Train_CNN_Digital-Readout-Small-v2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# CNN Training\n",
    "\n",
    "Target of this code is to train a CNN network to classify images of a digital readout to the digits 0 to 9. Additionally a category \"NaN\" is introduced, to mark images that are not amibiguous.\n",
    "\n",
    "### Preparing the training\n",
    "* First all libraries are loaded\n",
    "    * It is assumed, that they are installed during the Python setup\n",
    "* matplotlib is set to print the output inline in the jupyter notebook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "########### Basic Parameters for Running: ################################\n",
    "    \n",
    "TFliteNamingAndVersion = \"dig1320s2\"   # Used for tflite Filename\n",
    "Training_Percentage = 0.0              # 0.0 = Use all Images for Training\n",
    "Epoch_Anz = 500\n",
    "\n",
    "##########################################################################\n",
    "\n",
    "\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "import glob\n",
    "import os\n",
    "import numpy as np\n",
    "from sklearn.utils import shuffle\n",
    "from tensorflow.python import keras\n",
    "from tensorflow.python.keras import Sequential\n",
    "from tensorflow.python.keras.layers import Dense, InputLayer, Conv2D, MaxPool2D, Flatten, BatchNormalization\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "from sklearn.model_selection import train_test_split\n",
    "from tensorflow.keras.callbacks import History \n",
    "from tensorflow.keras.utils import to_categorical\n",
    "from PIL import Image \n",
    "from pathlib import Path\n",
    "\n",
    "loss_ges = np.array([])\n",
    "val_loss_ges = np.array([])\n",
    "\n",
    "%matplotlib inline\n",
    "np.set_printoptions(precision=4)\n",
    "np.set_printoptions(suppress=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load training data\n",
    "* The data is expected in the \"Input_dir\"\n",
    "* Inside subdirectories are expected from -1, 0, 1, ... 9 in which the pictures are sorted according to their values (=category)\n",
    "* Picture size must be 20x32 with 3 color channels (RGB)\n",
    "* The filename can be arbitrary\n",
    "\n",
    "* The images are stored in the x_data[]\n",
    "* The expected category for each image in the corresponding y_data[]\n",
    "\n",
    "* The last step is a shuffle (from sklearn.utils) and split the data into training and validation data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1582, 32, 20, 3)\n",
      "(1582, 11)\n"
     ]
    }
   ],
   "source": [
    "Input_dir='ziffer_sortiert_resize'\n",
    "\n",
    "files = glob.glob(Input_dir + '/*.jpg')\n",
    "x_data = []\n",
    "y_data = []\n",
    "\n",
    "for aktfile in files:\n",
    "    base = os.path.basename(aktfile)\n",
    "    target = base[0:1]\n",
    "    if target == \"N\":\n",
    "        category = 10                # NaN does not work --> convert to 10\n",
    "    else:\n",
    "        category = int(target)\n",
    "    test_image = Image.open(aktfile)\n",
    "    test_image = np.array(test_image, dtype=\"float32\")\n",
    "    x_data.append(test_image)\n",
    "    y_data.append(np.array([category]))\n",
    "\n",
    "x_data = np.array(x_data)\n",
    "y_data = np.array(y_data)\n",
    "y_data = to_categorical(y_data, 11)\n",
    "print(x_data.shape)\n",
    "print(y_data.shape)\n",
    "\n",
    "x_data, y_data = shuffle(x_data, y_data)\n",
    "\n",
    "if (Training_Percentage > 0):\n",
    "    X_train, X_test, y_train, y_test = train_test_split(x_data, y_data, test_size=Training_Percentage)\n",
    "else:\n",
    "    X_train = x_data\n",
    "    y_train = y_data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Define the model\n",
    "\n",
    "The layout of the network ist a typcial CNN network with alternating **Conv2D** and **MaxPool2D** layers. Finished after **flattening** with additional **Dense** layer.\n",
    "\n",
    "#### Important\n",
    "* Shape of the input layer: (32, 20, 3)\n",
    "* Number of output layers: 11\n",
    "* As loss function \"categorical_crossentropy\" is choosen, as it is a categories task"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "batch_normalization (BatchNo (None, 32, 20, 3)         12        \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 32, 20, 32)        896       \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 16, 10, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 16, 10, 32)        9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 8, 5, 32)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 8, 5, 32)          9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 4, 2, 32)          0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 256)               65792     \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 11)                2827      \n",
      "=================================================================\n",
      "Total params: 88,023\n",
      "Trainable params: 88,017\n",
      "Non-trainable params: 6\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "model.add(BatchNormalization(input_shape=(32,20,3)))\n",
    "model.add(Conv2D(32, (3, 3), padding='same', activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Conv2D(32, (3, 3), padding='same', activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Conv2D(32, (3, 3), padding='same', activation=\"relu\"))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(256,activation=\"relu\"))\n",
    "model.add(Dense(11, activation = \"softmax\"))\n",
    "model.summary()\n",
    "\n",
    "model.compile(loss=keras.losses.categorical_crossentropy, optimizer=tf.keras.optimizers.Adadelta(learning_rate=1.0, rho=0.95), metrics = [\"accuracy\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training\n",
    "The input pictures are randomly scattered for brightness, pixel shift variations and rotation angle. This is implemented with a ImageDataGenerator.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 2.2060 - accuracy: 0.3084\n",
      "Epoch 2/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 1.4547 - accuracy: 0.5454\n",
      "Epoch 3/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 1.0399 - accuracy: 0.6740\n",
      "Epoch 4/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.7859 - accuracy: 0.7539\n",
      "Epoch 5/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.6694 - accuracy: 0.8060 0s - loss:\n",
      "Epoch 6/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.6055 - accuracy: 0.8158\n",
      "Epoch 7/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.5277 - accuracy: 0.8366\n",
      "Epoch 8/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.4293 - accuracy: 0.8640\n",
      "Epoch 9/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.4308 - accuracy: 0.8623\n",
      "Epoch 10/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.4386 - accuracy: 0.8670\n",
      "Epoch 11/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.4072 - accuracy: 0.8794\n",
      "Epoch 12/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.4036 - accuracy: 0.8792 0s - loss: 0.4039 - accuracy: 0.\n",
      "Epoch 13/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.3317 - accuracy: 0.8973 0s - loss: 0.335\n",
      "Epoch 14/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.3083 - accuracy: 0.9198 0s - loss: 0.3087  - ETA: 0s - loss: 0.3083 - accu\n",
      "Epoch 15/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.2689 - accuracy: 0.9162\n",
      "Epoch 16/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.2928 - accuracy: 0.9159\n",
      "Epoch 17/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.2611 - accuracy: 0.9252\n",
      "Epoch 18/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.2492 - accuracy: 0.9173: 0s - loss:\n",
      "Epoch 19/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.2592 - accuracy: 0.9260\n",
      "Epoch 20/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.2654 - accuracy: 0.9290\n",
      "Epoch 21/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1897 - accuracy: 0.9427\n",
      "Epoch 22/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1640 - accuracy: 0.9444\n",
      "Epoch 23/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.2329 - accuracy: 0.9258\n",
      "Epoch 24/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1902 - accuracy: 0.9349\n",
      "Epoch 25/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1462 - accuracy: 0.9554\n",
      "Epoch 26/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1711 - accuracy: 0.9541 0s - loss: 0\n",
      "Epoch 27/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1957 - accuracy: 0.9388\n",
      "Epoch 28/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1804 - accuracy: 0.9491\n",
      "Epoch 29/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1628 - accuracy: 0.9532\n",
      "Epoch 30/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1518 - accuracy: 0.9563\n",
      "Epoch 31/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1502 - accuracy: 0.9686 0s - loss: 0.1495 - ac\n",
      "Epoch 32/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1252 - accuracy: 0.9607\n",
      "Epoch 33/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1443 - accuracy: 0.9555\n",
      "Epoch 34/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1786 - accuracy: 0.9547\n",
      "Epoch 35/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1363 - accuracy: 0.9545\n",
      "Epoch 36/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1459 - accuracy: 0.9508\n",
      "Epoch 37/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1057 - accuracy: 0.9565\n",
      "Epoch 38/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0745 - accuracy: 0.9776 0s - loss: 0.0726 - accu\n",
      "Epoch 39/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1258 - accuracy: 0.9657\n",
      "Epoch 40/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1213 - accuracy: 0.9711: 1s - loss: - ETA: 0s - l\n",
      "Epoch 41/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1328 - accuracy: 0.9674\n",
      "Epoch 42/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1824 - accuracy: 0.9524\n",
      "Epoch 43/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.1184 - accuracy: 0.9634\n",
      "Epoch 44/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1235 - accuracy: 0.9641\n",
      "Epoch 45/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1149 - accuracy: 0.9663\n",
      "Epoch 46/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1370 - accuracy: 0.9536\n",
      "Epoch 47/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1254 - accuracy: 0.9646\n",
      "Epoch 48/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1030 - accuracy: 0.9724\n",
      "Epoch 49/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.1074 - accuracy: 0.9673\n",
      "Epoch 50/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0816 - accuracy: 0.9739\n",
      "Epoch 51/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0949 - accuracy: 0.9610\n",
      "Epoch 52/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1168 - accuracy: 0.9632\n",
      "Epoch 53/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1053 - accuracy: 0.9659\n",
      "Epoch 54/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0777 - accuracy: 0.9748\n",
      "Epoch 55/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1150 - accuracy: 0.9670\n",
      "Epoch 56/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1211 - accuracy: 0.9740\n",
      "Epoch 57/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1078 - accuracy: 0.9616\n",
      "Epoch 58/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0948 - accuracy: 0.9692\n",
      "Epoch 59/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0806 - accuracy: 0.9764\n",
      "Epoch 60/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0906 - accuracy: 0.9714\n",
      "Epoch 61/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0707 - accuracy: 0.9833\n",
      "Epoch 62/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1105 - accuracy: 0.9742\n",
      "Epoch 63/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1241 - accuracy: 0.9713\n",
      "Epoch 64/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0727 - accuracy: 0.9817\n",
      "Epoch 65/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1024 - accuracy: 0.9689\n",
      "Epoch 66/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0985 - accuracy: 0.9687\n",
      "Epoch 67/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1016 - accuracy: 0.9786\n",
      "Epoch 68/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0748 - accuracy: 0.9777: 2s -\n",
      "Epoch 69/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0737 - accuracy: 0.9715\n",
      "Epoch 70/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0627 - accuracy: 0.9847\n",
      "Epoch 71/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0589 - accuracy: 0.9793\n",
      "Epoch 72/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0890 - accuracy: 0.9758 0s\n",
      "Epoch 73/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0765 - accuracy: 0.9763\n",
      "Epoch 74/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0617 - accuracy: 0.9745\n",
      "Epoch 75/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0595 - accuracy: 0.9782\n",
      "Epoch 76/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0749 - accuracy: 0.9837\n",
      "Epoch 77/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0932 - accuracy: 0.9750\n",
      "Epoch 78/500\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0666 - accuracy: 0.9784\n",
      "Epoch 79/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0739 - accuracy: 0.9809\n",
      "Epoch 80/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0695 - accuracy: 0.9683\n",
      "Epoch 81/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0703 - accuracy: 0.9809\n",
      "Epoch 82/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0817 - accuracy: 0.9750\n",
      "Epoch 83/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0735 - accuracy: 0.9819\n",
      "Epoch 84/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0698 - accuracy: 0.9821\n",
      "Epoch 85/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0735 - accuracy: 0.9772\n",
      "Epoch 86/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0660 - accuracy: 0.9786\n",
      "Epoch 87/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0615 - accuracy: 0.9846\n",
      "Epoch 88/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0869 - accuracy: 0.9770\n",
      "Epoch 89/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0797 - accuracy: 0.9802\n",
      "Epoch 90/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0481 - accuracy: 0.9843\n",
      "Epoch 91/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0357 - accuracy: 0.9891\n",
      "Epoch 92/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0682 - accuracy: 0.9804\n",
      "Epoch 93/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0827 - accuracy: 0.9792\n",
      "Epoch 94/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0729 - accuracy: 0.9805\n",
      "Epoch 95/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0599 - accuracy: 0.9826\n",
      "Epoch 96/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0856 - accuracy: 0.9761\n",
      "Epoch 97/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0825 - accuracy: 0.9751 0s - l\n",
      "Epoch 98/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0565 - accuracy: 0.9752\n",
      "Epoch 99/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0730 - accuracy: 0.9770\n",
      "Epoch 100/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0786 - accuracy: 0.9784\n",
      "Epoch 101/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0774 - accuracy: 0.9685\n",
      "Epoch 102/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0573 - accuracy: 0.9822\n",
      "Epoch 103/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0480 - accuracy: 0.9829\n",
      "Epoch 104/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0803 - accuracy: 0.9747\n",
      "Epoch 105/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0413 - accuracy: 0.9818\n",
      "Epoch 106/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0767 - accuracy: 0.9796\n",
      "Epoch 107/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0557 - accuracy: 0.9804\n",
      "Epoch 108/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0380 - accuracy: 0.9853\n",
      "Epoch 109/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0518 - accuracy: 0.9911\n",
      "Epoch 110/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0588 - accuracy: 0.9798\n",
      "Epoch 111/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0645 - accuracy: 0.9822\n",
      "Epoch 112/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0386 - accuracy: 0.9897\n",
      "Epoch 113/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0804 - accuracy: 0.9813\n",
      "Epoch 114/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0559 - accuracy: 0.9852\n",
      "Epoch 115/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0699 - accuracy: 0.9817\n",
      "Epoch 116/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0728 - accuracy: 0.9867\n",
      "Epoch 117/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0581 - accuracy: 0.9778\n",
      "Epoch 118/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0524 - accuracy: 0.9840\n",
      "Epoch 119/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0650 - accuracy: 0.9779\n",
      "Epoch 120/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0616 - accuracy: 0.9832\n",
      "Epoch 121/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0352 - accuracy: 0.9908\n",
      "Epoch 122/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0640 - accuracy: 0.9819\n",
      "Epoch 123/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.1038 - accuracy: 0.9754\n",
      "Epoch 124/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0369 - accuracy: 0.9925\n",
      "Epoch 125/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0507 - accuracy: 0.9891\n",
      "Epoch 126/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0322 - accuracy: 0.9899\n",
      "Epoch 127/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0219 - accuracy: 0.9946\n",
      "Epoch 128/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0419 - accuracy: 0.9873\n",
      "Epoch 129/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0247 - accuracy: 0.9927\n",
      "Epoch 130/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0387 - accuracy: 0.9884\n",
      "Epoch 131/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0540 - accuracy: 0.9812\n",
      "Epoch 132/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0413 - accuracy: 0.9872\n",
      "Epoch 133/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0531 - accuracy: 0.9879\n",
      "Epoch 134/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0292 - accuracy: 0.9885\n",
      "Epoch 135/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0385 - accuracy: 0.9887\n",
      "Epoch 136/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0527 - accuracy: 0.9881\n",
      "Epoch 137/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.1203 - accuracy: 0.9823\n",
      "Epoch 138/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0987 - accuracy: 0.9847\n",
      "Epoch 139/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0549 - accuracy: 0.9753\n",
      "Epoch 140/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0515 - accuracy: 0.9885\n",
      "Epoch 141/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0670 - accuracy: 0.9807 0s - los\n",
      "Epoch 142/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0533 - accuracy: 0.9844\n",
      "Epoch 143/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0381 - accuracy: 0.9903\n",
      "Epoch 144/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0481 - accuracy: 0.9816\n",
      "Epoch 145/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0617 - accuracy: 0.9829\n",
      "Epoch 146/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0455 - accuracy: 0.9844 0s - l\n",
      "Epoch 147/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0378 - accuracy: 0.9893\n",
      "Epoch 148/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0297 - accuracy: 0.9941 \n",
      "Epoch 149/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0220 - accuracy: 0.9944 2s - loss: 0.0194 - accura - ETA: 2s - loss: 0.0192 - accu - ETA: 1s - loss: 0.0\n",
      "Epoch 150/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0422 - accuracy: 0.9848\n",
      "Epoch 151/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0365 - accuracy: 0.9924\n",
      "Epoch 152/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0281 - accuracy: 0.9872 0s - loss: 0.0277 - accuracy: 0.\n",
      "Epoch 153/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0203 - accuracy: 0.9951\n",
      "Epoch 154/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0246 - accuracy: 0.9903\n",
      "Epoch 155/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0463 - accuracy: 0.9838\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 156/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0302 - accuracy: 0.9910\n",
      "Epoch 157/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0487 - accuracy: 0.9825 1s\n",
      "Epoch 158/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0564 - accuracy: 0.9832\n",
      "Epoch 159/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0539 - accuracy: 0.9880 0s - loss: 0.0544 \n",
      "Epoch 160/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0441 - accuracy: 0.9881\n",
      "Epoch 161/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0377 - accuracy: 0.9909\n",
      "Epoch 162/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0294 - accuracy: 0.9927\n",
      "Epoch 163/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0263 - accuracy: 0.9861 0s - los\n",
      "Epoch 164/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0577 - accuracy: 0.9865\n",
      "Epoch 165/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0358 - accuracy: 0.9876\n",
      "Epoch 166/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0510 - accuracy: 0.9853 0s - loss: 0.0511 - \n",
      "Epoch 167/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0289 - accuracy: 0.9879\n",
      "Epoch 168/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0323 - accuracy: 0.9891\n",
      "Epoch 169/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0451 - accuracy: 0.9866 0s - loss: 0.0455 - accu\n",
      "Epoch 170/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0420 - accuracy: 0.9874: 0s - loss: 0.042\n",
      "Epoch 171/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0184 - accuracy: 0.9951\n",
      "Epoch 172/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0439 - accuracy: 0.9895\n",
      "Epoch 173/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0630 - accuracy: 0.9839: 0s - l\n",
      "Epoch 174/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0192 - accuracy: 0.9934\n",
      "Epoch 175/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0443 - accuracy: 0.9865\n",
      "Epoch 176/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0262 - accuracy: 0.9915\n",
      "Epoch 177/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0177 - accuracy: 0.9975: 0s - loss: 0.0173 - accuracy: 0.99\n",
      "Epoch 178/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0585 - accuracy: 0.9895 0s - loss: 0.0590 - accuracy: 0.\n",
      "Epoch 179/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0745 - accuracy: 0.9831\n",
      "Epoch 180/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0277 - accuracy: 0.9886\n",
      "Epoch 181/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0518 - accuracy: 0.9944\n",
      "Epoch 182/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0224 - accuracy: 0.9945\n",
      "Epoch 183/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0517 - accuracy: 0.9878 0s - loss: 0.0522 \n",
      "Epoch 184/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0252 - accuracy: 0.9898\n",
      "Epoch 185/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0573 - accuracy: 0.9800\n",
      "Epoch 186/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0250 - accuracy: 0.9912\n",
      "Epoch 187/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0308 - accuracy: 0.9920\n",
      "Epoch 188/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0156 - accuracy: 0.9909\n",
      "Epoch 189/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0256 - accuracy: 0.9967\n",
      "Epoch 190/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0331 - accuracy: 0.9892 1s - loss: 0.0340  - ETA\n",
      "Epoch 191/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0146 - accuracy: 0.9967\n",
      "Epoch 192/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0329 - accuracy: 0.9899\n",
      "Epoch 193/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0382 - accuracy: 0.9913\n",
      "Epoch 194/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0295 - accuracy: 0.9896\n",
      "Epoch 195/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0310 - accuracy: 0.9893\n",
      "Epoch 196/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0387 - accuracy: 0.9886\n",
      "Epoch 197/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0200 - accuracy: 0.9909\n",
      "Epoch 198/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0155 - accuracy: 0.9931\n",
      "Epoch 199/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0366 - accuracy: 0.9906\n",
      "Epoch 200/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0314 - accuracy: 0.9922\n",
      "Epoch 201/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0229 - accuracy: 0.9895\n",
      "Epoch 202/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0425 - accuracy: 0.9894\n",
      "Epoch 203/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0354 - accuracy: 0.9926\n",
      "Epoch 204/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0340 - accuracy: 0.9889\n",
      "Epoch 205/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0504 - accuracy: 0.9865 0s - loss: 0.0509 - accuracy: \n",
      "Epoch 206/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0365 - accuracy: 0.9925: 0s - loss: 0.0363 - accuracy: 0.\n",
      "Epoch 207/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0422 - accuracy: 0.9871: 1s - loss: 0.0434 - accuracy:  - ETA: 1s\n",
      "Epoch 208/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0224 - accuracy: 0.9926\n",
      "Epoch 209/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0401 - accuracy: 0.9873\n",
      "Epoch 210/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0153 - accuracy: 0.9954: 0s - loss: 0.0139 - \n",
      "Epoch 211/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0259 - accuracy: 0.9916\n",
      "Epoch 212/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0515 - accuracy: 0.9861\n",
      "Epoch 213/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0665 - accuracy: 0.9844: \n",
      "Epoch 214/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0465 - accuracy: 0.9915: 2s - loss: 0.0320 - accu - ETA: 2s - loss: 0.0324 - ac - ETA: 2s - los - ETA: 1s - l - ETA: 0s - loss: 0.0460 - accuracy: 0. - ETA: 0s - loss: 0.0464 - accuracy: 0.99\n",
      "Epoch 215/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0236 - accuracy: 0.9904: 0s - loss: 0\n",
      "Epoch 216/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0190 - accuracy: 0.9941\n",
      "Epoch 241/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0344 - accuracy: 0.9905\n",
      "Epoch 242/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0184 - accuracy: 0.9939\n",
      "Epoch 243/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0104 - accuracy: 0.9958: 0s - loss: 0.0097 - ac\n",
      "Epoch 244/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0532 - accuracy: 0.9916\n",
      "Epoch 245/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0216 - accuracy: 0.9933\n",
      "Epoch 246/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0523 - accuracy: 0.9892\n",
      "Epoch 247/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0210 - accuracy: 0.9949\n",
      "Epoch 248/500\n",
      "396/396 [==============================] - 5s 13ms/step - loss: 0.0384 - accuracy: 0.9926\n",
      "Epoch 249/500\n",
      "396/396 [==============================] - 6s 13ms/step - loss: 0.0365 - accuracy: 0.9894\n",
      "Epoch 250/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0262 - accuracy: 0.9911\n",
      "Epoch 251/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0216 - accuracy: 0.9933\n",
      "Epoch 252/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0216 - accuracy: 0.9928 0s - loss:\n",
      "Epoch 253/500\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0198 - accuracy: 0.9940\n",
      "Epoch 254/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0318 - accuracy: 0.9923\n",
      "Epoch 255/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0722 - accuracy: 0.9862\n",
      "Epoch 256/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0173 - accuracy: 0.9942\n",
      "Epoch 257/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0120 - accuracy: 0.9964 1s - loss: 0.0084 - accuracy:  - ETA: 0s - l\n",
      "Epoch 258/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0273 - accuracy: 0.9873\n",
      "Epoch 259/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0158 - accuracy: 0.9933 0s - loss: 0.0141 \n",
      "Epoch 260/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0222 - accuracy: 0.9939\n",
      "Epoch 261/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0391 - accuracy: 0.9908\n",
      "Epoch 262/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0145 - accuracy: 0.9940 2s - loss: 0.0117 - accuracy: 0.99 - ETA: 2s -\n",
      "Epoch 263/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0337 - accuracy: 0.9927\n",
      "Epoch 264/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0224 - accuracy: 0.9903\n",
      "Epoch 265/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0225 - accuracy: 0.9953\n",
      "Epoch 266/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0248 - accuracy: 0.9893 1s - los - ETA: 0s - los\n",
      "Epoch 267/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0515 - accuracy: 0.9875\n",
      "Epoch 268/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0224 - accuracy: 0.9932\n",
      "Epoch 269/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0097 - accuracy: 0.9961 0s - loss: 0.0095 - accuracy: 0.\n",
      "Epoch 270/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0244 - accuracy: 0.9918\n",
      "Epoch 271/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0312 - accuracy: 0.9899 1s - loss: 0.0315 - ac - ETA: 0s - loss: 0.0317 \n",
      "Epoch 272/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0256 - accuracy: 0.9915\n",
      "Epoch 273/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0228 - accuracy: 0.9899 0s - loss: 0\n",
      "Epoch 274/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0180 - accuracy: 0.9947 0s - loss: 0.0\n",
      "Epoch 275/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0189 - accuracy: 0.9977 0s - loss: 0.0186 - accura\n",
      "Epoch 276/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0078 - accuracy: 0.9975\n",
      "Epoch 297/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0421 - accuracy: 0.9917\n",
      "Epoch 298/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0298 - accuracy: 0.9933 0s - loss: 0.0283 - ac\n",
      "Epoch 299/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0167 - accuracy: 0.9940\n",
      "Epoch 300/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0177 - accuracy: 0.9958 2s - loss: 0.0 - ETA: 0s - loss: 0\n",
      "Epoch 301/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0160 - accuracy: 0.9955\n",
      "Epoch 302/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0165 - accuracy: 0.9961: 1s - loss: - ETA: 0s - loss: 0.0144 - accuracy: 0. - ETA: 0s - loss: 0.0150 - ac\n",
      "Epoch 303/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0081 - accuracy: 0.9966: 2s - loss: 0\n",
      "Epoch 304/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0471 - accuracy: 0.9846\n",
      "Epoch 305/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0386 - accuracy: 0.9949\n",
      "Epoch 306/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0328 - accuracy: 0.9918\n",
      "Epoch 307/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0208 - accuracy: 0.9935\n",
      "Epoch 308/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0089 - accuracy: 0.9981\n",
      "Epoch 309/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0169 - accuracy: 0.9939\n",
      "Epoch 310/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0227 - accuracy: 0.9932\n",
      "Epoch 311/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0217 - accuracy: 0.9967\n",
      "Epoch 312/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0206 - accuracy: 0.9929 0s - loss:\n",
      "Epoch 313/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0222 - accuracy: 0.9943\n",
      "Epoch 314/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0184 - accuracy: 0.9946\n",
      "Epoch 315/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0308 - accuracy: 0.9930\n",
      "Epoch 316/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0187 - accuracy: 0.9924\n",
      "Epoch 317/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0363 - accuracy: 0.9870  - ETA: 1s - loss: 0 - E\n",
      "Epoch 318/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0058 - accuracy: 0.9973\n",
      "Epoch 319/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0187 - accuracy: 0.9957\n",
      "Epoch 320/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0277 - accuracy: 0.9913\n",
      "Epoch 321/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0215 - accuracy: 0.9947\n",
      "Epoch 322/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0221 - accuracy: 0.9957\n",
      "Epoch 323/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0113 - accuracy: 0.9963\n",
      "Epoch 324/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0413 - accuracy: 0.9925\n",
      "Epoch 325/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0130 - accuracy: 0.9951\n",
      "Epoch 326/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0202 - accuracy: 0.9892 0s - loss:\n",
      "Epoch 327/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0103 - accuracy: 0.9959\n",
      "Epoch 328/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0347 - accuracy: 0.9920\n",
      "Epoch 329/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0045 - accuracy: 0.9983\n",
      "Epoch 330/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0491 - accuracy: 0.9869\n",
      "Epoch 331/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0517 - accuracy: 0.9889\n",
      "Epoch 332/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0174 - accuracy: 0.9940\n",
      "Epoch 333/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0362 - accuracy: 0.9908: 0s - loss: 0.0375 - ac\n",
      "Epoch 334/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0746 - accuracy: 0.9889 0s - loss: 0\n",
      "Epoch 335/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0141 - accuracy: 0.9942 0s - loss: 0\n",
      "Epoch 367/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0103 - accuracy: 0.9965\n",
      "Epoch 368/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0232 - accuracy: 0.9905 0s - loss: 0.0232 - accuracy: 0.99\n",
      "Epoch 369/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0322 - accuracy: 0.9925\n",
      "Epoch 370/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0036 - accuracy: 0.9995\n",
      "Epoch 371/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0169 - accuracy: 0.9930\n",
      "Epoch 372/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0178 - accuracy: 0.9932\n",
      "Epoch 373/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0239 - accuracy: 0.9955\n",
      "Epoch 374/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0161 - accuracy: 0.9968 1s - - ETA: 0s -\n",
      "Epoch 375/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0215 - accuracy: 0.9957\n",
      "Epoch 376/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0129 - accuracy: 0.9970\n",
      "Epoch 377/500\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0457 - accuracy: 0.9954\n",
      "Epoch 378/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0205 - accuracy: 0.9958: 0s - loss: 0.019\n",
      "Epoch 379/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0437 - accuracy: 0.9905\n",
      "Epoch 380/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0312 - accuracy: 0.9931\n",
      "Epoch 381/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0068 - accuracy: 0.9976 0s - loss: 0.0056 - \n",
      "Epoch 382/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0093 - accuracy: 0.9974\n",
      "Epoch 383/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0223 - accuracy: 0.9940\n",
      "Epoch 384/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0350 - accuracy: 0.9930\n",
      "Epoch 385/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0200 - accuracy: 0.9938A\n",
      "Epoch 386/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0189 - accuracy: 0.9964\n",
      "Epoch 387/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0210 - accuracy: 0.9929 0s - loss:\n",
      "Epoch 388/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0111 - accuracy: 0.9933 0s - loss:\n",
      "Epoch 389/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0230 - accuracy: 0.9962\n",
      "Epoch 390/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0094 - accuracy: 0.9960\n",
      "Epoch 391/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0167 - accuracy: 0.9949 0s - l\n",
      "Epoch 392/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0286 - accuracy: 0.9937\n",
      "Epoch 393/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0202 - accuracy: 0.9929\n",
      "Epoch 394/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0139 - accuracy: 0.9944\n",
      "Epoch 395/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0378 - accuracy: 0.9912\n",
      "Epoch 396/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0151 - accuracy: 0.9962\n",
      "Epoch 397/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0129 - accuracy: 0.9946 0s - loss: 0.012\n",
      "Epoch 398/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0435 - accuracy: 0.9929\n",
      "Epoch 399/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0213 - accuracy: 0.9959 0s - l - ETA: 0s - loss: 0.0214 - accuracy: 0.99\n",
      "Epoch 400/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0115 - accuracy: 0.9972 0s - loss: 0.008\n",
      "Epoch 401/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0052 - accuracy: 0.9987\n",
      "Epoch 402/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0188 - accuracy: 0.9983\n",
      "Epoch 403/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0111 - accuracy: 0.9965 2s - los\n",
      "Epoch 404/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0236 - accuracy: 0.9967\n",
      "Epoch 405/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0225 - accuracy: 0.9940 0s - los\n",
      "Epoch 406/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0058 - accuracy: 0.9967\n",
      "Epoch 407/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0207 - accuracy: 0.9946\n",
      "Epoch 408/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0080 - accuracy: 0.9966 0s -\n",
      "Epoch 409/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0258 - accuracy: 0.9911 0s - los\n",
      "Epoch 410/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0134 - accuracy: 0.9953\n",
      "Epoch 411/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0056 - accuracy: 0.9988 1s - loss: 0.0048  - ETA: 0s - loss: 0.005\n",
      "Epoch 412/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0207 - accuracy: 0.9969\n",
      "Epoch 413/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0283 - accuracy: 0.9936\n",
      "Epoch 414/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0083 - accuracy: 0.9982 0s - l\n",
      "Epoch 415/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0305 - accuracy: 0.9960\n",
      "Epoch 416/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0307 - accuracy: 0.9928\n",
      "Epoch 427/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0122 - accuracy: 0.9961\n",
      "Epoch 428/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0102 - accuracy: 0.9980\n",
      "Epoch 429/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0521 - accuracy: 0.9905\n",
      "Epoch 430/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0272 - accuracy: 0.9906\n",
      "Epoch 431/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0170 - accuracy: 0.9927\n",
      "Epoch 432/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0115 - accuracy: 0.9967\n",
      "Epoch 433/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0100 - accuracy: 0.9978 0s - loss: 0.008\n",
      "Epoch 434/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0054 - accuracy: 0.9976\n",
      "Epoch 435/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0279 - accuracy: 0.9955 0s - loss: 0.0276 - \n",
      "Epoch 436/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0145 - accuracy: 0.9956\n",
      "Epoch 437/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0204 - accuracy: 0.9959\n",
      "Epoch 438/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0106 - accuracy: 0.9958\n",
      "Epoch 439/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0273 - accuracy: 0.9964\n",
      "Epoch 440/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0216 - accuracy: 0.9936 0s - loss: 0\n",
      "Epoch 441/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0340 - accuracy: 0.9888 1s - l - ETA: 0s - loss: 0.0351 - \n",
      "Epoch 442/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0363 - accuracy: 0.9900 0s - loss:\n",
      "Epoch 443/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0077 - accuracy: 0.9988\n",
      "Epoch 444/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0274 - accuracy: 0.9953\n",
      "Epoch 445/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0110 - accuracy: 0.9984\n",
      "Epoch 446/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0099 - accuracy: 0.9979\n",
      "Epoch 447/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0203 - accuracy: 0.9941 1s\n",
      "Epoch 448/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0064 - accuracy: 0.9977\n",
      "Epoch 449/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0081 - accuracy: 0.9992 0s - loss: 0.0058 - accu - ETA: 0s - loss: 0.0070 - accuracy: 0.99 - ETA: 0s - loss: 0.0071 - accuracy:  - ETA: 0s - loss: 0.0075 - accura\n",
      "Epoch 450/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0016 - accuracy: 0.9998\n",
      "Epoch 451/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0404 - accuracy: 0.9936\n",
      "Epoch 452/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0267 - accuracy: 0.9931\n",
      "Epoch 453/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0187 - accuracy: 0.9968 0s - l\n",
      "Epoch 454/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0282 - accuracy: 0.9892 0s - loss: 0.0288 - accu\n",
      "Epoch 455/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0076 - accuracy: 0.9961\n",
      "Epoch 456/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0295 - accuracy: 0.9924ETA: 3s - los - ETA: 0s - loss: 0.0305 - accu\n",
      "Epoch 457/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0358 - accuracy: 0.9912\n",
      "Epoch 458/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0463 - accuracy: 0.9920\n",
      "Epoch 459/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0309 - accuracy: 0.9944\n",
      "Epoch 460/500\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0233 - accuracy: 0.9943 0s - los\n",
      "Epoch 461/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0087 - accuracy: 0.9959\n",
      "Epoch 462/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0163 - accuracy: 0.9931\n",
      "Epoch 463/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0356 - accuracy: 0.9920\n",
      "Epoch 464/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0138 - accuracy: 0.9944\n",
      "Epoch 465/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0213 - accuracy: 0.9926\n",
      "Epoch 466/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0205 - accuracy: 0.9976\n",
      "Epoch 467/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0068 - accuracy: 0.9989 1s - loss: 0.0 - ETA: 0s - loss: 0.005\n",
      "Epoch 468/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0215 - accuracy: 0.9953\n",
      "Epoch 469/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0232 - accuracy: 0.9933\n",
      "Epoch 470/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0262 - accuracy: 0.9902\n",
      "Epoch 471/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0197 - accuracy: 0.9948 2s - loss: -\n",
      "Epoch 472/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0198 - accuracy: 0.9955 1s - loss: 0.0127 - ac - ETA: 1s - loss: 0.0 - ETA: 0s - los\n",
      "Epoch 473/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0050 - accuracy: 0.9970 0s - loss: 0\n",
      "Epoch 474/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0376 - accuracy: 0.9891\n",
      "Epoch 475/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0211 - accuracy: 0.9956\n",
      "Epoch 476/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0211 - accuracy: 0.9917 0s - loss: 0.0215 \n",
      "Epoch 477/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0226 - accuracy: 0.9930\n",
      "Epoch 478/500\n",
      "396/396 [==============================] - 5s 12ms/step - loss: 0.0188 - accuracy: 0.9954\n",
      "Epoch 479/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0436 - accuracy: 0.9898 0s - loss: 0.0458 \n",
      "Epoch 480/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0484 - accuracy: 0.9900 0s - loss: 0.0518 \n",
      "Epoch 481/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0429 - accuracy: 0.9900\n",
      "Epoch 482/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0128 - accuracy: 0.9968 0s - loss: 0.0126 - accura\n",
      "Epoch 483/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0146 - accuracy: 0.9968\n",
      "Epoch 484/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0473 - accuracy: 0.9876\n",
      "Epoch 485/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0397 - accuracy: 0.9846\n",
      "Epoch 486/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0136 - accuracy: 0.9967\n",
      "Epoch 487/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0247 - accuracy: 0.9947\n",
      "Epoch 488/500\n",
      "396/396 [==============================] - 5s 11ms/step - loss: 0.0249 - accuracy: 0.9937\n",
      "Epoch 489/500\n",
      "396/396 [==============================] - 4s 11ms/step - loss: 0.0183 - accuracy: 0.9981\n",
      "Epoch 490/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0199 - accuracy: 0.9930 0s - l\n",
      "Epoch 491/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0043 - accuracy: 0.9991\n",
      "Epoch 492/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0239 - accuracy: 0.9941\n",
      "Epoch 493/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0067 - accuracy: 0.9984: 0s - loss: 0\n",
      "Epoch 494/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0255 - accuracy: 0.9962\n",
      "Epoch 495/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0086 - accuracy: 0.9963\n",
      "Epoch 496/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0092 - accuracy: 0.9957\n",
      "Epoch 497/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0118 - accuracy: 0.9953\n",
      "Epoch 498/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0183 - accuracy: 0.9971 1s - - ETA: 0s - loss: 0.0179 - accura\n",
      "Epoch 499/500\n",
      "396/396 [==============================] - 4s 9ms/step - loss: 0.0036 - accuracy: 0.9985\n",
      "Epoch 500/500\n",
      "396/396 [==============================] - 4s 10ms/step - loss: 0.0111 - accuracy: 0.9981\n"
     ]
    }
   ],
   "source": [
    "Batch_Size = 4\n",
    "Shift_Range = 1\n",
    "Brightness_Range = 0.2\n",
    "Rotation_Angle = 5\n",
    "ZoomRange = 0.3\n",
    "\n",
    "datagen = ImageDataGenerator(width_shift_range=[-Shift_Range,Shift_Range], \n",
    "                             height_shift_range=[-Shift_Range,Shift_Range],\n",
    "                             brightness_range=[1-Brightness_Range,1+Brightness_Range],\n",
    "                             zoom_range=[1-ZoomRange, 1+ZoomRange],\n",
    "                             rotation_range=Rotation_Angle)\n",
    "\n",
    "if (Training_Percentage > 0):\n",
    "    train_iterator = datagen.flow(x_data, y_data, batch_size=Batch_Size)\n",
    "    validation_iterator = datagen.flow(X_test, y_test, batch_size=Batch_Size)\n",
    "    history = model.fit(train_iterator, validation_data = validation_iterator, epochs = Epoch_Anz)\n",
    "else:\n",
    "    train_iterator = datagen.flow(x_data, y_data, batch_size=Batch_Size)\n",
    "    history = model.fit(train_iterator, epochs = Epoch_Anz)\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Learing result\n",
    " \n",
    "* Visualization of the training and validation results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "loss_ges = np.append(loss_ges, history.history['loss'])\n",
    "plt.semilogy(history.history['loss'])\n",
    "\n",
    "if (Training_Percentage > 0):\n",
    "    val_loss_ges = np.append(val_loss_ges, history.history['val_loss'])\n",
    "    plt.semilogy(history.history['val_loss'])\n",
    "\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train','eval'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Check the model by hand\n",
    "\n",
    "* The following code uses the trained model to check the deviation for each picture.\n",
    "* x-axis walks through each pixel, y-axis shows the result"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Check each image for expected and deviation\n",
    "* setting the switch \"only_deviation = true\" will only print the images for which the classification and the CNN-result deviates\n",
    "\n",
    "The output contains the following information:\n",
    "\n",
    "| Filename      | Expected Category           | Predicted Category        |\n",
    "|------------- |:-----------------------------:|--------------|\n",
    "| ziffer_sortiert_resize_NaN/5\\Ziffer_4_0034.jpg | 4  | -1 |\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\0_1_ht5_20210627-220703.jpg 0 1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3F0F4F0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\3_0_ht6_20210702-055135.jpg 3 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF430FC10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\3_1_ht4_20210627-124615.jpg 3 4\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF404C250>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\3_6_ht6_20210702-044832.jpg 3 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3D59880>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\5_3_dig4_20210721-144728.jpg 5 9\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAABQAAAAgCAIAAACdAM/hAAADqElEQVR4nF1VubLcNhCcCwCX75W9VUr074od+NtcKh1+uyQwl4NZ0S4jIBPM1d3TwD/+/EJECBwRmQkADhgRQAgACJyZDgkAiEhEaAEAmZmZUjEIr4OIhISInoGIkICIkFkB7k6RiPgKBoCIQExmbjKYOQkBIJGqMgBYRp3MjKn56wgRuTsijjH223vvPQmJCIgzE5IAwCHd3czcXek0szlnRAgzRwQz7/t+v//WewcSRAzAzIwIAAggd1dVVTXk4zhUNTOFmc2MiMYY+7733j0RERPJ3d09MxOZmREREXMZERUEApCZAZBEyMwA4K6ZqR6qahaZCUSZaWaq6mpmFhGIKJWj2is811oR8e3HT1VdywAAiADAzMxsUK8LiEhm9qLXPTOYCRHnnIjprpmOmAmuNiOitQYARISIACBFbxV/cY5IRB8fH3NO9xQRFM5MAETEql+XBSAAItMj7BVMwEL7vjPzWgsRAzEizOZaS/KlxcyU/M+pyYlIRD59+qSqNZ4lzDmfz+ecE381i4hy9RARERZhAEhEbfTWZds2AAiEtdYY4/F4rA+NCHePCJlzuvsFppkF0jV8rQAStdZEhIhY0N3nnJkp53xGBFKqTVUVEQfMTHVzd10WEdxbaw0gELEAV1UAkDUtM4XDLc1C1QMhIr59/1F6JqL39/d2/30fGyU9YTFz9SVADTI9aWqoh0SutY7j+Pn978xca2VmOmzbTq1HxL7vFrHtu5mJqiOiWTweB8DX3ntEmNnz+bzaO89zHicBQiA1ukQhxQ0AzDlVlYhKQKqKiIVccdZaQ2QRKTm6u1z0XlIr8AuYMo36AgAzV+qXSFQ900Rk3/vt9iYiVSez9hmImLkBECITUWV5aenKPca43+/btp3n+Xg8Pj4+apl77zVL9bWWlvMQkVQbIjLGuN1ub29vvXcRqd109977GKPoMbOlUSgiotSE12JVVhG53+9lN2OMbdsKArcsRWYmEUmTUUany4/nFO6ttdFvwn0be2aKiIgg4Jq61vrr61dmrngpGUTEcRw13rZtRNRau0AuqR3HcZ5neWgRJL2P11OApGrnORFrDVptPwBEpKqtpWupqlY8AIiqFhgRsdZ6PB5jjNZamWEBU8XXWqoqvSEiMzOzvHATwZddWKH6Mt3MctyLqmr4RVX9Sjr/PldEl5+WUZa8mPmcsxgBAKk6pZjivGT0P81euWqisir5/PlzQVrtXS9DbfLVznUhIOt5EpF/AJpfqn9dbNv7AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF284E340>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\5_3_dig6_20210723-134707.jpg 5 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3AA3310>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\6_D8_2020-08-19_20-22-27.jpg 6 -1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF43B6700>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\7_D7_2020-08-20_15-23-28.jpg 7 -1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF4003D00>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\8_3_dig4_20210725-094630.jpg 8 -1\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF70C5700>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\8_D7_2020-08-14_00-01-25.jpg 8 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3C31AC0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\8_D8_2020-08-20_15-23-28.jpg 8 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3CCC0A0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\NaN_0_ht6_20210628-182853.jpg -1 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF7011880>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\NaN_4_ht7_20210701-112725.jpg -1 4\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF43A1FA0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\NaN_5_dig6_20210629-114809.jpg -1 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3AAF6D0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ziffer_sortiert_resize\\NaN_Ziffer_NaN_1250.jpg -1 9\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=20x32 at 0x27BF3DD22E0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "Input_dir='ziffer_sortiert_resize'\n",
    "res = []\n",
    "only_deviation = True\n",
    "show_wrong_image = True\n",
    "\n",
    "files = glob.glob(Input_dir + '/*.jpg')\n",
    "\n",
    "for aktfile in files:\n",
    "    base = os.path.basename(aktfile)\n",
    "    target = base[0:1]\n",
    "    if target == \"N\":\n",
    "        zw1 = -1\n",
    "    else:\n",
    "        zw1 = int(target)\n",
    "    expected_class = zw1\n",
    "    image_in = Image.open(aktfile)\n",
    "    test_image = np.array(image_in, dtype=\"float32\")\n",
    "    img = np.reshape(test_image,[1,32,20,3])\n",
    "    classes = np.argmax(model.predict(img), axis=-1)\n",
    "    classes = classes[0]\n",
    "    if classes == 10: \n",
    "        classes = -1\n",
    "    zw2 = classes\n",
    "    zw3 = zw2 - zw1\n",
    "    res.append(np.array([zw1, zw2, zw3]))\n",
    "    if only_deviation == True:\n",
    "        if str(classes) != str(expected_class):\n",
    "            print(aktfile + \" \" + str(expected_class) +  \" \" + str(classes))\n",
    "            if show_wrong_image == True:\n",
    "                display(image_in)\n",
    "    else:\n",
    "        print(aktfile + \" \" + aktsubdir +  \" \" + str(classes))\n",
    "        \n",
    "\n",
    "res = np.asarray(res)\n",
    "\n",
    "\n",
    "plt.plot(res[:,0])\n",
    "plt.plot(res[:,1])\n",
    "plt.title('Result')\n",
    "plt.ylabel('Digital Value')\n",
    "plt.xlabel('#Picture')\n",
    "plt.legend(['real','model'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save the model\n",
    "\n",
    "* Save the model to the file with the \"h5\" file format"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\Muell\\AppData\\Local\\Temp\\tmpid40mq__\\assets\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "356508"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "FileName = TFliteNamingAndVersion\n",
    "\n",
    "converter = tf.lite.TFLiteConverter.from_keras_model(model)\n",
    "tflite_model = converter.convert()\n",
    "open(FileName + \".tflite\", \"wb\").write(tflite_model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\Muell\\AppData\\Local\\Temp\\tmp2mc527qd\\assets\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Assets written to: C:\\Users\\Muell\\AppData\\Local\\Temp\\tmp2mc527qd\\assets\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dig1320s2q.tflite\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "98416"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "FileName = TFliteNamingAndVersion + \"q.tflite\"\n",
    "\n",
    "import tensorflow as tf\n",
    "\n",
    "def representative_dataset():\n",
    "    for n in range(x_data[0].size):\n",
    "      data = np.expand_dims(x_data[5], axis=0)\n",
    "      yield [data.astype(np.float32)]\n",
    "        \n",
    "converter2 = tf.lite.TFLiteConverter.from_keras_model(model)\n",
    "converter2.representative_dataset = representative_dataset\n",
    "converter2.optimizations = [tf.lite.Optimize.DEFAULT]\n",
    "converter2.representative_dataset = representative_dataset\n",
    "tflite_quant_model = converter2.convert()\n",
    "\n",
    "open(FileName, \"wb\").write(tflite_quant_model)\n",
    "print(FileName)\n",
    "Path(FileName).stat().st_size"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Checking the images shows, that this are border line images, which can be interpreted as a good digit or a faulty one."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}