DL4D_12_German_Traffic_Sign_Benchmark.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "45jXfJUS4wE0"
      },
      "source": [
        "# Distinguishing Traffic Signs"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BKKwtlIu4wE4"
      },
      "source": [
        "## Preparing image data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "570UTTOnZrie"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "import zipfile\n",
        "import pprint\n",
        "\n",
        "from skimage.transform import resize\n",
        "from imageio import imread\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "\n",
        "import warnings\n",
        "warnings.filterwarnings(\"ignore\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "SQB_Z5rWboUK"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.optimizers import Adam\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator \n",
        "from tensorflow.keras.utils import to_categorical\n",
        "from tensorflow.keras.layers import Conv2D, MaxPooling2D\n",
        "from tensorflow.keras.layers import Flatten, Dense, Dropout"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "_UNEK_X2ZP5P"
      },
      "outputs": [],
      "source": [
        "import urllib.request\n",
        "import os.path\n",
        "if not os.path.exists(\"GTSRB_Final_Training_Images.zip\"):\n",
        "    url = (\"https://sid.erda.dk/public/archives/\"\n",
        "    + \"daaeac0d7ce1152aea9b61d9f1e19370/\"\n",
        "    + \"GTSRB_Final_Training_Images.zip\")\n",
        "    filename = \"./GTSRB_Final_Training_Images.zip\"\n",
        "    urllib.request.urlretrieve(url, filename)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 5,
      "metadata": {
        "id": "VUiHtNGJsBuT"
      },
      "outputs": [],
      "source": [
        "IMG_SIZE = 32\n",
        "TEST_SIZE = 0.2"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "metadata": {
        "id": "pJJ8P5d2I1cF"
      },
      "outputs": [],
      "source": [
        "X, Xt, y, yt = list(), list(), list(), list()\n",
        "\n",
        "archive = zipfile.ZipFile(\n",
        "          'GTSRB_Final_Training_Images.zip', 'r')\n",
        "file_paths = [file for file in archive.namelist() \n",
        "              if '.ppm' in file]\n",
        "\n",
        "for filename in file_paths:\n",
        "    with archive.open(filename) as img_file:\n",
        "        img = imread(img_file.read())\n",
        "    img = resize(img, \n",
        "                 output_shape=(IMG_SIZE, IMG_SIZE),\n",
        "                 mode='reflect', anti_aliasing=True)\n",
        "    img_class = int(filename.split('/')[-2])\n",
        "    \n",
        "    if (hash(filename) % 1000) / 1000 > TEST_SIZE:\n",
        "        X.append(img)\n",
        "        y.append(img_class)\n",
        "    else:\n",
        "        Xt.append(img)\n",
        "        yt.append(img_class)\n",
        "\n",
        "archive.close()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5KZsP3GIZ6AI",
        "outputId": "69f38e30-0d9d-41d6-d0dd-07ca5982746c"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Train size:31219 test size:7990 (0.204)\n"
          ]
        }
      ],
      "source": [
        "test_ratio = len(Xt) / len(file_paths)\n",
        "print(\"Train size:{} test size:{} ({:0.3f})\".format(len(X), \n",
        "                len(Xt), \n",
        "                test_ratio))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "AeVvZi61s87C",
        "outputId": "1df52066-7a51-43dd-b0e2-7e64f19b92b0"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "No classes:43\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "classes, dist = np.unique(y+yt, return_counts=True)\n",
        "NUM_CLASSES = len(classes)\n",
        "print (\"No classes:{}\".format(NUM_CLASSES))\n",
        "\n",
        "plt.bar(classes, dist, align='center', alpha=0.5)\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "metadata": {
        "id": "z0ULObCbxO97"
      },
      "outputs": [],
      "source": [
        "class_weight = {c:dist[c]/np.sum(dist) for c in classes}"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zWYudzUH4wFi"
      },
      "source": [
        "## Running a classification task"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "metadata": {
        "id": "MW88y4HLbkpJ"
      },
      "outputs": [],
      "source": [
        "batch_size = 256\n",
        "tgen=ImageDataGenerator(rescale=1./255,\n",
        "                        rotation_range=5,\n",
        "                        width_shift_range=0.10, \n",
        "                        height_shift_range=0.10)\n",
        "\n",
        "train_gen = tgen.flow(np.array(X), \n",
        "                      to_categorical(y), \n",
        "                      batch_size=batch_size)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "metadata": {
        "id": "sSYl85NofHc7"
      },
      "outputs": [],
      "source": [
        "vgen=ImageDataGenerator(rescale=1./255)\n",
        "\n",
        "val_gen = vgen.flow(np.array(Xt), \n",
        "                    to_categorical(yt), \n",
        "                    batch_size=batch_size)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "metadata": {
        "id": "wP6BgtRib00-"
      },
      "outputs": [],
      "source": [
        "def small_cnn():\n",
        "    model = Sequential()\n",
        "    model.add(Conv2D(32, (5, 5), padding='same',\n",
        "                     input_shape=(IMG_SIZE, IMG_SIZE, 3),\n",
        "                     activation='relu'))\n",
        "    model.add(Conv2D(64, (5, 5), activation='relu'))   \n",
        "    model.add(Flatten())\n",
        "    model.add(Dense(768, activation='relu'))\n",
        "    model.add(Dropout(0.4))\n",
        "    model.add(Dense(NUM_CLASSES, activation='softmax'))\n",
        "    return model\n",
        "\n",
        "\n",
        "model = small_cnn()\n",
        "model.compile(loss='categorical_crossentropy', \n",
        "              optimizer=Adam(),\n",
        "              metrics=['accuracy'])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vifoZNqpcU1z",
        "outputId": "9b887cc9-acf3-437b-b681-df831bc29664"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "121/121 - 30s - loss: 0.1118 - accuracy: 0.0683 - val_loss: 3.4798 - val_accuracy: 0.1040 - 30s/epoch - 248ms/step\n",
            "Epoch 2/100\n",
            "121/121 - 15s - loss: 0.1002 - accuracy: 0.1420 - val_loss: 2.9420 - val_accuracy: 0.2597 - 15s/epoch - 127ms/step\n",
            "Epoch 3/100\n",
            "121/121 - 16s - loss: 0.0758 - accuracy: 0.2889 - val_loss: 1.9754 - val_accuracy: 0.4464 - 16s/epoch - 132ms/step\n",
            "Epoch 4/100\n",
            "121/121 - 16s - loss: 0.0603 - accuracy: 0.3902 - val_loss: 1.5274 - val_accuracy: 0.5340 - 16s/epoch - 129ms/step\n",
            "Epoch 5/100\n",
            "121/121 - 17s - loss: 0.0524 - accuracy: 0.4468 - val_loss: 1.3016 - val_accuracy: 0.6040 - 17s/epoch - 137ms/step\n",
            "Epoch 6/100\n",
            "121/121 - 16s - loss: 0.0474 - accuracy: 0.4952 - val_loss: 1.1302 - val_accuracy: 0.6384 - 16s/epoch - 129ms/step\n",
            "Epoch 7/100\n",
            "121/121 - 16s - loss: 0.0434 - accuracy: 0.5287 - val_loss: 0.9960 - val_accuracy: 0.6998 - 16s/epoch - 135ms/step\n",
            "Epoch 8/100\n",
            "121/121 - 16s - loss: 0.0397 - accuracy: 0.5654 - val_loss: 0.8540 - val_accuracy: 0.7253 - 16s/epoch - 132ms/step\n",
            "Epoch 9/100\n",
            "121/121 - 16s - loss: 0.0361 - accuracy: 0.6003 - val_loss: 0.7326 - val_accuracy: 0.7781 - 16s/epoch - 132ms/step\n",
            "Epoch 10/100\n",
            "121/121 - 17s - loss: 0.0328 - accuracy: 0.6371 - val_loss: 0.6883 - val_accuracy: 0.7839 - 17s/epoch - 140ms/step\n",
            "Epoch 11/100\n",
            "121/121 - 16s - loss: 0.0304 - accuracy: 0.6571 - val_loss: 0.5763 - val_accuracy: 0.8252 - 16s/epoch - 130ms/step\n",
            "Epoch 12/100\n",
            "121/121 - 16s - loss: 0.0276 - accuracy: 0.6862 - val_loss: 0.5009 - val_accuracy: 0.8586 - 16s/epoch - 130ms/step\n",
            "Epoch 13/100\n",
            "121/121 - 16s - loss: 0.0258 - accuracy: 0.7038 - val_loss: 0.4489 - val_accuracy: 0.8751 - 16s/epoch - 128ms/step\n",
            "Epoch 14/100\n",
            "121/121 - 16s - loss: 0.0242 - accuracy: 0.7257 - val_loss: 0.4200 - val_accuracy: 0.8841 - 16s/epoch - 130ms/step\n",
            "Epoch 15/100\n",
            "121/121 - 16s - loss: 0.0224 - accuracy: 0.7445 - val_loss: 0.3641 - val_accuracy: 0.9006 - 16s/epoch - 130ms/step\n",
            "Epoch 16/100\n",
            "121/121 - 17s - loss: 0.0208 - accuracy: 0.7609 - val_loss: 0.3420 - val_accuracy: 0.8991 - 17s/epoch - 140ms/step\n",
            "Epoch 17/100\n",
            "121/121 - 16s - loss: 0.0196 - accuracy: 0.7751 - val_loss: 0.3171 - val_accuracy: 0.9110 - 16s/epoch - 131ms/step\n",
            "Epoch 18/100\n",
            "121/121 - 16s - loss: 0.0185 - accuracy: 0.7836 - val_loss: 0.2894 - val_accuracy: 0.9205 - 16s/epoch - 134ms/step\n",
            "Epoch 19/100\n",
            "121/121 - 17s - loss: 0.0177 - accuracy: 0.7899 - val_loss: 0.2733 - val_accuracy: 0.9223 - 17s/epoch - 139ms/step\n",
            "Epoch 20/100\n",
            "121/121 - 17s - loss: 0.0173 - accuracy: 0.7978 - val_loss: 0.2466 - val_accuracy: 0.9307 - 17s/epoch - 139ms/step\n",
            "Epoch 21/100\n",
            "121/121 - 18s - loss: 0.0158 - accuracy: 0.8126 - val_loss: 0.2262 - val_accuracy: 0.9356 - 18s/epoch - 147ms/step\n",
            "Epoch 22/100\n",
            "121/121 - 16s - loss: 0.0155 - accuracy: 0.8167 - val_loss: 0.2222 - val_accuracy: 0.9424 - 16s/epoch - 136ms/step\n",
            "Epoch 23/100\n",
            "121/121 - 15s - loss: 0.0148 - accuracy: 0.8246 - val_loss: 0.2164 - val_accuracy: 0.9401 - 15s/epoch - 128ms/step\n",
            "Epoch 24/100\n",
            "121/121 - 15s - loss: 0.0141 - accuracy: 0.8331 - val_loss: 0.2149 - val_accuracy: 0.9409 - 15s/epoch - 127ms/step\n",
            "Epoch 25/100\n",
            "121/121 - 16s - loss: 0.0142 - accuracy: 0.8355 - val_loss: 0.1903 - val_accuracy: 0.9493 - 16s/epoch - 131ms/step\n",
            "Epoch 26/100\n",
            "121/121 - 17s - loss: 0.0135 - accuracy: 0.8386 - val_loss: 0.1795 - val_accuracy: 0.9486 - 17s/epoch - 137ms/step\n",
            "Epoch 27/100\n",
            "121/121 - 17s - loss: 0.0126 - accuracy: 0.8464 - val_loss: 0.1722 - val_accuracy: 0.9541 - 17s/epoch - 137ms/step\n",
            "Epoch 28/100\n",
            "121/121 - 16s - loss: 0.0123 - accuracy: 0.8515 - val_loss: 0.1649 - val_accuracy: 0.9578 - 16s/epoch - 133ms/step\n",
            "Epoch 29/100\n",
            "121/121 - 16s - loss: 0.0124 - accuracy: 0.8490 - val_loss: 0.1598 - val_accuracy: 0.9526 - 16s/epoch - 133ms/step\n",
            "Epoch 30/100\n",
            "121/121 - 16s - loss: 0.0117 - accuracy: 0.8610 - val_loss: 0.1439 - val_accuracy: 0.9633 - 16s/epoch - 129ms/step\n",
            "Epoch 31/100\n",
            "121/121 - 16s - loss: 0.0113 - accuracy: 0.8617 - val_loss: 0.1521 - val_accuracy: 0.9585 - 16s/epoch - 130ms/step\n",
            "Epoch 32/100\n",
            "121/121 - 17s - loss: 0.0110 - accuracy: 0.8668 - val_loss: 0.1392 - val_accuracy: 0.9640 - 17s/epoch - 140ms/step\n",
            "Epoch 33/100\n",
            "121/121 - 15s - loss: 0.0105 - accuracy: 0.8703 - val_loss: 0.1324 - val_accuracy: 0.9676 - 15s/epoch - 128ms/step\n",
            "Epoch 34/100\n",
            "121/121 - 16s - loss: 0.0106 - accuracy: 0.8715 - val_loss: 0.1316 - val_accuracy: 0.9631 - 16s/epoch - 133ms/step\n",
            "Epoch 35/100\n",
            "121/121 - 16s - loss: 0.0104 - accuracy: 0.8754 - val_loss: 0.1332 - val_accuracy: 0.9660 - 16s/epoch - 130ms/step\n",
            "Epoch 36/100\n",
            "121/121 - 16s - loss: 0.0096 - accuracy: 0.8815 - val_loss: 0.1154 - val_accuracy: 0.9700 - 16s/epoch - 133ms/step\n",
            "Epoch 37/100\n",
            "121/121 - 17s - loss: 0.0098 - accuracy: 0.8803 - val_loss: 0.1289 - val_accuracy: 0.9669 - 17s/epoch - 139ms/step\n",
            "Epoch 38/100\n",
            "121/121 - 16s - loss: 0.0097 - accuracy: 0.8814 - val_loss: 0.1268 - val_accuracy: 0.9645 - 16s/epoch - 136ms/step\n",
            "Epoch 39/100\n",
            "121/121 - 15s - loss: 0.0090 - accuracy: 0.8892 - val_loss: 0.1105 - val_accuracy: 0.9700 - 15s/epoch - 123ms/step\n",
            "Epoch 40/100\n",
            "121/121 - 16s - loss: 0.0092 - accuracy: 0.8863 - val_loss: 0.1170 - val_accuracy: 0.9699 - 16s/epoch - 131ms/step\n",
            "Epoch 41/100\n",
            "121/121 - 17s - loss: 0.0090 - accuracy: 0.8890 - val_loss: 0.1126 - val_accuracy: 0.9700 - 17s/epoch - 137ms/step\n",
            "Epoch 42/100\n",
            "121/121 - 18s - loss: 0.0087 - accuracy: 0.8918 - val_loss: 0.1030 - val_accuracy: 0.9728 - 18s/epoch - 147ms/step\n",
            "Epoch 43/100\n",
            "121/121 - 16s - loss: 0.0087 - accuracy: 0.8935 - val_loss: 0.1050 - val_accuracy: 0.9727 - 16s/epoch - 129ms/step\n",
            "Epoch 44/100\n",
            "121/121 - 16s - loss: 0.0086 - accuracy: 0.8928 - val_loss: 0.1096 - val_accuracy: 0.9680 - 16s/epoch - 128ms/step\n",
            "Epoch 45/100\n",
            "121/121 - 16s - loss: 0.0086 - accuracy: 0.8951 - val_loss: 0.0971 - val_accuracy: 0.9733 - 16s/epoch - 131ms/step\n",
            "Epoch 46/100\n",
            "121/121 - 16s - loss: 0.0080 - accuracy: 0.8995 - val_loss: 0.0949 - val_accuracy: 0.9739 - 16s/epoch - 132ms/step\n",
            "Epoch 47/100\n",
            "121/121 - 17s - loss: 0.0079 - accuracy: 0.9026 - val_loss: 0.0999 - val_accuracy: 0.9745 - 17s/epoch - 139ms/step\n",
            "Epoch 48/100\n",
            "121/121 - 18s - loss: 0.0078 - accuracy: 0.9030 - val_loss: 0.0950 - val_accuracy: 0.9762 - 18s/epoch - 148ms/step\n",
            "Epoch 49/100\n",
            "121/121 - 16s - loss: 0.0076 - accuracy: 0.9041 - val_loss: 0.0880 - val_accuracy: 0.9776 - 16s/epoch - 132ms/step\n",
            "Epoch 50/100\n",
            "121/121 - 16s - loss: 0.0077 - accuracy: 0.9061 - val_loss: 0.0894 - val_accuracy: 0.9761 - 16s/epoch - 129ms/step\n",
            "Epoch 51/100\n",
            "121/121 - 16s - loss: 0.0074 - accuracy: 0.9071 - val_loss: 0.0847 - val_accuracy: 0.9793 - 16s/epoch - 132ms/step\n",
            "Epoch 52/100\n",
            "121/121 - 17s - loss: 0.0072 - accuracy: 0.9100 - val_loss: 0.0857 - val_accuracy: 0.9771 - 17s/epoch - 138ms/step\n",
            "Epoch 53/100\n",
            "121/121 - 17s - loss: 0.0071 - accuracy: 0.9102 - val_loss: 0.0850 - val_accuracy: 0.9783 - 17s/epoch - 141ms/step\n",
            "Epoch 54/100\n",
            "121/121 - 16s - loss: 0.0072 - accuracy: 0.9092 - val_loss: 0.0819 - val_accuracy: 0.9786 - 16s/epoch - 128ms/step\n",
            "Epoch 55/100\n",
            "121/121 - 16s - loss: 0.0069 - accuracy: 0.9145 - val_loss: 0.0801 - val_accuracy: 0.9819 - 16s/epoch - 131ms/step\n",
            "Epoch 56/100\n",
            "121/121 - 16s - loss: 0.0067 - accuracy: 0.9144 - val_loss: 0.0794 - val_accuracy: 0.9812 - 16s/epoch - 131ms/step\n",
            "Epoch 57/100\n",
            "121/121 - 16s - loss: 0.0070 - accuracy: 0.9119 - val_loss: 0.0738 - val_accuracy: 0.9815 - 16s/epoch - 136ms/step\n",
            "Epoch 58/100\n",
            "121/121 - 16s - loss: 0.0066 - accuracy: 0.9178 - val_loss: 0.0759 - val_accuracy: 0.9795 - 16s/epoch - 130ms/step\n",
            "Epoch 59/100\n",
            "121/121 - 16s - loss: 0.0067 - accuracy: 0.9168 - val_loss: 0.0700 - val_accuracy: 0.9819 - 16s/epoch - 130ms/step\n",
            "Epoch 60/100\n",
            "121/121 - 16s - loss: 0.0066 - accuracy: 0.9175 - val_loss: 0.0836 - val_accuracy: 0.9772 - 16s/epoch - 134ms/step\n",
            "Epoch 61/100\n",
            "121/121 - 16s - loss: 0.0066 - accuracy: 0.9158 - val_loss: 0.0789 - val_accuracy: 0.9797 - 16s/epoch - 130ms/step\n",
            "Epoch 62/100\n",
            "121/121 - 16s - loss: 0.0067 - accuracy: 0.9154 - val_loss: 0.0745 - val_accuracy: 0.9808 - 16s/epoch - 129ms/step\n",
            "Epoch 63/100\n",
            "121/121 - 16s - loss: 0.0061 - accuracy: 0.9200 - val_loss: 0.0773 - val_accuracy: 0.9798 - 16s/epoch - 136ms/step\n",
            "Epoch 64/100\n",
            "121/121 - 16s - loss: 0.0060 - accuracy: 0.9236 - val_loss: 0.0655 - val_accuracy: 0.9832 - 16s/epoch - 130ms/step\n",
            "Epoch 65/100\n",
            "121/121 - 16s - loss: 0.0063 - accuracy: 0.9225 - val_loss: 0.0673 - val_accuracy: 0.9822 - 16s/epoch - 132ms/step\n",
            "Epoch 66/100\n",
            "121/121 - 16s - loss: 0.0058 - accuracy: 0.9255 - val_loss: 0.0658 - val_accuracy: 0.9835 - 16s/epoch - 136ms/step\n",
            "Epoch 67/100\n",
            "121/121 - 16s - loss: 0.0059 - accuracy: 0.9247 - val_loss: 0.0725 - val_accuracy: 0.9815 - 16s/epoch - 132ms/step\n",
            "Epoch 68/100\n",
            "121/121 - 16s - loss: 0.0059 - accuracy: 0.9234 - val_loss: 0.0735 - val_accuracy: 0.9826 - 16s/epoch - 129ms/step\n",
            "Epoch 69/100\n",
            "121/121 - 17s - loss: 0.0058 - accuracy: 0.9243 - val_loss: 0.0823 - val_accuracy: 0.9795 - 17s/epoch - 143ms/step\n",
            "Epoch 70/100\n",
            "121/121 - 16s - loss: 0.0057 - accuracy: 0.9281 - val_loss: 0.0729 - val_accuracy: 0.9821 - 16s/epoch - 132ms/step\n",
            "Epoch 71/100\n",
            "121/121 - 17s - loss: 0.0057 - accuracy: 0.9284 - val_loss: 0.0629 - val_accuracy: 0.9845 - 17s/epoch - 137ms/step\n",
            "Epoch 72/100\n",
            "121/121 - 16s - loss: 0.0057 - accuracy: 0.9265 - val_loss: 0.0609 - val_accuracy: 0.9849 - 16s/epoch - 131ms/step\n",
            "Epoch 73/100\n",
            "121/121 - 15s - loss: 0.0056 - accuracy: 0.9272 - val_loss: 0.0627 - val_accuracy: 0.9842 - 15s/epoch - 126ms/step\n",
            "Epoch 74/100\n",
            "121/121 - 18s - loss: 0.0053 - accuracy: 0.9313 - val_loss: 0.0604 - val_accuracy: 0.9864 - 18s/epoch - 146ms/step\n",
            "Epoch 75/100\n",
            "121/121 - 16s - loss: 0.0055 - accuracy: 0.9302 - val_loss: 0.0625 - val_accuracy: 0.9849 - 16s/epoch - 131ms/step\n",
            "Epoch 76/100\n",
            "121/121 - 15s - loss: 0.0054 - accuracy: 0.9316 - val_loss: 0.0833 - val_accuracy: 0.9778 - 15s/epoch - 128ms/step\n",
            "Epoch 77/100\n",
            "121/121 - 15s - loss: 0.0052 - accuracy: 0.9314 - val_loss: 0.0747 - val_accuracy: 0.9810 - 15s/epoch - 128ms/step\n",
            "Epoch 78/100\n",
            "121/121 - 16s - loss: 0.0051 - accuracy: 0.9330 - val_loss: 0.0675 - val_accuracy: 0.9829 - 16s/epoch - 132ms/step\n",
            "Epoch 79/100\n",
            "121/121 - 17s - loss: 0.0054 - accuracy: 0.9312 - val_loss: 0.0582 - val_accuracy: 0.9873 - 17s/epoch - 144ms/step\n",
            "Epoch 80/100\n",
            "121/121 - 16s - loss: 0.0052 - accuracy: 0.9320 - val_loss: 0.0697 - val_accuracy: 0.9812 - 16s/epoch - 130ms/step\n",
            "Epoch 81/100\n",
            "121/121 - 16s - loss: 0.0052 - accuracy: 0.9328 - val_loss: 0.0611 - val_accuracy: 0.9859 - 16s/epoch - 135ms/step\n",
            "Epoch 82/100\n",
            "121/121 - 16s - loss: 0.0051 - accuracy: 0.9327 - val_loss: 0.0788 - val_accuracy: 0.9812 - 16s/epoch - 132ms/step\n",
            "Epoch 83/100\n",
            "121/121 - 16s - loss: 0.0049 - accuracy: 0.9356 - val_loss: 0.0567 - val_accuracy: 0.9883 - 16s/epoch - 129ms/step\n",
            "Epoch 84/100\n",
            "121/121 - 17s - loss: 0.0048 - accuracy: 0.9366 - val_loss: 0.0591 - val_accuracy: 0.9864 - 17s/epoch - 138ms/step\n",
            "Epoch 85/100\n",
            "121/121 - 16s - loss: 0.0046 - accuracy: 0.9390 - val_loss: 0.0548 - val_accuracy: 0.9873 - 16s/epoch - 129ms/step\n",
            "Epoch 86/100\n",
            "121/121 - 16s - loss: 0.0045 - accuracy: 0.9416 - val_loss: 0.0549 - val_accuracy: 0.9866 - 16s/epoch - 133ms/step\n",
            "Epoch 87/100\n",
            "121/121 - 16s - loss: 0.0047 - accuracy: 0.9390 - val_loss: 0.0625 - val_accuracy: 0.9854 - 16s/epoch - 136ms/step\n",
            "Epoch 88/100\n",
            "121/121 - 16s - loss: 0.0045 - accuracy: 0.9402 - val_loss: 0.0578 - val_accuracy: 0.9854 - 16s/epoch - 130ms/step\n",
            "Epoch 89/100\n",
            "121/121 - 17s - loss: 0.0046 - accuracy: 0.9408 - val_loss: 0.0523 - val_accuracy: 0.9885 - 17s/epoch - 137ms/step\n",
            "Epoch 90/100\n",
            "121/121 - 17s - loss: 0.0046 - accuracy: 0.9403 - val_loss: 0.0558 - val_accuracy: 0.9882 - 17s/epoch - 137ms/step\n",
            "Epoch 91/100\n",
            "121/121 - 16s - loss: 0.0046 - accuracy: 0.9407 - val_loss: 0.0579 - val_accuracy: 0.9863 - 16s/epoch - 134ms/step\n",
            "Epoch 92/100\n",
            "121/121 - 15s - loss: 0.0043 - accuracy: 0.9435 - val_loss: 0.0505 - val_accuracy: 0.9893 - 15s/epoch - 127ms/step\n",
            "Epoch 93/100\n",
            "121/121 - 16s - loss: 0.0047 - accuracy: 0.9412 - val_loss: 0.0550 - val_accuracy: 0.9880 - 16s/epoch - 130ms/step\n",
            "Epoch 94/100\n",
            "121/121 - 16s - loss: 0.0045 - accuracy: 0.9404 - val_loss: 0.0564 - val_accuracy: 0.9884 - 16s/epoch - 129ms/step\n",
            "Epoch 95/100\n",
            "121/121 - 17s - loss: 0.0046 - accuracy: 0.9414 - val_loss: 0.0511 - val_accuracy: 0.9895 - 17s/epoch - 140ms/step\n",
            "Epoch 96/100\n",
            "121/121 - 16s - loss: 0.0042 - accuracy: 0.9428 - val_loss: 0.0501 - val_accuracy: 0.9894 - 16s/epoch - 132ms/step\n",
            "Epoch 97/100\n",
            "121/121 - 16s - loss: 0.0041 - accuracy: 0.9476 - val_loss: 0.0455 - val_accuracy: 0.9912 - 16s/epoch - 136ms/step\n",
            "Epoch 98/100\n",
            "121/121 - 16s - loss: 0.0044 - accuracy: 0.9427 - val_loss: 0.0456 - val_accuracy: 0.9889 - 16s/epoch - 130ms/step\n",
            "Epoch 99/100\n",
            "121/121 - 16s - loss: 0.0041 - accuracy: 0.9457 - val_loss: 0.0472 - val_accuracy: 0.9900 - 16s/epoch - 134ms/step\n",
            "Epoch 100/100\n",
            "121/121 - 18s - loss: 0.0041 - accuracy: 0.9461 - val_loss: 0.0490 - val_accuracy: 0.9888 - 18s/epoch - 145ms/step\n"
          ]
        }
      ],
      "source": [
        "history = model.fit_generator(train_gen,\n",
        "                              steps_per_epoch=len(X) // batch_size,\n",
        "                              validation_data=val_gen,\n",
        "                              validation_steps=len(Xt) // batch_size,\n",
        "                              class_weight=class_weight,\n",
        "                              epochs=100, \n",
        "                              verbose=2)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rQYvzGC5cuDU",
        "outputId": "f6ad1fa1-c371-4695-aadc-e6cc369bfa01"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Best validation accuracy: 0.991\n"
          ]
        }
      ],
      "source": [
        "print(\"Best validation accuracy: {:0.3f}\"\n",
        "       .format(np.max(history.history['val_accuracy'])))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 279
        },
        "id": "iJeKTTOPq_Tl",
        "outputId": "77fcbd78-b629-4305-daf3-2fb6e4877559"
      },
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEGCAYAAABo25JHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3dd3xc5Zno8d8zGmnUi1VsS7It27hiGxuEaYZQQrCBUHZTqIGExdxcsmF3WRJy00huWXKT5SbsEgKbOEAg1ITgBBMceggGLBdw70XFKlaXRjOjmXnvH+/Ylo1kj+wp1pzn+/nMZzRnzjnznDn2+8x52xFjDEoppZzLlewAlFJKJZcmAqWUcjhNBEop5XCaCJRSyuE0ESillMO5kx3AcJWUlJiqqqpkh6GUUiPKqlWr9htjSgd7b8QlgqqqKmpqapIdhlJKjSgismeo97RqSCmlHE4TgVJKOZwmAqWUcri4JQIRWSIizSKyfoj3RUQeFJHtIvKxiJwer1iUUkoNLZ5XBI8BC4/y/iJgSuSxGHg4jrEopZQaQtwSgTHmHaDtKKtcDTxhrPeBQhEZG694lFJKDS6ZbQQVQO2A13WRZZ8gIotFpEZEalpaWhISnFJKOcWIGEdgjHkUeBSgurpa581WSp1cgn7o90JmIYgM8n4A+trA3w2IXScchM5a6KgFbysUVUHJFPsc8IJ3P/R1QGY+5JRBTgmkpccl/GQmgnpg3IDXlZFlSqmRxBjwdUB/ny0Qw0FweyAjF9IyoKseWrdD2y5Iz4L8cvsom3nsgi0chvpVsGUZmBDM+xKUnHLoc5s32cK0agFk5NjlPS2w4j9g8zLIHW0L1vyxNj5fJwR6IM0DGdl2m+ziAQVtRmTfIWjeDHUr7ecf2CYtw65XNMHu199t329cD+F+cKXb9zNyIeSHUL8t1P2dsfmur3gAzrwtNvsaIJmJYCnwNRF5BjgL6DTG7EtiPEoljjHQWQe1H9hC0p1pCyV3pi1Iw0G7XuF4GDXJFpzte6Blk32uOB0mLIC0Af+Fg37obbG/Lr2ttlDKKoKsQrvN3vdg7/vg64osLwKM3aa3BQwwqgqKJkL2KOhrt/sJh6F8HlRWw6iJsH8bNK6Dpg22IG7ZbH/tDlfBOFjwzzDvJps4Ar3QsBb2b4H23faxZwX0NoOkgbjgbz+DyZfAmFmw+WX73YH93k75NOSNhTVP2kJ40kV2nzteh+59kJ4NmQX2ew4GoL8X/D123aEUjofKM22yCPntdr3NsO8j2PRH+7nl8+CcOyG37NB3Gei1icOdYT83u8R+p578yPkP2WMqqITCcZA1Cjr2QMsW6NgLnly7TWa+PV8H9ls+b/jfcxQkXncoE5GngQuBEqAJ+D6QDmCM+YWICPCf2J5FXuDLxphjzh1RXV1tdIoJlVCBXuish6466G60j55mW3jll0PeGPu69gP7CPqhZKq9zM8ssP+Be1rA32V/IYaDdv3uhhOLK2sUTL3MVkk0bYS2HWDCR9+mdIYtsHwdtqA3QG6p/UVswtC+yxbAoYD99ZtdYuPtbf7kvjwFUDYDSqfZY83Itd+JKx2CPhtXvxfyK6B4sk0wQR907YO2nfDhI/YXd165jaFxvS0gwX524XgYMxumXQ5TLrWF8OrHoebX0NMEE8+HGVfZ5LTlFVsw9zTDnC/A+XfbmA4Ih8E1RJOovydS0O4/lIDBJuC80UN/l+EQIEPv9yQjIquMMdWDvjfSblWpiUBFJdBrC5bGj22hA/YXZUYOTL4YKufb/8D9PlvtsGUZeNtswRXosZfz/V67H3/XJ/efnmMLy3D/oWV5Y2HcWbZA3L/V/rIN9EJOqX1kFtiqEFe6/buy2q5fNtPuJ+CFYB+43PYRDtlqj7adtnqlYDyUTbcF6+53YdNS2P6a/WVfNtM+Cirtr9fsUbZQ62u3j9zR9rOyRx37uwuH7LFn5Nq6bGPs59ettFcWJVPtL/KCcYPXh0fLGNj5Frz3HzbWcfPtr+/Rp9rkMFQBGwrahOLJPSLusP2V78k7/phSmCYCdXIIh6FpvS1QymbYgsmVduh9bxvsetsWbjvesoVm5Zn2UT7X/vLMLLD1vOtesFUATettAS9pkQa4kP1lO/ByPz3Hfk44ZAtaE7aFduWZ9vN8nfYXceE4exmfkXPoOSPH/oLOr4SCCrtd7mhbCIXDtuqke5+Nq3D84QXjgf9bJ1JYKhUjR0sEI6LXkBohuhpsAV15ZqSQd9nL+W2vwrrnYdc79tfpATmlMCVStdGwxlZLgC1UJ11oC+xd78C65w5tk1du9xHsg7JT4aw7ALHrGmM/U9JsAT5mNoyZY6tvDhTGvk7Yuhw2vQS1H8LUhXDa9TDxgsOTUjRcLlulkTvozL6aANSIoYlADU84bBssu/bZnhrpmXb5tr/A7xcfajTMGwsTzrWX/t5W+yt62hW2XrfyTNvYtvlPtl43qwDGzoXTb7YNoBVnHGoEPdCo2rT+UMOkJw/m3gDlpw+/sM0sgDmftw+lFKCJQEWj3xcptJfaumlvq13uyYeZV9u65A8etr/Qb/49tO6ADS/CjjdtwT/3JlsvP7CHS/FkmPV3x/5sEVtlUzgOpi2Kz/Ep5XCaCNQhxtguefWrbB14Rh60boOPn7O9TPLKYcpnoOp8W2++4UX7CPTAGbfCwvttP/HyeTD7c8k+GqVGhHDYsL/HT2meB0lSdaImAic5Wne35k3wyjdsnby4DnVDTPPAjM/aapuqCw7fdsqlcPlPbGNp8eSEHIJSJxt/MMTWxh78wRCnlOVSmJ1x1PWNMXT7g2xt7GbZukaWrdtHY5ePSSU5XDFnLJ+aWkp9Rx8f1XayubGLXn8QfzCMPxjmnz49havnDjoTzwnRROAE3jb44BfwwSO2213xlEi/7xz7a76vwyYAT54t2M/4su3P7e+2A2aO7KY3UEa2JgE1IgSCYXbu72FrUw9bGrvYvK+bXft7GVOQybQxeUwdnUfYGNp7A7T19tPj78cbCNEXCCEieNwuPG77QygQCtMfClPX3sfWpm76Q4d6X5bkeqgsyiLX4ybHk4bb5aLL109nXz/t3gAt3X58/faHVkaaiwumlnLLuVW8u72Fh97czn+8YQfJedwupo/NpygnA4/bRWZ6GiW5nrh8N9p9NFUZA/Wr4eNnYc1vbM+caVdERoZutY+g3xb+Gbm2AffCe6PrZ65UAgRDYQyQnnb0AVvhsEEERIQef5AN9Z2sq+9kW1MP7d4AHX397O/xs7fVSzBsyzu3S5hUmsOkklz2dfnY2thNX3/o4D5zMtLIy0wnKyONrPQ0DBAIhvAHw4jYmNJdLsryPcyqKGBWeQFZGS62N/ewvbmHfZ0+evxBvP4QwXCY/Kx0CrLSKcxKpyw/k9JIslgwpYS8zEPTbOzv8VOzu43xo3KYOjoX9zGOfTi0+6iT+Drhvf+0CaBjjx2heeq1dih/2YxkR6dSSDhsCIYNoUjhmpUxePfb5m4fb29pYcUO28lgVE4GRTkZ9AVCNHX5aOr20x8M404T3C6h2xekoaOPxi4fIsLYgkzGFWWTl+mm2xek299Pjy8Y+TtIIDj4aOqS3AyKczwUZKcztSyPRbPGMHW0/eU/qTQHj/tQvOGwob6jjwy3i8Ls9MPeG46Lpx9lJHIUSnI9LJyV+Nn4NRGkinAYPvotvHafHSo/+WL41Ddh+hV2rhmlsAUegMtlGyWNMezr9LG2toPmLh+5menkZbrJz0ynNC+DklwPOR43Pb4gXb5+6tr7eG/Hft7dtp919Z2EB1Qo5Ge6GTcqm7EFWYTCYXr9Idq8AbY39wC2YPa402jrDdDXHyLNJZTmeijL9+Bxu/AFDcGQIceTxjmTS6gozCRsoLbdy942L22tAfIy3ZTmephUkkteppvcTLf9xW7ssXjS05gxNo9ZFQWU5WVG/b24XMK4Udmx+6JHGE0EI1Wo3/bBb91uf/k3rLV97Svnw43Px21yKjWyGGPY3tzDX7ftZ8XOVj7Y2UqPP0hhdgajcjLo7Ounpfsok64NIs0lzBtXyOILJpPrScOd5iJsDPs6fNS2e6lr95Ke5iLHk0ZVcQ7XzqvgU1NLObU8/2CvGF9/iPQ0F2kuHXR3MtBEMBLt+xheutPOowOQO8ZOi3vtIzDnizqidQQxxtDutfMVuQQy3C6yMw79t+zy9fPi6nr+vL4Rd5qQn5VOfqabQNDQ1x+kLxCiKCeDisIsyguzyMt02wJWhJo97Szf0MjO/b0AjB+VzaJZYynL99DWG6CtN0BWehqnjStk7rhCKoqy8PpDdPv76fT209LjZ39PgF5/8OBVQkmeh9PHFx5Wr308MtOPr+pFxYcmgpEk1A/v/Bj++u92orHPPwZTFx0a3atOKn2BEPt7/ISNrUd3u1zkZ7nJy0ynqcvHi2vq+f3qOna09B62XXmkF0t+VjrLNzTR1x9i+pg8MtPTqO/oo9sXJCPNRXZGGp50F1sau2ns8h1WTQO2QfTsScV8ecFELppWSmVRFFUfR+kgplKXJoKRwtsGz99qJ0mbcx0s/Dft4ZNkxhj29wTY29ZLS7efDm8/HX397Gzp4eO6TrY19xxsSB3KmVVFfKF6HJnpaYSNodcfZFtzD1sau1lb28HVc8u58awJzK4sOOp+gqEwTd1+vP5gpGujoao4+5h92pUCTQQnL1+X7dbpctkbgfz2i/aGFdc8bOfZUXHV2uNnbW0HH9d10uXrJxAMEwiG6fL10+7tp8MboL69j95A6BPbFmWnM6eykEtnjmb8qGzSXEKaSwgEw3RHGl097jSumD2W8cWxaaB0p7moKMyKyb6U82giOBm9/j/hrz+x89bnjbGzbbo9cMsfYcI5yY4uJQVDYVbubucvG5t4Y3MTu1u9gK23z8lwk+F2keF2kZfppjA7g4klOZw7uYQJxdlUFedQlu+hKDuDwux0stLTkjZVgFLHQxPByaZmiU0CM6+2d0jq2mdH+V78XdsgrIalrTfAlsZuGjr62NfZR2tvAI87jeyMNNJcwt5WLztaetjS1G3r3t0uzptczPXzxzN3XCGzKwsOa7xVKhXpv/CTydbl8PLddo7+v19y+Gyd6jCBYJhnV+5lS1M3M8cWMLuigMLsdLY0drNpXxcbGrpYV99JfUffYdvlety2midkByGV5HqYXJrDZ08r5/xTSrhgaik5Hv3elbPov/iTxe53bWPwmNnwOU0CB6Y+ObKKJRw2/PHjBn6yfAu1bX1kZ6TxZGDvJ7avKs5m3vhCbjl3AjPHFlBRlMXYgsyD3RaDoTDBsNFujEqhiSD5vG3w2vdh9RP25t43PHf0Sd5SXEu3n2c+3MtvP9xLW2/ATkeQnYEB2nsDtHsD+INhZozN5/GvzOaCKSXUtvWxrr6Tjr4A08fkMW1MPrnH+FXvTnNxnLMIKJVyNBEk0/bXI3f1aodzv24nfcvISXZUCdfY6eOv21p4a0sLyzc20h8ynD+lhBlj8yMzQQYQgVnl+YzKyWB2ZQGXzxp7cJqE8cXZMet9o5QTaSJIlvrV8OxN9irgS3+wVUIprqGjjw93tbFqTzuNXT7aeu2UvHvbbA+dklwPN541gZvPmcDkUudeFSmVaJoIkqF9jx0XkF0CN78IeSc2Y+HJbHNjFy+tbWDZun3siXTJzPW4qSzKYlROBnMqC7jxrPFcMLWU6WPytNulUkmgiSDR+jrgqc9DyA+3/imlksDu/b384u0dtPUG6A0Eaez0saOllzSXsOCUEm45p4r5E0cxY2y+Tjam1ElEE0GivfINaNtpq4NKpyU7mphZsaOVrz61ikAwzPhR2eR43EwozuGWc6u4YvZYiuN0ZyWl1InTRJBIXftg/e9g/mKoWpDsaI6LNxBk+YYmNjR0Mqk0l2lj8ti8r5vvvbSeCcXZLLn1TCYUO6/BW6mRTBNBItUssTeQn397siMZto/rOvj133bz6oZGvIEQbpccvO0fwPlTSvjPG06nIOvEpidWSiWeJoJECfptIpi60E4dMULs2t/LT17dwsvr9pGX6ebqueVcM7eC6qpRNHT0sWlfF95AiCvnjI3p/VWVUomjiSBR1v8evPvhrDuSHclRtfcGWN/QyZq9Haze28672/aT4Xbx9UumcPv5Ew+7Icm4UdmOvr2fUqlCE0EiGAMf/AJKpsGkC5MdzWF8/SEee283b21pZntzD/t7AoC9ydnUsjxuPbeKOz41mdI8bexVKlVpIkiE2g9h31q44oGT6jaSr29q4od/2sieVi9zKgu4ZPpopozOZfqYfOaMKyD/BG9HqJQaGTQRJMKHj4KnAE67LtmRALC1qZv/s2wTb21pYXJpDr+5bT7nTylNdlhKqSTRRBBv/h7Y/LK9q1iS5xFq7PTx09e28lxNLTkeN9++fAa3nFtFhlsbeZVyMk0E8bZlGQT7YPbnkvLx25u7WbaukTe3NLO2tgO3S7jl3Cq+fvEUinL0frZKqTgnAhFZCPwMSAN+aYy5/4j3xwOPA4WRde41xiyLZ0wJt+4FyK+AcWcn/KPf2NzEHb9ZRTBsmFNZyD9dMpVr5pXrgC+l1GHilghEJA14CLgUqANWishSY8zGAat9B3jOGPOwiMwElgFV8Yop4bxtsON1OPur9ib0CfTO1hb+229WM31MPr+6pZqy/MyEfr5SauSIZ+k0H9hujNlpjAkAzwBXH7GOAfIjfxcADXGMJ/E2LYVwEGYltlrove37uf2JGiaX5fKb2+ZrElBKHVU8q4YqgNoBr+uAs45Y5z5guYj8I5ADfHqwHYnIYmAxwPjx42MeaNysewGKT4GxpyXk4zr7+vn5W9v59bu7qSrJ5snb5lOYre0ASqmjS3Z3keuBx4wxlcDlwG9E5BMxGWMeNcZUG2OqS0tHSDfHrn32PsSzPhf3sQN9gRC//tsuLvzxmzz6zk6uPG0sv739bJ3xUykVlXheEdQD4wa8rowsG+g2YCGAMWaFiGQCJUBzHONKjA0vAiauvYUaOvp4YsUenv5wL519/Zw7uZj/cfkMZlUUxO0zlVKpJ56JYCUwRUQmYhPAdcANR6yzF7gEeExEZgCZQEscY4q/pg1Q82v46GlbJVQyJeYf0eMP8uDr21jy7i7CxnDZqWP48nkTObOqSO/wpZQatrglAmNMUES+BryK7Rq6xBizQUR+CNQYY5YCdwP/JSL/jG04vtUYY4be60nuD3fC2ichzQOnXgMXfCOmuzfGsPSjBv73y5to7vbzhepK/vHiKTrxm1LqhMR1HEFkTMCyI5Z9b8DfG4Hz4hlDwnjbYO1TMPsLsOhHkD0qprs3xvC/Xt7Er97dxZzKAh79UjVzxxXG9DOUUs6kI4tjZdfbgIEz/yEuSeAHf9zIY+/t5tZzq/julTP1nr9KqZjRRBArO94ETz5UnBHT3YbDhu8tXc+T7+/lHxZM5NtXzNB2AKVUTGkiiAVjbCKYeAGkxfYr/dGfN/Pk+3u541OTuHfhdE0CSqmYS/Y4gtTQugM698Lki2K62ydW7OaRd3Zy89kTNAkopeJGE0Es7HzTPk+KXSL4y8Ym7lu6gU/PKOO+q07VJKCUihtNBLGw4w0onBCzm9Kvq+vkH59ezeyKAh68fp42DCul4koTwYkK9cOuv8Lki2MylUSnt5+vPrWK4hwPv7zlTLIztBlHKRVfWsqcqLoaCHTHpH0gHDbc/fxamrp8PP/fztUbxiulEkKvCE7UjjdAXLbH0Al69K87eW1TM9++fIYOFlNKJYwmghO14w0oPx2yik5oNyt3t/HjV7dwxeyx3HJuVWxiU0qpKGgiOBF7P4D6Gph+xQntxhsIcvdzH1FZlMX9fz9bewgppRJK2wiOVzgMr34LcsfAWXec0K7+fflW9rZ5eWbx2eRlpscoQKWUio4mguO1/ndQvwqu/jlkHP/N4FftaWfJ33Zx89kTOHtScQwDVEqp6GjV0PHo74PX7rP3Gzjt+uPeja8/xDde+Ijygiy+uWh67OJTSqlh0CuC47HiIeiqg797BFzHn0v/75+3sKOllye+Mp9cj54KpVRy6BXBcPX74N2fwvQroWrBce/md6vqWPK3Xdx6bhUXTB0h92FWSqUkTQTD1bDGDiCbe+RdN6O3traDb724jnMmFfPtK2bEMDillBo+TQTDtXeFfR531nFt3tTlY/ETNYzO9/DzG08nPU1PgVIqubQUGq7aD6BkKuSUHNfm335xHT3+IP/1pWqKcjJiHJxSSg2fJoLhCIdh7/vHfTWwvr6T1zY189VPTWb6mPwYB6eUUsdHE8Fw7N8Cvg4Yf85xbf7g69vIy3Rzy3lVsY1LKaVOgCaC4TjQPjD+7GFvurGhi+Ubm/jKeRPJ19HDSqmTiCaC4dj7AeSUHdcNaP7zzW3ketx85byJcQhMKaWOnyaC4di7wl4NDHNSuC2N3Sxb18it51ZRkK1XA0qpk4smgmh17YOOPcOuFjLG8ONXt5CdkcZtC/RqQCl18tFEEK3a9+3zMBPBczW1vLapibsumaLdRZVSJyVNBNHa+z6kZ8OYOVFvsqOlh/uWbuS8U4q5/fzY3NheKaViTRNBtPaugMpqSIuujt8fDPH1p9eQme7i3z8/F5dLbzajlDo5aSKIRl87NK6DcdFXC/30tW1saOjiR38/hzEFmXEMTimlTowmgmisfRpMGGZcGdXq3b5+Hvvbbq6eW85nTh0T5+CUUurEaCI4lnAYVv4SKufbG9FE4U8f76OvP6Q3oVdKjQiaCI5l11vQtgPm3x71Js+srGXq6FzmjSuMX1xKKRUjUSUCEfm9iFwhIs5LHB/+ErJLYObVUa2+ubGLj2o7+OKZ45FhDjxTSqlkiLZg/zlwA7BNRO4XkWlxjOnk0VELW1+B078Ebk9Umzy7spaMNBfXzquIc3BKKRUbUSUCY8xrxpgbgdOB3cBrIvKeiHxZRIbsTykiC0Vki4hsF5F7h1jnCyKyUUQ2iMhvj+cg4qZmiX2u/nJUq/v6Q7y4pp7PnDqaUTp4TCk1QkR9x3QRKQZuAm4G1gBPAQuAW4ALB1k/DXgIuBSoA1aKyFJjzMYB60wBvgWcZ4xpF5Gy4z+UGAv6YfUTMHUhFI6PapPlG5vo8PbzxTPHxTk4pZSKnagSgYi8CEwDfgN81hizL/LWsyJSM8Rm84HtxpidkX08A1wNbBywzu3AQ8aYdgBjTPPwDyFO6mrAux/m3hj1Js98uJeKwizOm3x8dy9TSqlkiLaN4EFjzExjzL8NSAIAGGOqh9imAqgd8LousmygqcBUEfmbiLwvIgsH25GILBaRGhGpaWlpiTLkE9Sw2j6Pmx/V6jtbenhvRyvXnTlORxErpUaUaBPBTBE52BdSRIpE5L/H4PPdwBRs1dL1wH8N/JwDjDGPGmOqjTHVpaWlMfjYKDSsgfxKyI2utuqpD/bidglfnK/VQkqpkSXaRHC7MabjwItIVc6xOtbXAwNLxcrIsoHqgKXGmH5jzC5gKzYxJF/9aqiYF9WqfYEQL6yq47JZYyjL0+kklFIjS7SJIE0GdIqPNAQfq1vMSmCKiEwUkQzgOmDpEev8gUhDs4iUYKuKdkYZU/z0tUP7LiiPLhH88eMGOvv6uemsCXEOTCmlYi/aXkN/xjYMPxJ5fUdk2ZCMMUER+RrwKpAGLDHGbBCRHwI1xpilkfc+IyIbgRBwjzGm9XgOJKYa1tjn8tOjWv2p9/dwSlkuZ08aFceglFIqPqJNBN/EFv5fjbz+C/DLY21kjFkGLDti2fcG/G2Af4k8Th71kYbi8rnHXPXjug4+quvkvs/O1JHESqkRKapEYIwJAw9HHqmvYY29QX1W0TFXffL9PWSlp/F3Z1QmIDCllIq9aMcRTAH+DZgJHGwNNcak5m23GtbA+HOOuVpnXz9LP2rgmrkV5GfqTemVUiNTtI3Fv8ZeDQSBi4AngCfjFVRSdTdBV31UDcVLP2rA1x/m+vnRjTxWSqmTUbSJIMsY8zogxpg9xpj7gCviF1YSHWgorjh2Q/GzK/cyY2w+cyoL4hyUUkrFT7SJwB+ZgnqbiHxNRK4FcuMYV/I0rAFxHfMm9evrO1lf38V1Z47TRmKl1IgWbSK4C8gGvg6cgZ187pZ4BZVUDauhZBp4jp7nnl1ZS4bbxTVzdbpppdTIdszG4sjgsS8aY/4V6AGim5N5JDLGdh2detlRV+sLhPjD2nounzWGgmxtJFZKjWzHvCIwxoSw002nvs46O+PoMRqKX1m/j25fkOu0kVgplQKiHVC2RkSWAs8DvQcWGmN+H5eokqVpvX0+xk3qn1lZS1VxNmdN1JHESqmRL9pEkAm0AhcPWGaA1EoELZvtc+nQd+Lc2tTNh7va+ObC6dpIrJRKCdGOLE7ddoGBWrZAXjlkDt0d9LH3duNxu/QuZEqplBHtyOJfY68ADmOM+UrMI0qmls1HvRro9Pbz4up6rplbofckVkqljGirhv404O9M4FqgIfbhJFE4DC1b4fQvDbnKczW19PWHuOXcqsTFpZRScRZt1dDvBr4WkaeBd+MSUbJ01UF/75BXBKGw4fEVu5k/cRQzy/MTG5tSSsVRtAPKjjQFiO4ejiNFyxb7XDp90Ldf39REXXsfX9arAaVUiom2jaCbw9sIGrH3KEgdx+gx9PiK3ZQXZHLpzNGJi0kppRIg2qqhvHgHknQtmyGnFLI/OTagts3L37a3cs9l03CnHe9FlFJKnZyiKtVE5FoRKRjwulBErolfWEnQsuWo1UIAV8wem8iIlFIqIaL9eft9Y0zngRfGmA7g+/EJKQmMiSSCwauFXt/czKTSHKpKchIcmFJKxV+0iWCw9aLtenry694H/q5Brwh6/UE+2NnGJdNTq21cKaUOiDYR1IjIAyIyOfJ4AFgVz8AS6igNxe9u308gFOYiTQRKqRQVbSL4RyAAPAs8A/iAO+MVVMIdpevoG5uayfO4ObNKJ5hTSqWmaHsN9QL3xjmW5GnZDFlFttfQAOGw4Y0tzVwwrZR07S2klEpR0fYa+ouIFA54XSQir8YvrAQ70GPoiNlENzR00dLt1/YBpVRKi/ZnbkmkpxAAxph2UmVksTHQvGnQ9vE4oUAAABJ3SURBVIHXNzchAhdOS41DVUqpwUSbCMIicvB2XCJSxSCzkY5IvS3g6xi8fWBzM6ePL9KZRpVSKS3aLqDfBt4VkbcBAc4HFsctqkQaosdQc7ePj+s6ueeyoaelVkqpVBBtY/GfRaQaW/ivAf4A9MUzsITp2mefCw6/0cyKHa0AXDCl9MgtlFIqpUQ76dw/AHcBlcBa4GxgBYffunJk6muzz9nFhy1esaOV/Ey3TjmtlEp50bYR3AWcCewxxlwEzAM6jr7JCOFtBXFBZuFhi1fsbGX+xGLSXHpfYqVUaos2EfiMMT4AEfEYYzYDqVF57m21Ywhch76Kho4+9rR6OWdy8VE2VEqp1BBtY3FdZBzBH4C/iEg7sCd+YSWQt23QaiGAcyZpIlBKpb5oG4uvjfx5n4i8CRQAf45bVInkbYWsw6ePWLGzlaLsdKaPSf3bMCil1LBnEDXGvB2PQJLG2wZFVYctWrGjlbMmFuPS9gGllAPEdQIdEVkoIltEZLuIDDlXkYj8vYiYSBfVxOprO+yuZLVtXuo7+rR9QCnlGHFLBCKSBjwELAJmAteLyMxB1svD9kr6IF6xDMkYWzU0IBEcbB/QRKCUcoh4XhHMB7YbY3YaYwLY6auvHmS9/wn8CDu1dWIFeiAUOKyxeMXOVkpyM5hSlpvwcJRSKhnimQgqgNoBr+siyw4SkdOBccaYl4+2IxFZLCI1IlLT0tISuwi9hw8mM8bY9oFJxYho+4BSyhmSNsm+iLiAB4C7j7WuMeZRY0y1Maa6tDSGUz54bTXQgUSwp9VLY5dPu40qpRwlnomgHhg4gU9lZNkBecAs4C0R2Y2dtmJpQhuMD1wRRLqPbm7sAuC0ysKhtlBKqZQTz0SwEpgiIhNFJAO4Dlh64E1jTKcxpsQYU2WMqQLeB64yxtTEMabDHTHP0O5WLwATSrITFoJSSiVb3BKBMSYIfA14FdgEPGeM2SAiPxSRq+L1ucNysGrIXhHs3t9LcU4G+ZnpSQxKKaUSa9gDyobDGLMMWHbEsu8Nse6F8YxlUEdMOLe7tZcJxXo1oJRyFmffkd3bdtiEc3tavVSV5CQ5KKWUSiyHJ4LWg+0Dvv4Q+zp9VBVrIlBKOYsmgkiPoT0HGoq1akgp5TDOTgR97QN6DPUC6BWBUspxnJ0IBswztEcTgVLKoZybCI6YcG53q5ei7HQKsrXrqFLKWZybCAK9h004t3t/LxP0akAp5UDOTQSDzDNUpQ3FSikH0kSQNQpff4iGzj4dQ6CUciTnJoIB8wzVtXsxRhuKlVLO5NxEMOBeBLv26xgCpZRzOTgRHJpwTruOKqWczMGJoO3ghHO7W3spyEqnKCcj2VEppVTCOTgRtB6ccE57DCmlnMzZiSDSdXSXjiFQSjmYcxNBXxtkjcIfDNHQ0adXBEopx3JuIvC2RbqO9hE26BgCpZRjOTgRtB7WY0i7jiqlnMqZicCYyBXBKBo7/QCMKchKclBKKZUczkwEgV4I+SG7mOZuHwCluZ4kB6WUUsnhzEQwYMK55m4/RdnpZLid+VUopZQzS78D8wxljaK5y09ZXmZy41FKqSRyZiIYcEXQ0u2jLF+rhZRSzuXQRHBgwrlRNHfrFYFSytmcmQh6WwAIZ5XQ0u3XKwKllKM5MxH0NENaBu3hLIJhQ1meJgKllHM5NxHkjqa5JwCgVUNKKUdzaCJogpxSmrvtYDKtGlJKOZkzE0Fv5Iqgyw4m06ohpZSTOTMR9DRD7oArAq0aUko5mPMSQTgEvfshdzQt3X7yPG6yMtKSHZVSSiWN8xKBtw1MCHLKaO72UartA0oph3NeIuhpss+5ZZHpJTQRKKWczXmJoLfZPueO1lHFSilFnBOBiCwUkS0isl1E7h3k/X8RkY0i8rGIvC4iE+IZD2AbigGTU0pzt0+vCJRSjhe3RCAiacBDwCJgJnC9iMw8YrU1QLUxZg7wAvB/4xXPQZGqoe70Ufj6wzqGQCnlePG8IpgPbDfG7DTGBIBngKsHrmCMedMY4428fB+ojGM8Vk8zuLNo9qUD2nVUKaXccdx3BVA74HUdcNZR1r8NeGWwN0RkMbAYYPz48ScWVU+zbSjuOTCGQK8IlHKC/v5+6urq8Pl8yQ4lrjIzM6msrCQ9PT3qbeKZCKImIjcB1cCnBnvfGPMo8ChAdXW1OaEP67WJoEWnl1DKUerq6sjLy6OqqgoRSXY4cWGMobW1lbq6OiZOnBj1dvGsGqoHxg14XRlZdhgR+TTwbeAqY4w/jvFYByac67IfVapVQ0o5gs/no7i4OGWTAICIUFxcPOyrnngmgpXAFBGZKCIZwHXA0oEriMg84BFsEmiOYyyH9DTZqqFuHx63i/zMk+KiSCmVAKmcBA44nmOMWyIwxgSBrwGvApuA54wxG0TkhyJyVWS1HwO5wPMislZElg6xu9gI9duRxTlldgxBvscR/zCUUupo4jqOwBizzBgz1Rgz2RjzvyPLvmeMWRr5+9PGmNHGmLmRx1VH3+MJ6t0PmAGjirVaSCmVGB0dHfz85z8f9naXX345HR0dcYjoEGeNLD44qrhMB5MppRJqqEQQDAaPut2yZcsoLCyMV1jASdJrKGF6Bk4v0c6CU0qSG49SKil+8McNbGzoiuk+Z5bn8/3Pnjrk+/feey87duxg7ty5pKenk5mZSVFREZs3b2br1q1cc8011NbW4vP5uOuuu1i8eDEAVVVV1NTU0NPTw6JFi1iwYAHvvfceFRUVvPTSS2RlZZ1w7M66IoiMKvZ7iun2BSnL16ohpVRi3H///UyePJm1a9fy4x//mNWrV/Ozn/2MrVu3ArBkyRJWrVpFTU0NDz74IK2trZ/Yx7Zt27jzzjvZsGEDhYWF/O53v4tJbI68ImgOFwBQqlVDSjnS0X65J8r8+fMP6+v/4IMP8uKLLwJQW1vLtm3bKC4uPmybiRMnMnfuXADOOOMMdu/eHZNYnJcIMvJo8tkLIW0jUEolS05OzsG/33rrLV577TVWrFhBdnY2F1544aBjATyeQ2VWWloafX19MYnFWVVDvXqLSqVUcuTl5dHd3T3oe52dnRQVFZGdnc3mzZt5//33Exqb864Ickezu7UXgIqiE29kUUqpaBQXF3Peeecxa9YssrKyGD169MH3Fi5cyC9+8QtmzJjBtGnTOPvssxMam8MSQROUTmdDfRfjRmVRkBX9pExKKXWifvvb3w663OPx8Morg865ebAdoKSkhPXr1x9c/q//+q8xi8tZVUORK4L1DZ3MKi9IdjRKKXVScE4iCPrB14HPU8yeVi+zKjQRKKUUOCkR9LYAUB/MB+DU8vxkRqOUUicN5ySCyGCyHV7bQHyqVg0ppRTgqERgB5Nt6MpiTH6mDiZTSqkIxyWClfvTmVWh1UJKKXWAcxKBdz8Aq1vTmKnVQkqpBDveaagBfvrTn+L1emMc0SHOSQQL/oU1N63HZzKYpQ3FSqkEO5kTgXMGlImwriUEoF1HlXK6V+6FxnWx3eeY2bDo/iHfHjgN9aWXXkpZWRnPPfccfr+fa6+9lh/84Af09vbyhS98gbq6OkKhEN/97ndpamqioaGBiy66iJKSEt58883Yxo2TEgGwvr6TUTkZjC3QOYaUUol1//33s379etauXcvy5ct54YUX+PDDDzHGcNVVV/HOO+/Q0tJCeXk5L7/8MmDnICooKOCBBx7gzTffpKQkPvdQcVgi6OLU8ny9T7FSTneUX+6JsHz5cpYvX868efMA6OnpYdu2bZx//vncfffdfPOb3+TKK6/k/PPPT0g8jkkE/mCIrU3d3D5tUrJDUUo5nDGGb33rW9xxxx2feG/16tUsW7aM73znO1xyySV873vfi3s8jmks3tbUQzBsdI4hpVRSDJyG+rLLLmPJkiX09PQAUF9fT3NzMw0NDWRnZ3PTTTdxzz33sHr16k9sGw+OuSJYX98JoGMIlFJJMXAa6kWLFnHDDTdwzjnnAJCbm8uTTz7J9u3bueeee3C5XKSnp/Pwww8DsHjxYhYuXEh5eXlcGovFGBPzncZTdXW1qampGfZ2yzc08vyqOh656QxcLm0jUMppNm3axIwZM5IdRkIMdqwissoYUz3Y+o65IvjMqWP4zKljkh2GUkqddBzTRqCUUmpwmgiUUo4x0qrCj8fxHKMmAqWUI2RmZtLa2prSycAYQ2trK5mZwxs065g2AqWUs1VWVlJXV0dLS0uyQ4mrzMxMKisrh7WNJgKllCOkp6czceLEZIdxUtKqIaWUcjhNBEop5XCaCJRSyuFG3MhiEWkB9hzn5iXA/hiGM1I48bideMzgzON24jHD8I97gjGmdLA3RlwiOBEiUjPUEOtU5sTjduIxgzOP24nHDLE9bq0aUkoph9NEoJRSDue0RPBosgNIEicetxOPGZx53E48ZojhcTuqjUAppdQnOe2KQCml1BE0ESillMM5JhGIyEIR2SIi20Xk3mTHEw8iMk5E3hSRjSKyQUTuiiwfJSJ/EZFtkeeiZMcaayKSJiJrRORPkdcTReSDyPl+VkQykh1jrIlIoYi8ICKbRWSTiJzjkHP9z5F/3+tF5GkRyUy18y0iS0SkWUTWD1g26LkV68HIsX8sIqcP9/MckQhEJA14CFgEzASuF5GZyY0qLoLA3caYmcDZwJ2R47wXeN0YMwV4PfI61dwFbBrw+kfA/zPGnAK0A7clJar4+hnwZ2PMdOA07PGn9LkWkQrg60C1MWYWkAZcR+qd78eAhUcsG+rcLgKmRB6LgYeH+2GOSATAfGC7MWanMSYAPANcneSYYs4Ys88Yszrydze2YKjAHuvjkdUeB65JToTxISKVwBXALyOvBbgYeCGySioecwFwAfArAGNMwBjTQYqf6wg3kCUibiAb2EeKnW9jzDtA2xGLhzq3VwNPGOt9oFBExg7n85ySCCqA2gGv6yLLUpaIVAHzgA+A0caYfZG3GoHRSQorXn4KfAMIR14XAx3GmGDkdSqe74lAC/DrSJXYL0UkhxQ/18aYeuAnwF5sAugEVpH65xuGPrcnXL45JRE4iojkAr8D/skY0zXwPWP7C6dMn2ERuRJoNsasSnYsCeYGTgceNsbMA3o5ohoo1c41QKRe/GpsIiwHcvhkFUrKi/W5dUoiqAfGDXhdGVmWckQkHZsEnjLG/D6yuOnApWLkuTlZ8cXBecBVIrIbW+V3MbbuvDBSdQCpeb7rgDpjzAeR1y9gE0Mqn2uATwO7jDEtxph+4PfYfwOpfr5h6HN7wuWbUxLBSmBKpGdBBrZxaWmSY4q5SN34r4BNxpgHBry1FLgl8vctwEuJji1ejDHfMsZUGmOqsOf1DWPMjcCbwOciq6XUMQMYYxqBWhGZFll0CbCRFD7XEXuBs0UkO/Lv/cBxp/T5jhjq3C4FvhTpPXQ20DmgCik6xhhHPIDLga3ADuDbyY4nTse4AHu5+DGwNvK4HFtn/jqwDXgNGJXsWON0/BcCf4r8PQn4ENgOPA94kh1fHI53LlATOd9/AIqccK6BHwCbgfXAbwBPqp1v4GlsG0g/9urvtqHOLSDYXpE7gHXYHlXD+jydYkIppRzOKVVDSimlhqCJQCmlHE4TgVJKOZwmAqWUcjhNBEop5XCaCJSKMxG58MCsqEqdjDQRKKWUw2kiUCpCRG4SkQ9FZK2IPBK5x0GPiPy/yPz3r4tIaWTduSLyfmT+9xcHzA1/ioi8JiIfichqEZkc2X3ugHsHPBUZFYuI3B+5f8THIvKTJB26cjhNBEoBIjID+CJwnjFmLhACbsROalZjjDkVeBv4fmSTJ4BvGmPmYEdzHlj+FPCQMeY04Fzs6FCwM8H+E/Z+GJOA80SkGLgWODWyn/8V36NUanCaCJSyLgHOAFaKyNrI60nYqa2fjazzJLAgci+AQmPM25HljwMXiEgeUGGMeRHAGOMzxngj63xojKkzxoSxU39UYadQ9gG/EpG/Aw6sq1RCaSJQyhLgcWPM3MhjmjHmvkHWO945WfwD/g4BbmPnz5+PnTn0SuDPx7lvpU6IJgKlrNeBz4lIGRy8P+wE7P+RA7Na3gC8a4zpBNpF5PzI8puBt429K1ydiFwT2YdHRLKH+sDIfSMKjDHLgH/G3m5SqYRzH3sVpVKfMWajiHwHWC4iLuysj3dib/gyP/JeM7YdAew0wL+IFPQ7gS9Hlt8MPCIiP4zs4/NH+dg84CURycRekfxLjA9Lqajo7KNKHYWI9BhjcpMdh1LxpFVDSinlcHpFoJRSDqdXBEop5XCaCJRSyuE0ESillMNpIlBKKYfTRKCUUg73/wGo5istYB1I9QAAAABJRU5ErkJggg==\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "source": [
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "plt.ylabel('accuracy'); plt.xlabel('epochs')\n",
        "plt.legend(['train', 'test'], loc='lower right')\n",
        "plt.show()"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "name": "German_traffic_sign_benchmark.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python [conda env:dl4dummies]",
      "language": "python",
      "name": "conda-env-dl4dummies-py"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.6"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}