svhn-single-digit-classification.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D\n",
    "from keras.layers import Dropout, Flatten, Dense, Activation, Reshape, Input\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.models import Sequential, load_model\n",
    "from keras.utils  import to_categorical\n",
    "from datetime     import datetime\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import cv2\n",
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_dir = 'numpy_arrays'\n",
    "X  = 'train_digit_features.npy'\n",
    "y  = 'train_digit_labels.npy'\n",
    "\n",
    "train_features = np.load(os.path.join(data_dir, X))\n",
    "train_labels   = np.load(os.path.join(data_dir, y))\n",
    "\n",
    "X  = 'test_digit_features.npy'\n",
    "y  = 'test_digit_labels.npy'\n",
    "\n",
    "test_features = np.load(os.path.join(data_dir, X))\n",
    "test_labels   = np.load(os.path.join(data_dir, y))\n",
    "\n",
    "#print(\"Features:\",train_features.shape)\n",
    "#print(\"Labels\", train_labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LABEL: 3.0\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAANYAAAD8CAYAAAAL1Fp+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nCUuIDEhYQmRAwhIiAxKWEBn4fxdB3/SfWNA5AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9caa839048>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "def display(feature_array, label_array, num):\n",
    "    print(\"LABEL: {}\".format(label_array[num] ))\n",
    "    fig,ax  = plt.subplots(1)\n",
    "    ax.imshow(feature_array[num])\n",
    "\n",
    "display(train_features, train_labels, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Preprocessing features\n",
    "from skimage.transform import resize\n",
    "%matplotlib inline\n",
    "\n",
    "def remove_zero_dims(features, labels):\n",
    "    skimmed_features = []\n",
    "    skimmed_labels   = []\n",
    "    \n",
    "    for i, feature in enumerate(features):\n",
    "        height, width = feature.shape[0], feature.shape[1]\n",
    "        to_height = 25\n",
    "        try:\n",
    "            to_width = to_height * ( (1.0 * width) / height)\n",
    "            tmp = resize(feature, (int(to_height), int(to_width), 3), mode='constant')\n",
    "            skimmed_features.append(feature)\n",
    "            skimmed_labels.append(labels[i])\n",
    "        except Exception as e:\n",
    "            print(\"ERROR: {}. DIMS: {} {} {}\".format(e, i, height, width))\n",
    "\n",
    "    return np.array(skimmed_features), np.array(skimmed_labels)\n",
    "\n",
    "        \n",
    "#tmp_features, tmp_labels = remove_zero_dims(train_features, train_labels)\n",
    "#print(tmp_features.shape)\n",
    "#print(tmp_labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def crop(features, to_height):\n",
    "    #mean_width = np.mean([f.shape[1] for f in features])\n",
    "    #print(\"MEAN WIDTH: {}\".format(mean_width))\n",
    "    \n",
    "    tmp_resized_features = []\n",
    "    for i, feature in enumerate(features):\n",
    "        #print(feature.shape)\n",
    "        height, width = feature.shape[0], feature.shape[1]\n",
    "\n",
    "        to_width = to_height * ( (1.0 * width) / height)\n",
    "        #tmp = resize(feature, (int(to_height), int(to_width), 3), mode='constant')\n",
    "        tmp = resize(feature, (50, 25, 3), mode='constant')\n",
    "        #print(tmp.shape)\n",
    "        tmp_resized_features.append(tmp)\n",
    "        \n",
    "    return np.array(tmp_resized_features)\n",
    "\n",
    "def pad_features(features):\n",
    "    new_features = []\n",
    "    for feature in features:\n",
    "        #print(feature.shape)\n",
    "        height, width, _ = feature.shape\n",
    "        if height > width:\n",
    "            pad = np.zeros((int(height), int((height - width) / 2), 3))\n",
    "            tmp = np.concatenate((pad, feature, pad), axis=1)\n",
    "#            if tmp.shape[0] > tmp.shape[1]:\n",
    "#                one_px_pad = np.zeros((int(height), 1, 3))\n",
    "#                tmp = np.concatenate((tmp, one_px_pad), axis=1)\n",
    "                \n",
    "        else:\n",
    "#            print(feature.shape)\n",
    "            pad = np.zeros((int((width - height) / 2), int(width), 3))\n",
    "            tmp = np.concatenate((pad, feature, pad), axis=0)\n",
    "#            print(tmp.shape)\n",
    "        new_features.append(tmp)\n",
    "    return np.array(new_features)\n",
    "\n",
    "\n",
    "def resize_32x32(features):\n",
    "    return np.array([resize(f, (32, 32, 3), mode='constant') for f in features])\n",
    "\n",
    "def resize_64x64(features):\n",
    "    return np.array([resize(f, (64, 64, 3), mode='constant') for f in features])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprocess_features(features, labels):\n",
    "    features, labels = remove_zero_dims(features, labels)\n",
    "    features = pad_features(features)\n",
    "    features = resize_32x32(features)\n",
    "    \n",
    "    return features, labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprocess_labels(labels):\n",
    "    # Removing labels that are not 0 - 9.\n",
    "    labels[labels == 10.0] = 0.0\n",
    "\n",
    "    # One hot encoding\n",
    "    encoded_labels = to_categorical(labels)\n",
    "    \n",
    "    return encoded_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ERROR: float division by zero. DIMS: 543 30 0\n",
      "ERROR: float division by zero. DIMS: 1571 108 0\n",
      "ERROR: float division by zero. DIMS: 6875 38 0\n",
      "ERROR: float division by zero. DIMS: 11710 40 0\n",
      "ERROR: float division by zero. DIMS: 15150 69 0\n",
      "ERROR: float division by zero. DIMS: 20600 46 0\n",
      "ERROR: float division by zero. DIMS: 26065 33 0\n",
      "ERROR: float division by zero. DIMS: 28859 32 0\n",
      "ERROR: float division by zero. DIMS: 33178 17 0\n",
      "ERROR: float division by zero. DIMS: 40034 68 0\n",
      "ERROR: float division by zero. DIMS: 40296 30 0\n",
      "ERROR: float division by zero. DIMS: 42428 46 0\n",
      "ERROR: float division by zero. DIMS: 44972 39 0\n",
      "ERROR: float division by zero. DIMS: 45752 49 0\n",
      "ERROR: float division by zero. DIMS: 45780 53 0\n",
      "ERROR: float division by zero. DIMS: 49123 16 0\n",
      "ERROR: float division by zero. DIMS: 52328 46 0\n",
      "ERROR: float division by zero. DIMS: 58190 64 0\n",
      "ERROR: float division by zero. DIMS: 60511 50 0\n",
      "ERROR: float division by zero. DIMS: 60866 54 0\n",
      "ERROR: float division by zero. DIMS: 62449 37 0\n",
      "ERROR: float division by zero. DIMS: 63075 79 0\n",
      "ERROR: float division by zero. DIMS: 63121 34 0\n",
      "ERROR: float division by zero. DIMS: 63190 173 0\n",
      "ERROR: float division by zero. DIMS: 64847 29 1\n",
      "ERROR: float division by zero. DIMS: 65226 32 0\n",
      "ERROR: float division by zero. DIMS: 66885 27 0\n",
      "ERROR: float division by zero. DIMS: 67866 97 0\n",
      "ERROR: float division by zero. DIMS: 70566 83 0\n",
      "ERROR: float division by zero. DIMS: 71125 29 0\n",
      "ERROR: float division by zero. DIMS: 662 18 0\n",
      "ERROR: float division by zero. DIMS: 842 60 0\n",
      "ERROR: float division by zero. DIMS: 1017 52 0\n",
      "ERROR: float division by zero. DIMS: 5186 71 0\n",
      "ERROR: float division by zero. DIMS: 10451 16 0\n",
      "ERROR: float division by zero. DIMS: 15199 36 0\n",
      "ERROR: float division by zero. DIMS: 17710 21 0\n",
      "ERROR: float division by zero. DIMS: 24752 60 0\n"
     ]
    }
   ],
   "source": [
    "processed_train_features, labels = preprocess_features(train_features, train_labels)  \n",
    "processed_train_labels = preprocess_labels(labels)\n",
    "\n",
    "processed_test_features, labels = preprocess_features(test_features, test_labels)  \n",
    "processed_test_labels = preprocess_labels(labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train feature shape: (73227, 32, 32, 3)\n",
      "Train label shape: (73227, 10)\n",
      "Test feature shape: (26024, 32, 32, 3)\n",
      "Test label shape: (26024, 10)\n",
      "LABEL: [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  1.]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  1.])"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca659fba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"Train feature shape:\", processed_train_features.shape)\n",
    "print(\"Train label shape:\", processed_train_labels.shape)\n",
    "print(\"Test feature shape:\", processed_test_features.shape)\n",
    "print(\"Test label shape:\", processed_test_labels.shape)\n",
    "\n",
    "pic = -1\n",
    "display(processed_train_features, processed_train_labels, pic)\n",
    "processed_train_labels[pic]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Models\n",
    "def dense_various_size_model(in_shape):\n",
    "    inputs = Input(shape=(3,None,None))\n",
    "    x = Dense(32, activation='relu')(inputs)\n",
    "    x = Dense(32, activation='relu')(x)\n",
    "    x = Flatten()(x)\n",
    "    x = Dense(10, activation='softmax')(x)\n",
    "    \n",
    "    x = Model(inputs=inputs, outputs=x)\n",
    "    x.summary()\n",
    "    return x\n",
    "    \n",
    "def dense_model(in_shape):\n",
    "    model = Sequential()\n",
    "    model.add(Dense(32, input_shape=in_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dense(32))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dense(8))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(10, activation='softmax'))\n",
    "    model.summary()\n",
    "    return model\n",
    "\n",
    "def pooling_model(in_shape):\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(filters=16, kernel_size=2, padding='same', activation='relu', input_shape=in_shape))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(Dropout(0.2))\n",
    "    model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(GlobalAveragePooling2D())\n",
    "    model.add(Dropout(0.2))\n",
    "    model.add(Dense(10, activation='softmax'))\n",
    "    model.summary()\n",
    "    return model\n",
    "\n",
    "def cnn_model(in_shape):\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(filters=48, kernel_size=2, padding='same', activation='relu', input_shape=in_shape))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Conv2D(filters=128, kernel_size=2, padding='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Conv2D(filters=160, kernel_size=2, padding='same', activation='relu'))\n",
    "    model.add(MaxPooling2D(pool_size=2))\n",
    "    model.add(BatchNormalization())\n",
    "    \n",
    "    #model.add(Conv2D(filters=192, kernel_size=2, padding='same', activation='relu'))\n",
    "    #model.add(Conv2D(filters=192, kernel_size=2, padding='same', activation='relu'))\n",
    "    #model.add(Conv2D(filters=192, kernel_size=2, padding='same', activation='relu'))\n",
    "    #model.add(Conv2D(filters=192, kernel_size=2, padding='same', activation='relu'))\n",
    "\n",
    "    model.add(GlobalAveragePooling2D())\n",
    "    model.add(Dense(512, activation='relu'))\n",
    "    model.add(Dense(10, activation='softmax'))\n",
    "    model.summary()\n",
    "    return model\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_1 (Conv2D)            (None, 32, 32, 48)        624       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 16, 16, 48)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 16, 16, 48)        192       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 16, 16, 64)        12352     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 8, 8, 64)          0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 8, 8, 64)          256       \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 8, 8, 128)         32896     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 4, 4, 128)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 4, 4, 128)         512       \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 4, 4, 160)         82080     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2 (None, 2, 2, 160)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_4 (Batch (None, 2, 2, 160)         640       \n",
      "_________________________________________________________________\n",
      "global_average_pooling2d_1 ( (None, 160)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 512)               82432     \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                5130      \n",
      "=================================================================\n",
      "Total params: 217,114.0\n",
      "Trainable params: 216,314.0\n",
      "Non-trainable params: 800.0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = cnn_model(processed_train_features.shape[1:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "MODEL FILE: models/singledigits_20epochs_20190717_164218.hdf5\n",
      "Train on 54920 samples, validate on 18307 samples\n",
      "Epoch 1/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.8197 - acc: 0.7341Epoch 00000: acc improved from -inf to 0.73412, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 214s - loss: 0.8196 - acc: 0.7341 - val_loss: 0.6829 - val_acc: 0.7806\n",
      "Epoch 2/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.4281 - acc: 0.8677Epoch 00001: acc improved from 0.73412 to 0.86770, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 216s - loss: 0.4280 - acc: 0.8677 - val_loss: 0.4695 - val_acc: 0.8550\n",
      "Epoch 3/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.3561 - acc: 0.8921Epoch 00002: acc improved from 0.86770 to 0.89212, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 214s - loss: 0.3561 - acc: 0.8921 - val_loss: 0.4172 - val_acc: 0.8742\n",
      "Epoch 4/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.3119 - acc: 0.9052- ETA:Epoch 00003: acc improved from 0.89212 to 0.90526, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 395s - loss: 0.3119 - acc: 0.9053 - val_loss: 0.3709 - val_acc: 0.8932\n",
      "Epoch 5/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.2823 - acc: 0.9155Epoch 00004: acc improved from 0.90526 to 0.91551, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 412s - loss: 0.2823 - acc: 0.9155 - val_loss: 0.3903 - val_acc: 0.8850\n",
      "Epoch 6/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.2570 - acc: 0.9232Epoch 00005: acc improved from 0.91551 to 0.92318, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 414s - loss: 0.2570 - acc: 0.9232 - val_loss: 0.3854 - val_acc: 0.8927\n",
      "Epoch 7/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.2351 - acc: 0.928 - ETA: 0s - loss: 0.2350 - acc: 0.9289Epoch 00006: acc improved from 0.92318 to 0.92891, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 415s - loss: 0.2351 - acc: 0.9289 - val_loss: 0.5194 - val_acc: 0.8446\n",
      "Epoch 8/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.2174 - acc: 0.9352Epoch 00007: acc improved from 0.92891 to 0.93520, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 415s - loss: 0.2174 - acc: 0.9352 - val_loss: 0.5173 - val_acc: 0.8508\n",
      "Epoch 9/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1996 - acc: 0.9394Epoch 00008: acc improved from 0.93520 to 0.93944, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 416s - loss: 0.1996 - acc: 0.9394 - val_loss: 0.6479 - val_acc: 0.8314\n",
      "Epoch 10/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1839 - acc: 0.9444Epoch 00009: acc improved from 0.93944 to 0.94439, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 418s - loss: 0.1839 - acc: 0.9444 - val_loss: 0.4428 - val_acc: 0.8811\n",
      "Epoch 11/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1716 - acc: 0.9481Epoch 00010: acc improved from 0.94439 to 0.94812, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 419s - loss: 0.1716 - acc: 0.9481 - val_loss: 0.8023 - val_acc: 0.7833\n",
      "Epoch 12/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1624 - acc: 0.9507Epoch 00011: acc improved from 0.94812 to 0.95073, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 420s - loss: 0.1624 - acc: 0.9507 - val_loss: 0.4140 - val_acc: 0.9012\n",
      "Epoch 13/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1480 - acc: 0.9548Epoch 00012: acc improved from 0.95073 to 0.95481, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 422s - loss: 0.1482 - acc: 0.9548 - val_loss: 0.4462 - val_acc: 0.8886\n",
      "Epoch 14/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1417 - acc: 0.9564Epoch 00013: acc improved from 0.95481 to 0.95643, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 420s - loss: 0.1417 - acc: 0.9564 - val_loss: 0.5388 - val_acc: 0.8837\n",
      "Epoch 15/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1326 - acc: 0.9594Epoch 00014: acc improved from 0.95643 to 0.95938, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 421s - loss: 0.1326 - acc: 0.9594 - val_loss: 0.4454 - val_acc: 0.9018\n",
      "Epoch 16/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1255 - acc: 0.9626Epoch 00015: acc improved from 0.95938 to 0.96265, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 421s - loss: 0.1255 - acc: 0.9627 - val_loss: 0.4446 - val_acc: 0.9009\n",
      "Epoch 17/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1189 - acc: 0.9639Epoch 00016: acc improved from 0.96265 to 0.96395, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 420s - loss: 0.1189 - acc: 0.9639 - val_loss: 0.4963 - val_acc: 0.8982\n",
      "Epoch 18/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1140 - acc: 0.9657Epoch 00017: acc improved from 0.96395 to 0.96571, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 420s - loss: 0.1140 - acc: 0.9657 - val_loss: 0.6079 - val_acc: 0.8657\n",
      "Epoch 19/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1058 - acc: 0.9678Epoch 00018: acc improved from 0.96571 to 0.96781, saving model to models/singledigits_20epochs_20190717_164218.hdf5\n",
      "54920/54920 [==============================] - 420s - loss: 0.1058 - acc: 0.9678 - val_loss: 0.4887 - val_acc: 0.9010\n",
      "Epoch 20/20\n",
      "54912/54920 [============================>.] - ETA: 0s - loss: 0.1052 - acc: 0.9677Epoch 00019: acc did not improve\n",
      "54920/54920 [==============================] - 420s - loss: 0.1052 - acc: 0.9677 - val_loss: 0.5450 - val_acc: 0.9039\n",
      "TIME: 7741.820248842239\n"
     ]
    }
   ],
   "source": [
    "from keras.callbacks import ModelCheckpoint\n",
    "from time import time\n",
    "\n",
    "start      = time()\n",
    "epochs     = 20\n",
    "time_now   = datetime.now().strftime(\"%Y%m%d_%H%M%S\")\n",
    "model_path = os.path.join('models', 'singledigits_' + str(epochs) + 'epochs_' + time_now +'.hdf5')\n",
    "\n",
    "print(\"MODEL FILE: {}\".format(model_path))\n",
    "\n",
    "checkpointer = ModelCheckpoint(filepath=model_path, \n",
    "                               verbose=1, save_best_only=True, monitor='acc',)\n",
    "\n",
    "model.fit(processed_train_features, \n",
    "          processed_train_labels,\n",
    "          epochs=epochs, \n",
    "          batch_size=32,\n",
    "          validation_split=0.25,\n",
    "          callbacks=[checkpointer], \n",
    "          verbose=1)\n",
    "\n",
    "print(\"TIME: {}\".format(time() - start))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SCORE: 0.6667160498151682, ACCURACY: 0.8869505072241008\n"
     ]
    }
   ],
   "source": [
    "score, acc = model.evaluate(processed_test_features, processed_test_labels, verbose=0)\n",
    "print(\"SCORE: {}, ACCURACY: {}\".format(score, acc))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 5.  2.  1.  0.  6.  1.  9.  1.  1.  8.]\n"
     ]
    }
   ],
   "source": [
    "def predict(array_set):\n",
    "    y_pred      = model.predict(array_set)\n",
    "    return np.array([np.argmax(pred) * 1.0 for pred in y_pred])\n",
    "\n",
    "samples = 10\n",
    "print(predict(processed_test_features[:samples]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LABEL: 5\n",
      "LABEL: 2\n",
      "LABEL: 1\n",
      "LABEL: 0\n",
      "LABEL: 6\n",
      "LABEL: 1\n",
      "LABEL: 9\n",
      "LABEL: 1\n",
      "LABEL: 1\n",
      "LABEL: 8\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAYAk3t0AutoxAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9cd82d05c0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca6302390>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nyS+SKSW/SKaU/CKZUvKLZErJL5Kp/wUfA0Oxca6VTwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca5810978>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca5e74e48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/CKFUvKLFErJL1Ko/wNpnXHedyqGxAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca607c2b0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAFeNJREFUeJztnV+sXNV1xr81/65R7NRQqGUZt/wpUoWixqAriyooookS\nuSgSIEUIHpAfUG5UBalI6QOiUqFSH0hVQDxRXWorTkX50wACVagNtSJZeXEw1BiD24RYprFlbCJM\noA3Yd2ZWH86xdE1nfTN3zzln7u3+fpLluWfPOWfNnvnmzOxv1lrm7hBC5Edr1gEIIWaDxC9Epkj8\nQmSKxC9Epkj8QmSKxC9Epkj8QmSKxC9Epkj8QmRKZ5qdzWwHgMcAtAH8vbs/NOb+4c8JW612uF+v\nNzd6n068D40DlrafrXy/lH2ah8VY7S9A2S9K6Zkq/iVq479sJVM8HAxGbj937uyK9wEAd5/oRWep\nk2BmbQA/A/A1AMcBvArgTnd/m+wTnmzD5zeG59ryu1eN3P75jRezAMOhlsUfeFqteKzdHv1mw964\n2PHYGwN906BDwSB9mskBE8XqweiAvGj7/X44NhwO43OxGIOx1OOxR03f58nYx78+M3L7if86Svb5\nMBybVPzTfOzfDuAddz/q7ucAPA3glimOJ4RokGnEvwXAL5f9fbzcJoRYA0z1nX8SzGwBwELd5xFC\nrIxpxH8CwNZlf19ebrsAd18EsAjw7/xCiGaZ5mP/qwCuMbMrzawH4A4AL1UTlhCibpKv/O7eN7N7\nAPwrCqtvt7u/lRxItxuOrd/wWyO3b7zksnAfa8ULnu12/LA7nTiOTmf0ftH24lyxE9CmTkC1P8Hw\neHGbLttza46tso/ePhjEK/rnls6FY/0+s7ZWvnLPXAd6vDEeRwpRjJ1OL+l4kzLVd353fxnAyxXF\nIoRoEP3CT4hMkfiFyBSJX4hMkfiFyBSJX4hMqf0XfpPCcimGYXJGvBO10UgiDrP6er3R1kuX2JSd\nbpoNmGr1eTAnw35sX7F5ZAkw3NoabbWyJCgSBoDYInSPbbsofpY4NRiS45FzrbUWGLryC5EpEr8Q\nmSLxC5EpEr8QmSLxC5Epq2a1n62+RqWw2qSGH0sU6vXisW6wos/G2Go/TewhYwy2qhytVFM3hSyz\n85JWLHkqKnmWdr3pkLkaktX5aD5YglFrwJyAcAjDhAQjgCSh1Vz+UVd+ITJF4hciUyR+ITJF4hci\nUyR+ITJF4hciU9aE1RfVyOt1mS3H7LzR7b8AYG6OHDM4H6vhl2ptMYfNSbLNcDB6x6Wl2Nri9ezi\nOJhV2emMftxsPmjSD7Xz2GNbGrm9349fb0tkzOIwaEIQs/qiDlKpbeUmRVd+ITJF4hciUyR+ITJF\n4hciUyR+ITJF4hciU6ay+szsGICPAQwA9N19fopjhWORpcQy8ObWxXZej1h9kZ0HAK3I2iL19lhd\nOicpYsx+W1qKx86dG93y6tNPzsbnIq2wWNuzHstmDDIuaSZmOz4ee6kOhvEx+/3ARiOPizpsrNcs\nsQGZVRm+9GvO6qvC5/9jd/9VBccRQjSIPvYLkSnTit8B/MjMXjOzhSoCEkI0w7Qf+2909xNm9jsA\nXjGz/3D3fcvvUL4p6I1BiFXGVFd+dz9R/n8awAsAto+4z6K7z0+zGCiEqJ5k8ZvZ58xsw/nbAL4O\n4HBVgQkh6mWaj/2bALxQWnQdAP/o7v+SerAUq49ZTevWrQvHmNXXapMpCWJkBTCZjTYgmXZLS6Oz\n0QDg7NnRdh4QW3q/+eSTOA4SI8vCu+iieI67c6PnuDcXP8/Mum0Ray7K3AOAVjsqjkmKlpI2ZGyM\ntS9zkMqfM1p2Txa/ux8F8MUKYxFCNIisPiEyReIXIlMkfiEyReIXIlMkfiEyZdUU8GRWTlQgk/XI\nSx0zUkQyKsLoiYUbWcFHVnAzytwDgE/Pfjpy+2/+J7b6+v34XLw4aTxXUVYiS1RjtmJUELQ4KClo\n6kF2YWIPxcEgjoPZ1TRRMGGfKtCVX4hMkfiFyBSJX4hMkfiFyBSJX4hMWTWr/Ubq4LWDZBu2as+S\nfrpd9rBJPb6gTdaAJW2QpJ8hreGXlvTz6acrT+zpE2eBzXGnE48N+tFji9ewaSsvsgLf8nge24Ej\nEW0fGwcZY6/huFDf7NCVX4hMkfiFyBSJX4hMkfiFyBSJX4hMkfiFyJRVY/UxC2UuqAc3NxfXkGNJ\nJ6zjEnNrot2MHNBa8Rh5yMnOUJRINGS1BEliT5tMiAfWZzrkQXtKasy4sWphz5mtkhiXoyu/EJki\n8QuRKRK/EJki8QuRKRK/EJki8QuRKWOtPjPbDeAbAE67+xfKbZcAeAbAFQCOAbjd3c9MEwi1+oL2\nWrTtVurbGqm5F1l6LWL1sTjCVlIYlz0WHzPyI5ktx7ILh+14jNUnTKtmlwixAaMInUXILDs6+eyY\n5PkMr8H1WoCTSOT7AHZ8Ztt9APa6+zUA9pZ/CyHWEGPF7+77AHzwmc23ANhT3t4D4NaK4xJC1Ezq\nh+NN7n6yvP0eio69Qog1xNQ/73V3N/L7VjNbALAw7XmEENWSeuU/ZWabAaD8/3R0R3dfdPd5d59P\nPJcQogZSxf8SgJ3l7Z0AXqwmHCFEU0xi9T0F4CYAl5rZcQAPAHgIwLNmdjeAdwHcPm0gzEKJ2id1\nSFFHlrrnpODmkLTQGoR2WXwuFmNrjlg5JGGOFfDszY0uqtntxcU2nVhl3W4vHGMFPMOsSvI8M+tw\nSAqhsv3IU0NIK9IZFZodR9QSzUgLuyoYG6273xkMfbXiWIQQDaJf+AmRKRK/EJki8QuRKRK/EJki\n8QuRKaumgCe1+qJ+a21SXJLaeXEcA4+tPg96wnGbkllDsQ04HMQeFeuf1w1so8hOKs4VDtH92q14\nLLLEWCFL7tgleXaIM+NYBl5atiV7PlkyYLszej+eQTg9uvILkSkSvxCZIvELkSkSvxCZIvELkSkS\nvxCZsmqsvqoxYg3xMXbMaIDZUMy/YnYTKQrKegMGh+RFKVNttJi4cCaBBskqoRLvNsqMYxlzJA5m\nv/GxlWcK8v5+06MrvxCZIvELkSkSvxCZIvELkSkSvxCZsiZW+8OkDrZ0nLaYSxeBU94quQ9ARtkK\nfMLKfWKXqTEdoxJdjpSTcbtixWM0aSZ51T52HXj7tciiCXepBF35hcgUiV+ITJH4hcgUiV+ITJH4\nhcgUiV+ITJmkXdduAN8AcNrdv1BuexDAtwC8X97tfnd/eapIklyjNK8vtUZbeDrWGoyEyFpQpdpo\nYcU66mwxT4k8thoSgqomTnQirwFi5znxglse7zckhSPrrtUXMcmV//sAdozY/qi7byv/TSd8IUTj\njBW/u+8D8EEDsQghGmSa7/z3mNkhM9ttZhdXFpEQohFSxf84gKsBbANwEsDD0R3NbMHMDpjZgcRz\nCSFqIEn87n7K3QdedLF4AsB2ct9Fd5939/nUIIUQ1ZMkfjPbvOzP2wAcriYcIURTTGL1PQXgJgCX\nmtlxAA8AuMnMtqHwgY4B+HaNMcIDv4zZaC1iyZjFbZUYUU01lp03DFp8AYCxB0Br/xH7LdovOSku\nzTKtGjZXPIpq23XVMhbEUncNv7Hid/c7R2zeVUMsQogG0S/8hMgUiV+ITJH4hcgUiV+ITJH4hciU\nVVPAM8UuY/u0iJ3HOj8x38sRxEHsvGCXcigtGzCKoxhLyLRjNhS1TBPGWHgsy7HiBMLUgqarxfqs\nAl35hcgUiV+ITJH4hcgUiV+ITJH4hcgUiV+ITFk1Vl9SvUpavzPNvmqxgptJdlNsyw2Jn8eyAQek\nGGRoi5JimzSrL9HqS3vSErIVx5BSwDMlA28toiu/EJki8QuRKRK/EJki8QuRKRK/EJmyelb7CSkr\nvekJGCtf6XWP92EOAWvhNBgOksbCY7JMoVQngLbriuoupiYlpa72B/XxUl8fiYv9zCWYlX+gK78Q\nmSLxC5EpEr8QmSLxC5EpEr8QmSLxC5Epk7Tr2grgBwA2ofBbFt39MTO7BMAzAK5A0bLrdnc/U1+o\n/5fqjaG0M1L7itp5JEFnwI658kfO3CvW2qyVbJmmJPaQgof0OpUwH8RgY495uAotu1QmufL3AXzX\n3a8FcAOA75jZtQDuA7DX3a8BsLf8WwixRhgrfnc/6e6vl7c/BnAEwBYAtwDYU95tD4Bb6wpSCFE9\nK/rOb2ZXALgOwH4Am9z9ZDn0HoqvBUKINcLEP+81s/UAngNwr7t/tPz7nru7Bb/1NLMFAAvTBiqE\nqJaJrvxm1kUh/Cfd/fly8ykz21yObwZwetS+7r7o7vPuPl9FwEKIahgrfisu8bsAHHH3R5YNvQRg\nZ3l7J4AXqw9PCFEXk3zs/xKAuwC8aWYHy233A3gIwLNmdjeAdwHcPk0gaSX80mq+OTFlaBxBZhzL\nzqNjxM4bEjsvioPB7Dz2oLkNuOIwpsjcI5mHxCK0YK5Y7Kl2Hq3vR56zhKezEsaK391/gvgxf7Xa\ncIQQTaFf+AmRKRK/EJki8QuRKRK/EJki8QuRKauogCdpXRXYZakWG01GI7ZRbLGlWo5pmMXv2e3W\n6LFOux3uMyTz0WnH52qRsTjExJZctAApGQqfbGYdsrE6SMmAnB5d+YXIFIlfiEyR+IXIFIlfiEyR\n+IXIFIlfiExZNVYfc3KSsukGcT873mOOxbGy7cVg0qmoH0kz7QJLr92Jn+oWySDsdGKLsM2svihE\nNve0j1881iLWZ9oTkPr6SMvEjMbqzvbTlV+ITJH4hcgUiV+ITJH4hcgUiV+ITFk1q/1saTOqdZfa\n0oqtvPKkn3Avcry0tlCtIEEHiFf0gTiBh63as3qB7S5Z7e+QBKPACWCPmSZBsdVy4vqE+9HjpY3x\nY7IYozEl9gghakDiFyJTJH4hMkXiFyJTJH4hMkXiFyJTxlp9ZrYVwA9QtOB2AIvu/piZPQjgWwDe\nL+96v7u/nBoIs3mGw9FJOix5ZzgkFhUrWpfQg4rZeaSYHa3Fx6y+NrH62oGl1+nGTzW1+lhiD7H6\nWu3Rc0JzcBiJ7a4iG41Zb2MOGA9ReznBlq45sWcSn78P4Lvu/rqZbQDwmpm9Uo496u5/W194Qoi6\nmKRX30kAJ8vbH5vZEQBb6g5MCFEvK/oQZmZXALgOwP5y0z1mdsjMdpvZxRXHJoSokYnFb2brATwH\n4F53/wjA4wCuBrANxSeDh4P9FszsgJkdqCBeIURFTCR+M+uiEP6T7v48ALj7KXcfeLGS8QSA7aP2\ndfdFd5939/mqghZCTM9Y8VuxlL0LwBF3f2TZ9s3L7nYbgMPVhyeEqItJVvu/BOAuAG+a2cFy2/0A\n7jSzbSgMiWMAvj1NIMwmGfRHW3pLS0vhPkn15QC0iSUWWXqplh2txUf2Y623er1eMJLWaKrd7oZj\nc3Nz4Vi3O3q/aPu4MZoNyOy32A1OhLUNYxZh1XFMzySr/T/B6FdOsqcvhJg9+oWfEJki8QuRKRK/\nEJki8QuRKRK/EJmyagp4MrtmEGTv9QMLEACW+v1wrEVswFabFPcM9mPWIbPsLDFzr0Nab0XTyOJg\nNmC7HZ+r14utvl5vtG3HYmePmc0xbc2W4LHR7mtpjikvThqM1e0O6sovRKZI/EJkisQvRKZI/EJk\nisQvRKZI/EJkyqqx+hihFcKKIrLef7QYZEJmFvGhjNhoLSPFMYnl2OkQOzJ4Px+2ScFKavWRDEKS\nhReNdVi2JbPDBmxs5cUxU3vu8UKiqa+raEy9+oQQNSDxC5EpEr8QmSLxC5EpEr8QmSLxC5Epq8bq\no1lPgS3DLLsh6cXGxgbENooKdfL+c8zqI1l9rdhic2L1Rcek1haBWX1dUuw0KjLaJtmF7HlhvfXY\nWHRMfrzq7Tw+Fg7Viq78QmSKxC9Epkj8QmSKxC9Epkj8QmTK2NV+M1sHYB+AufL+P3T3B8zsSgBP\nA/htAK8BuMvdzyVHQlfuR9doGwziOn39AanT12etvFZepM1I2y0GW+VlCUHMCQhbhyWu9tMahCyh\nKRijCVeJDs0wIcGLJ36xOGpY7Z9RL69JrvxnAXzF3b+Ioh33DjO7AcD3ADzq7r8P4AyAu+sLUwhR\nNWPF7wX/Xf7ZLf85gK8A+GG5fQ+AW2uJUAhRCxN95zezdtmh9zSAVwD8AsCH7n7+c/dxAFvqCVEI\nUQcTid/dB+6+DcDlALYD+INJT2BmC2Z2wMwOJMYohKiBFa32u/uHAH4M4I8AbDSz8wuGlwM4Eeyz\n6O7z7j4/VaRCiEoZK34zu8zMNpa3LwLwNQBHULwJfLO8204AL9YVpBCieiZJ7NkMYI+ZtVG8WTzr\n7v9sZm8DeNrM/hrAvwPYNU0g587FLuGZM78auZ213eoG7aIAoNNhY6ydVNSui9lh4VBysgdNgop2\nTMweYXYeswHjeYyPF1m6ADBkdfqYbRfsx841YGOkNVifWc+ktdyvz3wwcvsS0UQVjBW/ux8CcN2I\n7UdRfP8XQqxB9As/ITJF4hciUyR+ITJF4hciUyR+ITLFaBuhqk9m9j6Ad8s/LwUw2sNrFsVxIYrj\nQtZaHL/n7pdNcsBGxX/Bic0OrIZf/SkOxZFrHPrYL0SmSPxCZMosxb84w3MvR3FciOK4kP+3cczs\nO78QYrboY78QmTIT8ZvZDjP7TzN7x8zum0UMZRzHzOxNMzvYZLERM9ttZqfN7PCybZeY2Stm9vPy\n/4tnFMeDZnainJODZnZzA3FsNbMfm9nbZvaWmf1Zub3ROSFxNDonZrbOzH5qZm+UcfxVuf1KM9tf\n6uYZM+tNdSJ3b/QfgDaKMmBXAegBeAPAtU3HUcZyDMClMzjvlwFcD+Dwsm1/A+C+8vZ9AL43ozge\nBPDnDc/HZgDXl7c3APgZgGubnhMSR6NzgiLveX15uwtgP4AbADwL4I5y+98B+NNpzjOLK/92AO+4\n+1EvSn0/DeCWGcQxM9x9H4DPJnHfgqIQKtBQQdQgjsZx95Pu/np5+2MUxWK2oOE5IXE0ihfUXjR3\nFuLfAuCXy/6eZfFPB/AjM3vNzBZmFMN5Nrn7yfL2ewA2zTCWe8zsUPm1oPavH8sxsytQ1I/YjxnO\nyWfiABqekyaK5ua+4Heju18P4E8AfMfMvjzrgIDinR+YUScH4HEAV6Po0XASwMNNndjM1gN4DsC9\n7v7R8rEm52REHI3PiU9RNHdSZiH+EwC2Lvs7LP5ZN+5+ovz/NIAXMNvKRKfMbDMAlP+fnkUQ7n6q\nfOENATyBhubEzLooBPekuz9fbm58TkbFMas5Kc+94qK5kzIL8b8K4Jpy5bIH4A4ALzUdhJl9zsw2\nnL8N4OsADvO9auUlFIVQgRkWRD0vtpLb0MCcWFEocBeAI+7+yLKhRuckiqPpOWmsaG5TK5ifWc28\nGcVK6i8A/MWMYrgKhdPwBoC3mowDwFMoPj4uofjudjeKnod7AfwcwL8BuGRGcfwDgDcBHEIhvs0N\nxHEjio/0hwAcLP/d3PSckDganRMAf4iiKO4hFG80f7nsNftTAO8A+CcAc9OcR7/wEyJTcl/wEyJb\nJH4hMkXiFyJTJH4hMkXiFyJTJH4hMkXiFyJTJH4hMuV/AaP/+zqb/R2tAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca5a29780>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nxC9EoUj8QhSKxC9Eofw/Y2qtgrp42pMAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca5673d30>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca5a5d208>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nhVLxixTq/wCj0Hp6dBkJ+AAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca56a2f98>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9ca523fd68>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "for i in range(samples):\n",
    "    tmp_labels = [ np.argmax(label) for label in processed_test_labels ]\n",
    "    display(processed_test_features,  tmp_labels, i)"
   ]
  },
  {
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}