Gc5BQdHEvIK6/qqeGBgFY/3L6E2jVHot/pXmjt744KBdtjoWMJHc66GBTfEc2ZpPIXWBE/BNgpjkWw2RuFUxB/R4zJ9cfeZMz+avrRuDLETgBPzs/jmNfO4fcV0br39q1x7zfv1CS4XFM7UlRlV58SHzoOeTeCP6C5fO2luj/Ysme9sDm6XkBPy4XG7yLOUV6NpcgqgN4pufPxUrcK9+3nY/jQceps5nWt4uO8SYuGspLJau6v5UFs3uw63OzHudDjy2cBXntrEwdZuvnvdAmeQUE6Wj6b2Hid5OFC6QP/cQFZOCRPyQry445ATLtB9CvpntNVb/7z1ABd/+y+s+t6L9PQNcKit2xlpahs+u5s5lZDPw1dXz+bjFwxOkq6YXYJL4LG1cbFGrWSbxkuwmX45Czte4iPN92pBR6WgaTfv9OUnGSXbQOw53OFUotk3CzOsUt2t9S1stxLdieGjdJRY5adb67SnYP+uU6nOz2JiQRbRoDdhiFOC4m17P38o/RT15FHbOHhoVl1zJ31W74YtpMiMK2D21Wl1xEbEyrv137vbn7ak1fYUIgGP83s1noLhlGJifphPLp3Ef924yBHvK8wOkJU4tnTBB7RWlN+6Q6y+EBA9uwJ00nzvWr3x7FuvNaeCMSeMkGdVx4DeEN0ucTbiZTOK+PD51c6wIdtI/DGwUuvf/PxaePT9DLi8PNJ/UVJTF8RLWjftb6GnbyApZp2KLZ+tlOK5rQd594Iy5llllKDDMO09/dRaiqR5OTFH64hwIedPyeeVnYc50NKNz+0i4vcQ8XsIet1sqWvhnx97k5seXItS+o5yfU0Th1p7nE23IMEoDMXqBWWsnDN4MyqOBnjXtEIee32vc1d8VKOw/D9YU/4h3qOeYeDhq62JbV3s7C9wVG9Bz2mekBfij5vqHU/BDvVNtYzClrrWBKOQPqluUxjxIwLbLWmOWCi9UfjMRZP53WeWADjeX2JX8+H2HqYVR/B7XM7vJJGahGPO68oXwdUPjLz0PBVfCD7wa7j+l2lnUqfLKYyF/pExCoYxw+USbl8xnUnD3GFzxo26x8PmzJvho3+O341VnaPLAJ+6VUtRW6qxdggicSMvzPaTHfA4MfPKvBBfWjXTMRp2SMEXyoabn9fe1aFtMHM1Xf585+7Mxr7DfN2qkJk4nFGwPIWG1m6aOnqd5K+NvTHZKqCFkYA2eKCNwmQtXvenLQfIC+updCJCUbafx9fv4zfr9/GxCyfyx89dgEt0qOlQWzf5jjEIWNfg2ORK3ru4gobWbp6zJDHsXokhcbl5Z+4/8YXej+Da9bwjblirCslPmJUhIlw2p4SX3znM5roW8rJ8TpgnO6CT8NvqW9l+oI2yWHDwnPMUvG4XBWE/bd19STcAqXjcLoI+/X3sZLntKfT2D3Cko5f8sJ/ynKCTtE4k0VCMZPjTiMku1YoGabBVdsMBHQqbUhh2/nZPJCMU4DcYxglvIC4XDjB5Gcx6t5Yp7+3QZavoZJxdumdz9sQ8/J6h73ucpraQT1d8rfoOLP4IrmgFX53cOihBa3sKa6xZEcOGjwJeWjp7nVj3tOJko2AP7dnqGAW/rhL7239CpJhzyrSEw57DHUkG5fK5JexsaOe25dMc4zqvIsbz2xto7Ih7Cnb+YThPYTjeNb2QgoifX6ypYdnMIhrbewYZtlSKogEe7b+Iz89qp2D7I4BuXEutWLpsTgnf/8s7PL35gNPdbTOjJMLW+hZcIkw9SujIpiQW5GBrN9Ggd0SDoRxPwVbJtTb5/LCfitxQ2pxCWk/BYuO+Zh5bW8sdK6cT8h3bllrb2MFv1u/jMxdNdn4GO6cQsQzdM/904TG992gxRsFwcuELwTUP6vruhq06Jmtx2ewSppfEN5mPXzhp2LeyQw1JIQerAuqq+YM3wOyAB49L2NvUScTvGeRJJJ0b9DCg4iJvqZtf3FNowSW6P4CpK2H1D6B6KVG3h3kVMdbXHNHPWfzz8sHyJ+dNyuc/n3sbgIKUnMKxCht63S6uPqOc+59/hx0HWnX4aJifF6DIMqKvz7yDFc0bGDi4hb2qIGn9oEtmK3ND1DR2OLkgm2nFEf6yrQGXiCPJcTRKsgO8ydD5hFTs37ddltqQUOFVkRNKK8xX09hJyOemo6c/qfFtXU0TN/7kNVq7+lhYlcNV88tGtIZUfvTCTv77lT2snl9GZZ5ulrNzCuGRDs86TpjwkeHkxOPXMXh3fCP4+tVz+dB5I68ESS1VPRp2KSnAxIKsYe9KbamL13Y1Upgwuc3Gjs9vq9fS0G6XWCMm3+/El5dYsy3yhwvbAOdNjjdzObkEyxgUZR+bpwBw4zkTyM3yccMDr9HTPzB8+Cjhe9W1A9f/ihcX30cP3kGvs0NIEC8ptZlenE3fgKKnP7nyaDjs98geoVHwul1Eg17HKNgVXnlhPxW5QVq6+pwpbjY1jR3MLtWl3LansHl/Czf8+FVys3zkh308vSllmhraAxhq1KiNUsqZE7L3SNwjaevqwyUQ/DsFKkeLMQqG0xZbBiEWGn6zS8SuQBouyQzxDWp9bRPTigdvbraRaO/pH9RcZrNkit7s0zaMJbCwKkbAqz/KdvioyrrbtJPqx0JxNMBPP3imU0WVOEc7HblZPrxu0VIX2SVsCp055Povt4xCSTRZHj7RoxqpUXBkLUZoFOy1OkbBrvDK0p4CMCjZXNvYwaTCcFyeBPjDpno6evv55cfO4dJZxfxl28EkA9DV28+Ke17ge3/ewXBs2t/iKL0m5jPaLDG80c5K/3s5oUZBRFaIyDYReVtEBg2lFZF/EpHNIvKWiDwrIuMjymM4LYml9C+MhJysZMmFobA9ha7egUGho9Tvma65DGB+RYwzqnJYNCG9VLaN3+NmsXWObRRmlUb5w61LOKt6+NcejTnlUb5//UKiQW9SI1k6RITCSMDpozjcppvp0sXZZ5dl8/X3zOGaM5I7wavzs/C5XYiMfPqf3aMy0vARpBiFBE+h3DIKiQ1srV29NLb3UJkbSpIn2X+kk6JIgKLsAJfOLKK9p5+X3znkvG77gVbae/r589b4/Ip0PL35AC4Bl8DexsTv23fURPuJ4IQZBRFxA/cBK4GZwPtEZGbKaeuBRUqpucCvgLtP1HoMhlTssFHOKDwFu/fB1vcfCltaAwYnmUFXw0RTmulS8bpd/PoT57J8VvFR13XpzCLCfk9SDmF6cfZxuctcOq2QN768jDnlR++EL8r2O0ZhuDJWEeG9iysHVUd53C6mFIWpzA051UJHw85LjMYo5CTIjDS06bLf7ICHilxtYBKF+ey+hcrcUJI8yf4jnY6Xcu6kfCJ+T1IIyRYk3FLXMmx/wTObD3BGVQ4l0WCKp9A75vkEOLGewpnA20qpnUqpHuBR4KrEE5RSzymlbNP4ClCOwTBGTMjPoiDid1RFR4LtKRw1fJQwGjWdpwDxvMKxVgglcv1ZVbx4x7tGvJGOlpEal+JowOk/ONTeM2wyfihuvWQqt106bcTn2+XII80NQcJsDKxyW6vsNxbysWpuCT96YSe/XKOb9+zKI+0p+JI8BXt2g8/jYun0Qp7ZfIB+q8lt475mp6v+pbcPkY7axg621LWwbGYRZTmpRuEU8xSAMqA24fFe69hQfBj4fbonRORmEVkrImsbGoZ3xQyGkZIf9rPmzktGNY+6ODuA1y0jzim4hgmD2EYhndTEaHG5ZFS5kRPF1KIIuw6309LVy+G27kGVRyNh2cwirphXevQTLYqjAeaVR50ZCyPB3tyVUtY649fuW9fOY8mUfL7w+Fv89s39Tn6hMk97CofbehgYUOxv7nImCoL21g6397CuRlcvbdrfwuKqXHJCXl7YkX7fshPMy2YWWz0SyYnmcOrc9THgRBqFdLcWKs0xROQDwCLgG+meV0r9SCm1SCm1qKBgZGVqBsOJ4B/PncBjHz/3qPXo9kyFCflZQ443tcNWx8NTyBQWT8hFWaW4h9uO0gV9nPC6XfzPp8/n4hlFI35NXpaP3n5Fa3cfh9t7nLAg6BzN/TecwaKqXG55dD2PrqkhGvQSDXodY3LY0q0qTTAKS6cV4HO7+OPGevoHFFvrW5hdFuXcyfm8uONQksCgze831DO1KEx1fhblOSHqW7ocob627j6nR2EsOZFGYS9QkfC4HNifepKIXALcCVyplBqbKRIGwzGSHfCO6I7U63YR8rmZkSafYGM3sJ1KRmF+RQy3S3h9TxON7T1HrZwaL+x17Wxo51Br96B1hnweHrxpMedNzuedhnYqc3UCOjfLR0dPP7sOaRHCRKMQCXg5d3IeT28+wDsNbXT1DjC7LJsLpuRzsLXbkeKwqW/uYs2eRlbN1V5RRU6QAYUTfjsVw0drgCkiUi0iPuA64MnEE0RkAXA/2iAcPIFrMRjGnM9fOo0PnjdhyOftstRjlaLIRLL8HmaURHhu20F6+geSJC4yiYumFxLxe7jvubc51N6TNIvbJuTz8OMbF3HtonKumKdLaG3vbtN+LUtuJ5ptls8qpqaxg1+v0+NJZ5VGOd/qN/lrSgjpdxvqUApWzdXvbVc+2R3VOnx0ChkFpVQf8Gngj8AW4JdKqU0icpeIXGmd9g0gDDwmIm+IyJNDvJ3BcNLx4fOrVcUQaQAACqdJREFUnVLRdMwsyaYo2592QzqZWVSVy8Z9uvImUz2FWMjHRy+YyDObD9DTNzDkOv0eN3dfPY+bL9Dd8bZ3Z/98iTkFgEtmFCECP3t5D36Pi0kFWZTFgpTnBHmj9kjSuU+9tZ+ZJdlOeXN5jn6vvU0d9A8o2nv6TzlPAaXU/1NKTVVKTVJK/bt17MtKqSetry9RShUppeZb/64c/h0NhlOHq+aX8eq/XIJvGH2mk5FFE+JDcMYip3Cs3HR+tbO+vBF6NImeQtDrHlQGWxDxc0ZlDp29/UwvjjizHypzQ+xPGEW7t6mD9TVHWDUvrlJbEg3gtmRUbDG8yKnkKRgMhtOTRVVx7yg/g72gsN/DJyx9rKG6ylOxjciOg22UxgJpS3UvnaUT3jNL41VtpbEg+xKMwu/eqgNg1Zzk2d7F2QH2NnUmyWaPNcYoGAyG40pxNOCEQjI1fGTzwfMmcM9753POxLwRnW/ngfoHVFKSOZEVs0rwuIRFCWNDyywlV7uy6NVdjUwtCjvidzZ2WaotLZJljILBYDgVsDfETA4fga4SW72gLGnE53AkypOk5hNsKvNCPH/7u3j3gnhbVllOEJVQWbT7UHvauSLlOSFqGzu555kdeFzCjJKR6T8dT4x0tsFgOO7cdH41kwrC6Wc6n8R43C5iIS9HOnqH9BRgsMGwH+890kFpLEBtUwfLZw+WLynPCVLf0sUfNtXzL5dNZ3KhMQoGg+EUYG55jLnlI+8wPpnIDfmOahRSsY3CvqZO9se66O1XVKdRsLXDbhdPL+Qj5w+enz0WmPCRwWAwjAI7r5DaozAcJda5+450suuwbnybkEYqZem0Qm44u4pvXjMP1xiM3kyH8RQMBoNhFNhlqUPlFNLh97gpjPjZf6TTef2ElCQz6JLW/7169vFZ6DFijILBYDCMAruBLXVq3NGwy1JDPg9ZPnfSPPFMwhgFg8FgGAVLpxXSN6BGnUQvywmyaV8zPreLqrzhx7mOJ8YoGAwGwyi4bE6JM2N6NJTHgjyzWUtlzyoduVz7WGMSzQaDwTAGlMaC9PQNsPtwhzNDOxMxRsFgMBjGgMTEdLrKo0zBGAWDwWAYA8py4kbhaJP7xhNjFAwGg2EMSGx2y+TwkUk0GwwGwxgQDXqJ+D0MKJXRMzSMUTAYDIYxoiwniNslGVuOCsYoGAwGw5jx2Yun4MpggwDGKBgMBsOYcSz9DWONSTQbDAaDwcEYBYPBYDA4GKNgMBgMBgdjFAwGg8HgYIyCwWAwGByMUTAYDAaDgzEKBoPBYHAwRsFgMBgMDqKUGu81jAoRaQD2HOPL84FDx3E5J5KTZa1mncefk2WtZp3HlxO9ziqlVMHRTjrpjMLfg4isVUotGu91jISTZa1mncefk2WtZp3Hl0xZpwkfGQwGg8HBGAWDwWAwOJxuRuFH472AUXCyrNWs8/hzsqzVrPP4khHrPK1yCgaDwWAYntPNUzAYDAbDMBijYDAYDAaH08YoiMgKEdkmIm+LyBfGez02IlIhIs+JyBYR2SQit1jHc0XkGRHZYf2fM95rBRARt4isF5GnrMfVIvKqtc5fiIhvvNcIICIxEfmViGy1ru05mXhNReRz1u99o4g8IiKBTLimIvITETkoIhsTjqW9fqK51/psvSUiCzNgrd+wfvdvichvRCSW8NwXrbVuE5Hl47nOhOduExElIvnW43G7pqeFURARN3AfsBKYCbxPRGaO76oc+oDPK6VmAGcDn7LW9gXgWaXUFOBZ63EmcAuwJeHx14HvWOtsAj48LqsazHeBPyilpgPz0GvOqGsqImXAZ4FFSqnZgBu4jsy4pg8CK1KODXX9VgJTrH83Az8YozXaPMjgtT4DzFZKzQW2A18EsD5b1wGzrNd839ofxmudiEgFsAyoSTg8btf0tDAKwJnA20qpnUqpHuBR4KpxXhMASqk6pdQ66+tW9OZVhl7fQ9ZpDwGrx2eFcUSkHLgc+LH1WICLgF9Zp2TKOrOBC4AHAJRSPUqpI2TgNUWPxA2KiAcIAXVkwDVVSr0ANKYcHur6XQX8TGleAWIiMmZzJ9OtVSn1tFKqz3r4ClCesNZHlVLdSqldwNvo/WFc1mnxHeB2ILHqZ9yu6eliFMqA2oTHe61jGYWITAAWAK8CRUqpOtCGAygcv5U53IP+4x2wHucBRxI+fJlyXScCDcBPrVDXj0Ukiwy7pkqpfcA30XeIdUAz8DqZeU1h6OuX6Z+vm4DfW19n1FpF5Epgn1LqzZSnxm2dp4tRkDTHMqoWV0TCwK+BW5VSLeO9nlREZBVwUCn1euLhNKdmwnX1AAuBHyilFgDtZE74zcGKyV8FVAOlQBY6bJBKJlzT4cjUvwNE5E50iPZh+1Ca08ZlrSISAu4Evpzu6TTHxmSdp4tR2AtUJDwuB/aP01oGISJetEF4WCn1uHX4gO0uWv8fHK/1WZwHXCkiu9Hht4vQnkPMCn1A5lzXvcBepdSr1uNfoY1Epl3TS4BdSqkGpVQv8DhwLpl5TWHo65eRny8RuRFYBVyv4g1ZmbTWSegbgjetz1U5sE5EihnHdZ4uRmENMMWq6vChE01PjvOaACcu/wCwRSn17YSnngRutL6+EfifsV5bIkqpLyqlypVSE9DX789KqeuB54CrrdPGfZ0ASql6oFZEplmHLgY2k2HXFB02OltEQtbfgb3OjLumFkNdvyeBf7QqZs4Gmu0w03ghIiuAO4ArlVIdCU89CVwnIn4RqUYncl8bjzUqpTYopQqVUhOsz9VeYKH19zt+11QpdVr8Ay5DVyG8A9w53utJWNf5aLfwLeAN699l6Hj9s8AO6//c8V5rwpqXAk9ZX09Ef6jeBh4D/OO9Pmtd84G11nV9AsjJxGsKfAXYCmwE/hvwZ8I1BR5B5zl60ZvVh4e6fuhQx33WZ2sDuppqvNf6Njomb3+mfphw/p3WWrcBK8dznSnP7wbyx/uaGpkLg8FgMDicLuEjg8FgMIwAYxQMBoPB4GCMgsFgMBgcjFEwGAwGg4MxCgaDwWBwMEbBYBhDRGSpWAqzBkMmYoyCwWAwGByMUTAY0iAiHxCR10TkDRG5X/QciTYR+ZaIrBORZ0WkwDp3voi8kqDdb88ZmCwifxKRN63XTLLePizxWQ8PW93MBkNGYIyCwZCCiMwA3gucp5SaD/QD16MF69YppRYCzwP/y3rJz4A7lNbu35Bw/GHgPqXUPLSmkS1TsAC4FT3bYyJaV8pgyAg8Rz/FYDjtuBg4A1hj3cQH0eJvA8AvrHP+L/C4iESBmFLqeev4Q8BjIhIBypRSvwFQSnUBWO/3mlJqr/X4DWAC8OKJ/7EMhqNjjILBMBgBHlJKfTHpoMiXUs4bTiNmuJBQd8LX/ZjPoSGDMOEjg2EwzwJXi0ghOLOJq9CfF1u99P3Ai0qpZqBJRJZYx28Anld6JsZeEVltvYff0s83GDIac4diMKSglNosIv8KPC0iLrSq5afQw3pmicjr6Clp77VeciPwQ2vT3wl8yDp+A3C/iNxlvcc1Y/hjGAzHhFFJNRhGiIi0KaXC470Og+FEYsJHBoPBYHAwnoLBYDAYHIynYDAYDAYHYxQMBoPB4GCMgsFgMBgcjFEwGAwGg4MxCgaDwWBw+P9YzF5GoXk1uwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f35d4a62278>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#####################################################\n",
    "\n",
    "# print(np.array(history.losses))\n",
    "loss = train_history.history['loss']\n",
    "val_loss = train_history.history['val_loss']\n",
    "plt.plot(loss)\n",
    "plt.plot(val_loss)\n",
    "plt.legend(['loss', 'val_loss'])\n",
    "plt.title('AdaDelta')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer number: 2 Layer name: conv2d_81\n",
      "(3, 2, 3, 6)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5,0.98,'Convolution kernels in layer conv2d_81')"
      ]
     },
     "execution_count": 123,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f35d460f780>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualize a few kernels from the first convolutional layer:\n",
    "# enc.layers[2].get_config()\n",
    "# dir(enc.layers[2])\n",
    "x = enc.layers[2].get_weights()\n",
    "print ('Layer number:', 2, 'Layer name:', enc.layers[2].name)\n",
    "sh = x[0].shape\n",
    "print(sh) # kernel length x input channels x output channels\n",
    "\n",
    "c = 0\n",
    "fig, axs = plt.subplots(sh[2],sh[3])\n",
    "fig.set_size_inches(18.5, 10.5)\n",
    "axs = axs.ravel()\n",
    "for i in range(sh[2]):\n",
    "    for j in range(sh[3]):\n",
    "        axs[c].imshow(x[0][:,:,i,j])\n",
    "        axs[c].get_xaxis().set_visible(False)\n",
    "        axs[c].get_yaxis().set_visible(False)        \n",
    "        axs[c].set_title(str(i)+str(j))\n",
    "        c = c + 1\n",
    "plt.suptitle(str('Convolution kernels in layer ' + enc.layers[2].name), fontsize=16)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy on the test data: 0.875\n",
      "Accuracy on the training data: 1.0\n",
      "Elapsed 76.21398878097534 seconds\n"
     ]
    }
   ],
   "source": [
    "##########################################################\n",
    "\n",
    "# now test sample by sample\n",
    "cor_test = []\n",
    "L = x_test.shape\n",
    "for sample_no in range(x_test.shape[0]):\n",
    "    pred = enc.predict(np.expand_dims(x_test[sample_no],0))\n",
    "    if ((pred[0][0]<pred[0][1]) and (y_test[sample_no,0] < y_test[sample_no,1]) or\n",
    "       ((pred[0][0]>pred[0][1]) & (y_test[sample_no,0] > y_test[sample_no,1]))):\n",
    "        cor_test.append(1)\n",
    "    else:\n",
    "        cor_test.append(0)\n",
    "\n",
    "cor_train = []\n",
    "L = x_train.shape\n",
    "for sample_no in range(x_train.shape[0]):\n",
    "    pred = enc.predict(np.expand_dims(x_train[sample_no],0))\n",
    "    if ((pred[0][0]<pred[0][1]) and (y_train[sample_no,0] < y_train[sample_no,1]) or\n",
    "       ((pred[0][0]>pred[0][1]) & (y_train[sample_no,0] > y_train[sample_no,1]))):\n",
    "        cor_train.append(1)\n",
    "    else:\n",
    "        cor_train.append(0)\n",
    "\n",
    "# Report accuracies\n",
    "print('Accuracy on the test data:', np.mean(cor_test))\n",
    "print('Accuracy on the training data:', np.mean(cor_train))\n",
    "t1 = time.time()\n",
    "print('Elapsed', t1-t0, 'seconds')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}