FashionMNIST_CNN.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "env37",
      "language": "python",
      "name": "env37"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.2"
    },
    "colab": {
      "name": "FashionMNIST_CNN.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/tfredrick112/DL-Assignments/blob/master/FashionMNIST_CNN.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o8-fZJWTRrLB",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 306
        },
        "outputId": "df91f1c3-edc4-4cd3-c996-ce443288b3c4"
      },
      "source": [
        "!nvidia-smi"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Sun Mar 29 17:11:29 2020       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 440.64.00    Driver Version: 418.67       CUDA Version: 10.1     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla P100-PCIE...  Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   33C    P0    26W / 250W |      0MiB / 16280MiB |      0%      Default |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                       GPU Memory |\n",
            "|  GPU       PID   Type   Process name                             Usage      |\n",
            "|=============================================================================|\n",
            "|  No running processes found                                                 |\n",
            "+-----------------------------------------------------------------------------+\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eSTLY3MGRhb1",
        "colab_type": "text"
      },
      "source": [
        "Import necessary libraries"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C1sKLTr8Rhb2",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from tqdm import tqdm"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gtY9fxQ3Rhb7",
        "colab_type": "text"
      },
      "source": [
        "Load the training and validation images, which are stored as rows in a dataframe. We will later reshape each image to get images of shape (28, 28)."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XkisLtCIRhb8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "dftrain = pd.read_csv('train.csv')\n",
        "dfval = pd.read_csv('val.csv')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yUyY4wOYRhcA",
        "colab_type": "text"
      },
      "source": [
        "Number of images belonging to each class:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eHavHxl7RhcB",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "outputId": "8162b462-b1ff-444a-f854-a236b4e54fab"
      },
      "source": [
        "dftrain['label'].value_counts()"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "7    5550\n",
              "6    5533\n",
              "2    5510\n",
              "1    5503\n",
              "5    5497\n",
              "3    5492\n",
              "8    5485\n",
              "0    5479\n",
              "9    5478\n",
              "4    5473\n",
              "Name: label, dtype: int64"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "q0i7LpIhRhcG",
        "colab_type": "text"
      },
      "source": [
        "We see that the images are evenly distributed into all the 10 classes. There is no skew in the dataset, so accuracy will be a good evaluation metric for our classification model.\n",
        "\n",
        "Defining a few constants:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fsC7KRJoRhcH",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# number of training examples\n",
        "M = dftrain.shape[0]\n",
        "\n",
        "# number of validation examples\n",
        "Mval = dfval.shape[0]\n",
        "\n",
        "# dimensions of the image\n",
        "img_rows = 28\n",
        "img_cols = 28\n",
        "\n",
        "# number of classes\n",
        "num_classes = len(dftrain['label'].value_counts())"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ChxLW8TURhcL",
        "colab_type": "text"
      },
      "source": [
        "Preparing the training and validation dataset and corresponding labels."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MxNFvMbmRhcM",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# pick only the pixel columns\n",
        "data_columns = [col for col in list(dftrain.columns) if col not in ['label', 'id']]\n",
        "\n",
        "dftrain_data = dftrain[data_columns]\n",
        "dftrain_labels = dftrain['label']\n",
        "\n",
        "dfval_data = dfval[data_columns]\n",
        "dfval_labels = dfval['label']\n",
        "\n",
        "# Training images: X_train is of shape (55000, 28, 28)\n",
        "X_train = np.array(dftrain_data).reshape((M, img_rows, img_cols, 1))\n",
        "# Validation images: X_val is of shape (5000, 28, 28)\n",
        "X_val = np.array(dfval_data).reshape((Mval, img_rows, img_cols, 1))\n",
        "\n",
        "# Labels for the training examples\n",
        "y_train = np.array(dftrain_labels)\n",
        "# Labels for the validation examples\n",
        "y_val = np.array(dfval_labels)\n",
        "\n",
        "# Convert the labels to one-hot vectors\n",
        "from tensorflow.keras.utils import to_categorical\n",
        "y_train_one_hot = to_categorical(y_train)\n",
        "y_val_one_hot = to_categorical(y_val)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qEBpTtHsRhcQ",
        "colab_type": "text"
      },
      "source": [
        "Import necessary layers, models and callbacks from tensorflow."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M1M6FIHxRhcR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import tensorflow as tf\n",
        "# layers in the CNN\n",
        "from tensorflow.keras.layers import Conv2D, Dense, Flatten, MaxPool2D, BatchNormalization\n",
        "from tensorflow.keras.models import Sequential\n",
        "\n",
        "# optimizer\n",
        "from tensorflow.keras.optimizers import Adam\n",
        "\n",
        "# fine tuning: patience, early stopping, saving the best performing model etc.\n",
        "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EDeEciXcRhcV",
        "colab_type": "text"
      },
      "source": [
        "Build the model layer-by-layer."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8079gTmiRhcW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def create_model(choice_init, batch_norm):\n",
        "    model = Sequential()\n",
        "\n",
        "    # padding = 'same', because the stride = 1 and kernel_size=(3, 3) and we want a padding of 1 layer.\n",
        "    model.add(Conv2D(filters=64, kernel_size=(3, 3), strides=(1, 1), padding='same',\n",
        "                     activation='relu', kernel_initializer=choice_init,\n",
        "                     input_shape=(img_rows, img_cols, 1)))\n",
        "    model.add(MaxPool2D(pool_size=(2, 2)))\n",
        "\n",
        "    # second convolutional layer\n",
        "    model.add(Conv2D(filters=128, kernel_size=(3, 3), strides=(1, 1), padding='same',\n",
        "                     activation='relu', kernel_initializer=choice_init))\n",
        "    model.add(MaxPool2D(pool_size=(2, 2)))\n",
        "\n",
        "    # third convolutional layer\n",
        "    model.add(Conv2D(filters=256, kernel_size=(3, 3), strides=(1, 1), padding='same',\n",
        "                     activation='relu', kernel_initializer=choice_init))\n",
        "\n",
        "    # fourth convolutional layer\n",
        "    model.add(Conv2D(filters=256, kernel_size=(3, 3), strides=(1, 1), padding='same',\n",
        "                     activation='relu', kernel_initializer=choice_init))\n",
        "\n",
        "\n",
        "    model.add(MaxPool2D(pool_size=(2, 2)))\n",
        "\n",
        "    model.add(Flatten())\n",
        "\n",
        "    # fully connected layers\n",
        "    model.add(Dense(256, activation='relu', kernel_initializer=choice_init))\n",
        "    model.add(Dense(1024, activation='relu', kernel_initializer=choice_init))\n",
        "\n",
        "    # Batch normalization before the softmax layer.\n",
        "    if batch_norm:\n",
        "        model.add(BatchNormalization())\n",
        "\n",
        "    # softmax layer for classification\n",
        "    model.add(Dense(num_classes, activation='softmax', kernel_initializer=choice_init))\n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-u6LmQNsRhca",
        "colab_type": "text"
      },
      "source": [
        "Compile the model."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UShpuqx9Rhcb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from tensorflow.keras.optimizers import Adam\n",
        "\n",
        "adam = Adam(learning_rate=0.001)\n",
        "model.compile(optimizer=adam, loss='categorical_crossentropy', metrics=['accuracy'])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JTl2m0xtRhce",
        "colab_type": "text"
      },
      "source": [
        "Actual training step"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "scrolled": true,
        "id": "C25PQa6IRhcf",
        "colab_type": "code",
        "colab": {},
        "outputId": "b5bd9ff3-704b-4ce7-de88-d6915dc32a65"
      },
      "source": [
        "history = model.fit(X_train, y_train_one_hot, batch_size=50, epochs=10, \n",
        "                    verbose=1, validation_data=(X_val, y_val_one_hot))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 55000 samples, validate on 5000 samples\n",
            "Epoch 1/10\n",
            "55000/55000 [==============================] - 249s 5ms/sample - loss: 0.4995 - accuracy: 0.8350 - val_loss: 0.2986 - val_accuracy: 0.8876\n",
            "Epoch 2/10\n",
            "55000/55000 [==============================] - 242s 4ms/sample - loss: 0.2849 - accuracy: 0.8963 - val_loss: 0.2754 - val_accuracy: 0.8942\n",
            "Epoch 3/10\n",
            "55000/55000 [==============================] - 236s 4ms/sample - loss: 0.2521 - accuracy: 0.9074 - val_loss: 0.2881 - val_accuracy: 0.8974\n",
            "Epoch 4/10\n",
            "55000/55000 [==============================] - 242s 4ms/sample - loss: 0.2262 - accuracy: 0.9178 - val_loss: 0.2531 - val_accuracy: 0.9060\n",
            "Epoch 5/10\n",
            "55000/55000 [==============================] - 241s 4ms/sample - loss: 0.2038 - accuracy: 0.9255 - val_loss: 0.2473 - val_accuracy: 0.9114\n",
            "Epoch 6/10\n",
            "55000/55000 [==============================] - 238s 4ms/sample - loss: 0.1900 - accuracy: 0.9285 - val_loss: 0.2764 - val_accuracy: 0.9006\n",
            "Epoch 7/10\n",
            "55000/55000 [==============================] - 240s 4ms/sample - loss: 0.1747 - accuracy: 0.9347 - val_loss: 0.2463 - val_accuracy: 0.9198\n",
            "Epoch 8/10\n",
            "55000/55000 [==============================] - 238s 4ms/sample - loss: 0.1596 - accuracy: 0.9417 - val_loss: 0.2771 - val_accuracy: 0.9070\n",
            "Epoch 9/10\n",
            "55000/55000 [==============================] - 242s 4ms/sample - loss: 0.1460 - accuracy: 0.9460 - val_loss: 0.2742 - val_accuracy: 0.9108\n",
            "Epoch 10/10\n",
            "55000/55000 [==============================] - 242s 4ms/sample - loss: 0.1345 - accuracy: 0.9499 - val_loss: 0.3053 - val_accuracy: 0.9094\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "a1yBn8DcRhcj",
        "colab_type": "text"
      },
      "source": [
        "Early Stopping:\n",
        "\n",
        "In order to prevent overfitting, we must ensure that we do not train the model for too many epochs. We must stop once the validation loss reaches its minimum value. That is precisely the role of EarlyStopping. It **monitors** the validation loss, and stops training when its **minimum** is reached. The role of the **patience** parameter is as follows: Sometimes, we can reach a plateau in the validation loss curve, where the loss remains constant for a few epochs and then starts decreasing again. When we set the patience epoch to 5(for example), we will stop training only if the validation loss does not improve over the next 5 epochs."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "62kv1dL8Rhck",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=5)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SryjQKFURhcp",
        "colab_type": "text"
      },
      "source": [
        "Model Checkpoint\n",
        "\n",
        "Since we are using 'patience', when we finally stop training, we might not stop at the epoch that gives us the best performing model. The loss curve might have started to increase again within those 'patience' number of epochs. So we can use the ModelCheckpoint class to save the best performing model."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PSNsN9RjRhcq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "mc = ModelCheckpoint('best_model.h5', monitor='val_accuracy', mode='max', verbose=1, save_best_only=True)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cQYbVU86Rhct",
        "colab_type": "text"
      },
      "source": [
        "Plotting Loss curves"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Q0V9_1KCRhcu",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def plot_loss_curves(history):\n",
        "    x_axis = list(range(1, len(history.history['loss'])+1))\n",
        "    plt.plot(x_axis, history.history['loss'], color='red', label='Training Loss')\n",
        "    plt.plot(x_axis, history.history['val_loss'], color='lime', label='Validation Loss')\n",
        "    plt.title('Training Loss and Validation Loss', size=18)\n",
        "    plt.xlabel('Number of epochs', size=14)\n",
        "    plt.ylabel('Loss', size=14)\n",
        "    plt.legend()\n",
        "    plt.grid()\n",
        "    plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lJtkNSLORhcx",
        "colab_type": "text"
      },
      "source": [
        "Putting everything together"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "90aqu5VDRhcy",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train_model(X_train, y_train, X_val, y_val, learning_rate=0.001, batch_size=50, \n",
        "                num_epochs=20, choice_init='glorot_uniform', batch_norm=True):\n",
        "    # Create the model\n",
        "    fashion_mnist = create_model(choice_init, batch_norm)\n",
        "    \n",
        "    # Compile the model\n",
        "    adam = Adam(learning_rate=learning_rate)\n",
        "    fashion_mnist.compile(optimizer=adam, loss='categorical_crossentropy', metrics=['accuracy'])\n",
        "    \n",
        "    # Early stopping with patience and model checkpoint\n",
        "    es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=5)\n",
        "    mc = ModelCheckpoint('best_model.h5', monitor='val_accuracy', mode='max', verbose=1, save_best_only=True)\n",
        "    \n",
        "    # Fit training data\n",
        "    history = fashion_mnist.fit(X_train, y_train, batch_size=batch_size, epochs=num_epochs, \n",
        "                                verbose=1, validation_data=(X_val, y_val), callbacks=[es, mc])\n",
        "    \n",
        "    # plot graphs\n",
        "    plot_loss_curves(history)\n",
        "    return history"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ydC045PARhc1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "ba00fb77-7db4-4f20-88a4-53b8eb5a71af"
      },
      "source": [
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 20\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "\n",
        "train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "1098/1100 [============================>.] - ETA: 0s - loss: 0.4363 - accuracy: 0.8441\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.81720, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.4361 - accuracy: 0.8441 - val_loss: 0.4966 - val_accuracy: 0.8172\n",
            "Epoch 2/20\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.2826 - accuracy: 0.8987\n",
            "Epoch 00002: val_accuracy improved from 0.81720 to 0.89640, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2826 - accuracy: 0.8987 - val_loss: 0.2944 - val_accuracy: 0.8964\n",
            "Epoch 3/20\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.2360 - accuracy: 0.9151\n",
            "Epoch 00003: val_accuracy did not improve from 0.89640\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2360 - accuracy: 0.9151 - val_loss: 0.2873 - val_accuracy: 0.8942\n",
            "Epoch 4/20\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.2086 - accuracy: 0.9251\n",
            "Epoch 00004: val_accuracy improved from 0.89640 to 0.91220, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2084 - accuracy: 0.9251 - val_loss: 0.2358 - val_accuracy: 0.9122\n",
            "Epoch 5/20\n",
            "1098/1100 [============================>.] - ETA: 0s - loss: 0.1825 - accuracy: 0.9348\n",
            "Epoch 00005: val_accuracy improved from 0.91220 to 0.91660, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1825 - accuracy: 0.9348 - val_loss: 0.2408 - val_accuracy: 0.9166\n",
            "Epoch 6/20\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.1598 - accuracy: 0.9422\n",
            "Epoch 00006: val_accuracy did not improve from 0.91660\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1597 - accuracy: 0.9421 - val_loss: 0.2890 - val_accuracy: 0.9044\n",
            "Epoch 7/20\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.1360 - accuracy: 0.9511\n",
            "Epoch 00007: val_accuracy improved from 0.91660 to 0.91940, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1357 - accuracy: 0.9513 - val_loss: 0.2290 - val_accuracy: 0.9194\n",
            "Epoch 8/20\n",
            "1092/1100 [============================>.] - ETA: 0s - loss: 0.1142 - accuracy: 0.9584\n",
            "Epoch 00008: val_accuracy did not improve from 0.91940\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1141 - accuracy: 0.9584 - val_loss: 0.2603 - val_accuracy: 0.9170\n",
            "Epoch 9/20\n",
            "1095/1100 [============================>.] - ETA: 0s - loss: 0.0988 - accuracy: 0.9644\n",
            "Epoch 00009: val_accuracy did not improve from 0.91940\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0986 - accuracy: 0.9645 - val_loss: 0.3055 - val_accuracy: 0.9192\n",
            "Epoch 10/20\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.0823 - accuracy: 0.9704\n",
            "Epoch 00010: val_accuracy improved from 0.91940 to 0.91960, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0823 - accuracy: 0.9704 - val_loss: 0.2764 - val_accuracy: 0.9196\n",
            "Epoch 11/20\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.0686 - accuracy: 0.9753\n",
            "Epoch 00011: val_accuracy improved from 0.91960 to 0.92480, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0685 - accuracy: 0.9754 - val_loss: 0.2911 - val_accuracy: 0.9248\n",
            "Epoch 12/20\n",
            "1098/1100 [============================>.] - ETA: 0s - loss: 0.0585 - accuracy: 0.9787\n",
            "Epoch 00012: val_accuracy did not improve from 0.92480\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0587 - accuracy: 0.9786 - val_loss: 0.2951 - val_accuracy: 0.9236\n",
            "Epoch 00012: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEeCAYAAACOtbLLAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOzdd3wU1drA8d9DEjoC0kQBaQIiPRAE\nlKKoKBJUQEVFEaXdCwpexXJREcWGrw1QVEAUS66iIgIqoEG4YKEIXqogRCmKFIXQ2/P+cSZhEzYh\nZUs2eb757CfZmdmZ5+xM9tk558wZUVWMMcaYFIXCHYAxxpi8xRKDMcaYNCwxGGOMScMSgzHGmDQs\nMRhjjEnDEoMxxpg0LDGEkYhUFxEVkRG5WMdkEbE+xxHC29+Tw7h9v8dcduISkRHe8tWDEF9vb93t\nA71uk3WWGHx4B2RWH9XDHW9e4r0nM8IdR34gIh9672eTTJYREdkkIn+LSLFQxpdbItLeSy5lwh2L\nPz7Jc2y4YwmX6HAHkMf0Svf8YqAf8DqwIN28HQHY3q9AMeBYLtbRFxgQgFhM3jER6A7cDtydwTId\ngOrAa6p6MADbLAYcD8B6sqI98CgwGfg73bwpQAJwJESxGD8sMfhQ1Xd8n4tINC4xfJt+XnoiUkpV\nk7O5PQUOZTvQtOs4ChzNzTpMnjMb2AzcLCL3qaq/D8nbvd8TA7FBVc3VcRgoqnqc0CUokwGrSsoB\nEUkSkXki0lREvhSRPcBP3rxSIvKEiHwvIjtF5LCIbBCRp0WkeLr1nFLf6ztNRK4WkcUickhEfheR\n0V6y8l3HKW0MKdNEpLSIvCoif3rrWCgiLf2Up5yITBKRXSKyT0S+9so2T0SSAvjWpZRviohs996b\nX0TkST/vzZki8oI3/5AX21IRuS/dcreKyA9elcp+EdkoIu+KSIUsxHKDiEwXkd+8WHaKyDQRaeRn\n2ZR9Xk9EZopIsojsEZGpInKWn+UvEJEvvJh2ezFVzMp7pKoncN+mywHxftZ9BtANWKmqi7NzzGXy\nXpzSxiAihUTkQa/K6pCIrBSRmzN4fT0ReUVEVnnvzQFvf92ZbrnJuLMFgE1ysmp2hDffbxuDiJQX\nkXEisllEjni/x4lIuXTLpbz+EhG51zt+DovIzyJyW1bei+wQkUYi8ol3fB4SkdUiMkxEotItV9X7\nH/vVi+dPEVnkG5P3fg8RkZ+893CviKwTkYkiEhPo2DNjZww5Vw34GvgQ+Ago6U0/B7jTm/Yerpqo\nHTAMaApckcX1XwX8AxgPTAK6AvcCfwFPZnEdX+KqvEbiPmTuAWaKSI2UsxsRKQLMBZrgPox+ABp5\n03ZncTtZIiLneusvDbwCrMdVKzwItBGRS1U1pVrtQ6Atrvw/4ao6zveWH+2trxfwFq6a7xHgIFAV\n995V5PTVfYOAXbiqwj+AWrgzxIUi0kxV16db/hxgHvAJcB/QGOgPnAFc7lPOGl5MRYCxuG//XYAv\nThOPrzeB4bgzg6np5t2Iez9SzhYCdcyl9zyuKms+8ALuPR0HbPSzbHvc/poBbAJKAD2AN0Skgqo+\n5S33Gu79uhYYCuz0pv+UURAiUhpYBNTG/S8s88o1ELhEROL8nK0/iXuPXgMOe8tOFpENqrowi+XP\nlIg0B77BnbGPwx1DXYBncMfGzd5y0cAc3H56BfgZ9z/QCFdd/Za3yn/j/lc/wx33x4EauC8HRQhl\nzYCq2iODB9AbUKB3uulJ3vQ7/bymMBDjZ/rj3mvifKZV96aN8DNtP1DdZ7oAK4Hf0613Ml6tVPpp\nwCvppvfwpvf3mfYPb9q/0y2bMj0pi++VAjNOs8y73nJXpZs+2pt+h/e8tL/4/azvY2AvEJ3D/VvC\nz7TzcR8k6d+7lH1+fbrp47zpdX2mvedN65Bu/33iTZ+cxfi+wn3IV043/VsvxvKBOOZ89t9kn+d1\ngRNeDFE+05t50zXd8envvSyES6R7fOMDRqR/vZ//ufY+00Z50/6Rbtl/etMf9/P6H4HCPtPP8d6z\n97Pwvqe8R2NPs9xCb/80SrefP/Bef6k3rZH3fNhp1rcMWJ2TYznQD6tKyrnduG91aajqEXX1/ohI\ntIiUFZHyuG/gAKdU5WRgmqom+axXgUTgLBEpmeGr0noh3fOvvd/n+Uzrgvtm8lK6ZSfg/qEDQkQK\n4b75/Kiqs9LNfgr3YXOt9/wg7p+4pWTe+2sPUBzoLCKS3ZhUdb8Xm4jIGd5+2gGsw/9+2qaqH6Sb\nluY99crZBViiqok+21Lg2WyGOBGIAm5NmSAi9YALgemqutNbd6COOV9dcR9yz6ur908pxzLct980\nUt5LL4aiXhXPmbj2kjOAejmIIcW1uP3yerrpr3nTrz3lFS6xp7bNqOpW3Df18/wsm21etWBr3H5I\nPdvx9vMon7jh5P9Rh9NUJ+4BzhGRiwIRY25YYsi5X3z/YXyJyD9E5Cfch9tu3ME7z5tdNovr93e6\nvsv7Xc7PvNOuQ1X9vb4G7gNvX7plj+CqBAKlAq66bVX6Gaq6G/gdqOmz7SFAA1w99CoRGSMil6Z7\n6ZO4nl3TgB0i8pGI3CkipbISkLh2lBlAMu6fcof3aIj//ZSVfVLRK+daP8uuzkpcPj7G9dq53Wda\nH+/3JN8FA3TM+arp/c5SOUSkpIg8JyK/4RL7Ti+GlA/JnMSQogawTk9WMwLgPf/ZJ1ZfGe2rrP7v\nZCUm8HM8A2twX3RSjudfce/D5cDvXtvLsyLSIt3rHsJ1RlkgIlu9dqmbRKRwgGLOMksMOXfA30QR\nuQdXvfA7rv65M3AZ7hQXsv6eZ9YzI0vfjjNKXFl9fTip6njcKX1f3Cl2d2CuiCT4LLMeqI97j98C\nzgXeANaKSK3M1i8i1XB1501xVS7X4v5xL8P9s/vbT7neJ9mhrqfQe0BdEWntNWj2Arbg2o/chgN3\nzOXGe7g2rFm4uvVOXgwpZ62h/qzJU8e+qg7Hna0MAX7BtQn9ICLP+CzzLa6dqzuu2rEJrvp1uYic\nGcp4rfE58Hrh6qOvVNe7BAAR6RS2iDKXBHQUkZK+Zw1eL4ganNrPPKd24L6ZX5B+hoiUBSoDy32n\nq+rvuCqtCd6H4hSgp4j8n6ou9pY5jPswmuWt6ypgJu5D6p+ZxHMt7pt9vG+Vj7eOcrhv3jmxA9iH\n/6qT+jlY30Rce8/tuKqZs4BRvscWwTnmUr5x18N9kPlKUw5xF6pdDUxR1QHp5nX0s+7sXqm/EZcc\no33PGrxG3Tr4PzsItpSz6VOOZ9x7VohTz9g3AmOAMSJSFJfch3nH85/eMvtwnQg+AncmiEv6d+B1\nuggFO2MIvOO4Az/1m4l3AD8Qtogy9xmuHjv9hVR9cY3AAeF9YH0GNPXzgfUA7lj8BEBEiku6bpbe\n2U9KXe6Z3nLl/Wxqme8ymUj5RpnmG6SI9MV9+OaIF+cMoLmIdPBZr+B6CWV3fctwCfMGTja2Tkq3\nWDCOueneOu/x7XopIs2A9B/2Gb2XlXHfjNNL+QKS1W/B03BVkenX1deb/kkW1xMw3gf5IqCLiDRI\nme7t5we9pynHc+n03U29s8E13tOy3nK5OZ4Dys4YAm8qrjH1cxH5GNfwdhN59yK0CbjqhydEpDYn\nu6teD2wge8dIbREZnsG8F3B1qJcB00TkFW/9bXEfevM52W2vDvCNiHyC64n1F6630EDcN7WUq9Bn\ni8jf3vPNQBlO9kqZcppYP8dVB04RN/TBX0AbXFfXX8jd/8Zw4EpghoiMwVX9dMF9iOXERNw3zU7A\nPO+bp6+AH3OqulZExuG69H4tIh/h2k8GAStwVXApyyaLyGzgFhE5CCzGVev1x+2v9PX633m/nxGR\nd3H16itVdWUG4TyL61E3zktMP3rbvwPXUSC7jfpZ1TyD4/mYqj6N+zL1Da5NIKW76tW47sHvqepX\n3vIdgNe993AdLjHG4hLd96q6zltujYh8B3wPbMOdRffDXQWeWoUaEuHuFpWXH2TeXXVeBq+Jwn1j\n2ICrjvgVd+CeT8ZdUzOd5jNvBKd2E5xMBt1VM4jvlO6SuA+sybhGy/24njZNgCVksfuct97MHmd5\ny9XAfWj/iTvgN+IakYv7rKscLpEsx1VlHfTezxfx6bqJ+8Y4B/cPeQRXxz4Ln26ip4m5LfBfXBXX\n37gqqAa4RtukrOxzXP99f8dIQ1yPnP3e+/ou7oM1y91VfdZV1nsPFOgV6GMuk+OiEK5v/a/eelfi\n2g/8HYflcV8ytuE+6P/n7Z/epOt+6i0/zNv3R33jyWT5CrhrALZ4r9mCq2Ipn8H/bHs/79Mp+zWD\n9zvlPcrocchn2ca4M5rd3nu0xiubbxffGrjrEtbgulfv9/4eCZT2We4B3BekP711bcZdz9MsO8dL\nIB7iBWRMGl71wU7cN5q82j5ijAkCa2MwiP/ROQfgqmZO6bNujMnf7IzBICLvAEVxjWmHgVa4Oupf\ncKex2Roc0BgT2SwxGETkVlyPlzq4LpzbcXX1D6vq9nDGZowJPUsMxhhj0oj47qrly5fX6tWrhzuM\nLNm/fz8lSpQIdxhBkZ/LBvm7fFa2yJWb8i1dunSnqvrtQh3xiaF69eosWbIk3GFkybx582jfvn24\nwwiK/Fw2yN/ls7JFrtyUT0R+zWie9UoyxhiThiUGY4wxaVhiMMYYk0bEtzEYY0Lj6NGjbNmyhUOH\nDoU7lCwrXbo0a9asOf2CESor5StatChVqlQhJibrt40OaWLwRtV8CTe2ywR1A1H5zu+NG1p2qzdp\nrKpOCGWMxhj/tmzZQqlSpahevTo5uGFeWCQnJ1OqVJbu2xSRTlc+VWXXrl1s2bKFGjVqZLhceiGr\nSvLG3hmHG3WyPm5cfX/j0/9HVZt4D0sKxuQRhw4doly5chGTFAyICOXKlcv2WV4o2xjigA2qulHd\nrRsTcPeVNcZECEsKkScn+yyUVUnn4IaRTbEF/zcp7yYibXH3ch2qqpvTLyAi/XDjlFOpUiXmzZuX\n7WBWn7GaheUW0ndT32y/Nqf27duXo1gjQX4uG+Tv8mW1bKVLlyY5ObKGzTp+/HjExZwdWS3foUOH\nsnf8hmp8b9x9TCf4PO+Fa0PwXaYcUMT7uz/w9enWGxsbqzkxTscpiq7VtTl6fU4kJiaGbFuhlp/L\nppq/y5fVsq1evTq4gZzGzp07tXHjxtq4cWOtVKmSnn322anPDx8+7Pc1e/fuVVXVxYsX6+DBg0+7\njVatWgUk1sTERO3cuXNA1pWZlPKdjr99ByzRDD5XQ1mVtBWo6vO8CicbmQFQ1V3q7uEL7qYfscEK\npgtdAJjO9GBtwhgTQOXKlWP58uUsX76cAQMGMHTo0NTnhQsX5tixYxm+tnnz5rz88sun3caiRYsC\nGXLECmViWAycJyI1RKQwcCOk/VT27hGbIp6T90QNuKpUpQlN+IzPgrUJY0yQ9e7dmwEDBtCyZUuG\nDRvGDz/8QKtWrWjatCmtW7dm/fr1gBs64uqrrwZgxIgR9OnTh/bt21OzZs00CaNkyZKpy7dv357u\n3btTr149br755pRaDWbNmkW9evWIjY3lrrvuSl1vVrz//vs0bNiQBg0acP/99wOuOqh37940aNCA\nhg0b8sILLwDw8ssvU79+fRo1asSNN96Y+zcrG0LWxqCqx0RkEPAlrrvqJFVdJSIjcac004G7RCQe\nOIa7VV7vYMYUTzxP8AQ72Ul5/N2H2xjj15AhsHx5YNfZpAm8+GK2X7ZlyxYWLVpEVFQUe/fuZcGC\nBURHRzN37lwee+wxPv3001Nes3btWhITE0lOTqZu3boMHDjwlH7+P/74I6tWreLss8+mTZs2LFy4\nkObNm9O/f3/mz59PjRo16NmzZ5bj3LZtG/fffz9Lly6lbNmyXH755UybNo2qVauydetWVq50t7z+\n+++/AXj66afZtGkTRYoUSZ0WKiG98llVZ6lqHVWtpaqjvGmPeEkBVX1QVS9Q1caq2kFV1wYznnji\nOcEJZjErmJsxxgRRjx49iIqKAmDPnj306NGDBg0aMHTo0Awv/urcuTNFihShfPnyVKxYke3bT73t\nSFxcHFWqVKFQoUI0adKEpKQk1q5dS82aNVOvCchOYli8eDHt27enQoUKREdHc/PNNzN//nxq1qzJ\nxo0bGTx4MF988QVnnHEGAI0aNeLmm2/mnXfeITo6tNciF+grn5vRjLM5m8/4jFu5NdzhGBM5cvDN\nPlh8h51++OGH6dChA5988glJSUm0a9fO72uKFCmS+ndUVJTf9omsLBMIZcuWZcWKFXz55ZeMHz+e\nDz74gEmTJjFz5kzmz5/PZ599xqhRo/jf//4XsgRRoMdKEoQudOELvuAwh0//AmNMnrZnzx7OOecc\nACZPnhzw9detW5eNGzeSlJQEwH/+858svzYuLo5vvvmGnTt3cvz4cd5//33atWvHzp07OXHiBN26\ndeOJJ55g2bJlnDhxgs2bN9OhQweeeeYZ9uzZw759+wJenowU6MQArjppH/uYx7xwh2KMyaVhw4bx\n4IMP0rRp06B8wy9WrBivvPIKnTp1IjY2llKlSlG6dGm/y3711VdUqVIl9ZGUlMTTTz9Nhw4daNy4\nMbGxsXTt2pWtW7fSvn17mjRpwi233MJTTz3F8ePHueWWW2jYsCFNmzblrrvuokyZMgEvT4Yy6sca\nKY+cXseQ4qAe1OJaXP+p/8zVerLC+sJHrvxcvki5jiEnstrPPzuSk5NVVfXEiRM6cOBAff755wO+\njazKD9cx5ElFKcrlXM50pqPY/a+NMZl74403aNKkCRdccAF79uyhf//+4Q4p4Ap043OKLnRhGtNY\nwQqa0CTc4Rhj8rChQ4cydOjQcIcRVAX+jAGgM50RxC52M8YYLDEAUIlKXMiFNjyGMcZgiSFVF7qw\nhCVsY1u4QzHGmLCyxOCJJx6AGcwIcyTGGBNelhg89alPTWpadZIxeVSHDh348ssv00x78cUXGThw\nYIavueqqq1iyZEnq3/7GHBoxYgTPPfdcptueNm0aq1evTn3+yCOPMHfu3OyE75fv4H55iSUGT8pV\n0F/xFfvZH+5wjDHp9OzZk4SEhDTTEhISsjxe0axZs3J8kVj6xDBy5Eg6duyYo3VFAksMPuKJ5xCH\nmEvuvwkYYwKre/fuzJw5kyNHjgCQlJTEtm3buPjiixk4cCDNmzfnggsu4NFHH/X7+urVq7Nz504A\nRo0aRZ06dbjoootYt25d6jJvvPEGLVq0oHHjxnTr1o0DBw6waNEipk+fzn333UeTJk345Zdf6N27\nN1OnTgXcFc5NmzalYcOG9OnTh8OHD6du79FHH6VZs2Y0bNiQtWuzPiZouIfntusYfFzMxZSmNNOZ\nTle7HbUxGRrCEJYT2GG3m9CEF8l4cL4zzzyTuLg4Pv/8c7p27UpCQgLXX389IsKoUaM488wzOX78\nOJdeeik//fQTjRo18ruepUuXkpCQwPLlyzl27BjNmjUjNtbdE+y6666jb193u9/hw4czceJEBg8e\nTHx8PFdffTXdu3dPs65Dhw7Ru3dvvvrqK+rUqcOtt97Kq6++ypAhQwAoX748y5Yt45VXXuG5555j\nwoQJp30f8sLw3HbG4COGGK7kSmYwgxOcCHc4xph0fKuTfKuRPvjgA5o1a0bTpk1ZtWpVmmqf9BYs\nWMC1115L8eLFOeOMM4iPj0+dt3LlSi6++GIaNmzIu+++y6pVqzKNZ926ddSoUYM6deoAcNtttzF/\n/vzU+ddddx0AsbGxqQPvnU5eGJ7bzhjSiSeeBBL4gR+4kAvDHY4xeVJm3+yDqWvXrgwdOpRly5Zx\n4MABYmNj2bRpE8899xyLFy+mbNmy9O7dm0OHDuVo/b1792batGk0btyYyZMnM2/evFzFmzJ0dyCG\n7fY3PPdLL70UlOG57YwhnU50Iooo651kTB5UsmRJOnToQJ8+fVLPFvbu3UuJEiUoXbo027dv5/PP\nP890HW3btmXatGkcPHiQ5ORkPvvs5IgHycnJVK5cmaNHj/Luu++mTi9VqhTJycmnrKtu3bokJSWx\nYcMGAKZMmZLhPSCyKi8Mz21nDOmUpSxtactnfMaTPBnucIwx6fTs2ZNrr702tUqpcePGNG3alHr1\n6lG1alXatGmT6eubNWvGDTfcQOPGjalYsSItWrRInff444/TsmVLKlSoQMuWLVOTwY033kjfvn15\n+eWXUxudAYoWLcqbb75Jjx49OHbsGC1atGDAgAHZKk/K8NwpPvzww9ThuVWVzp0707VrV1asWMHt\nt9/OiROumtt3eO49e/agqoEbnjujYVcj5ZHbYbf9eUFfUBTdqBsDul4bujly5efy2bDbkcuG3Q6h\nLnQBsEH1jDEFkiUGP2pRi/rUt3YGY0yBVHATw6FDkEmPgy504Ru+YQ97QheTMXmcq4EwkSQn+6zg\nJoYnn4SOHWGb/9FU44nnGMf4gi9CHJgxeVPRokXZtWuXJYcIoqrs2rWLokWLZut1BbdX0m23weOP\nw6RJMHz4KbNb0pIKVGA607mBG8IQoDF5S5UqVdiyZQs7duwIdyhZdujQoWx/KEaSrJSvaNGiaXo9\nZUXBTQy1arkzhgkT4MEHISoqzewoouhMZ6YxjaMcJYaYMAVqTN4QExNDjRo1wh1GtsybN4+mTZuG\nO4ygCVb5Cm5VEkC/fvDrrzBnjt/Z8cTzN3+zkIUhDswYY8KnYCeGrl2hQgV4/XW/sy/jMopQxHon\nGWMKlIKdGAoXhttvh+nT4fffT5ldkpJcwiVMZzqKNbgZYwqGgp0YAPr2hePHXSO0H/HE8wu/sJas\nj6VujDGRzBJD7dpw6aXwxhtw4tShtq/G3XbPqpOMMQWFJQbItBG6ClVoRjMbHsMYU2BYYgC45ppM\nG6HjiWcRi9hB5PTfNsaYnLLEAK4RunfvDBuhu9AFRZnJzNDHZowxIWaJIUXfvnDsGEyefMqspjTl\nHM6x6iRjTIFgiSHFeefBJZf4bYQWhHji+ZIvOUTObhlojDGRwhKDr379YNMmmDv3lFld6MJ+9pNI\nYhgCM8aY0AlpYhCRTiKyTkQ2iMgDmSzXTURURJqHMj6uuQbKl/fbCN2BDpSghFUnGWPyvZAlBhGJ\nAsYBVwL1gZ4iUt/PcqWAu4HvQxVbqiJFXCP0p5/CH3+kmVWUolzO5XzGZ3YVtDEmXwvlGUMcsEFV\nN6rqESAB6OpnuceBZyBMlfmZNELHE88WtrCc5aGPyxhjQiSUw26fA2z2eb4FaOm7gIg0A6qq6kwR\nuS+jFYlIP6AfQKVKlZiXyZ3YcqJx06YUHTOG7+PioNDJ3FkmpgzSWng56WVu+/W2bK933759AY81\nr8jPZYP8XT4rW+QKWvlUNSQPoDswwed5L2Csz/NCwDyguvd8HtD8dOuNjY3VgHv/fVVQnT37lFmt\ntbU202Y5Wm1iYmIuA8u78nPZVPN3+axskSs35QOWaAafq6GsStoKVPV5XsWblqIU0ACYJyJJwIXA\n9JA3QANcey2UK+e3ETqeeJaxjC1sCXlYxhgTCqFMDIuB80SkhogUBm6EkyPTqeoeVS2vqtVVtTrw\nHRCvqktCGKOT0gg9bRps355mVjzxAMxgRsjDMsaYUAhZYlDVY8Ag4EtgDfCBqq4SkZEiEh+qOLIs\ng0boetSjFrVstFVjTL4V0usYVHWWqtZR1VqqOsqb9oiqnvIpq6rtw3K2kKJuXWjXzlUn+VwJnXIV\n9Nd8zT72hS08Y4wJFrvyOTP9+8PGjfD112kmxxPPYQ4zB//3ijbGmEhmiSEzGTRCt6ENZShj1UnG\nmHzJEkNmihaF226DTz5J0wgdQwxXcRUzmclxjocxQGOMCTxLDKeT0gj91ltpJnehCzvYwfdhGLnD\nGGOCyRLD6dSrB23bntII3YlORBNtg+oZY/IdSwxZ0a8f/PILJJ4ccrsMZWhHO2tnMMbkO5YYsqJb\nNzjzzFMaobvQhdWs5hd+CVNgxhgTeJYYssK3EfrPP1Mnd6ELgFUnGWPyFUsMWdW3Lxw9mqYRuiY1\naUADq04yxuQrlhiy6vzz4eKLT2mE7kIX5jOfv/grjMEZY0zgWGLIjn79YMMG8Bn/PJ54jnOcL/gi\nfHEZY0wAWWLIju7doWzZNI3QccRRkYpWnWSMyTcsMWRHSiP0xx/Djh0AFKIQV3M1n/M5Rzka5gCN\nMSb3LDFkl59G6Hji2cMeFrAgjIEZY0xgWGLIrvr14aKLXHWSuwUpHelIEYpYdZIxJl+wxJAT/frB\n+vWpjdAlKEFHOjKd6Sga3tiMMSaXLDHkhJ9G6Hji2cQmVrM6jIEZY0zuWWLIiWLF4NZb0zRCX83V\nAFadZIyJeJYYcqpvXzhyBN5+G4CzOZvmNLfhMYwxEc8SQ05dcAG0aZOmEboLXfiO79jO9tO82Bhj\n8i5LDLnRrx/8/DN88w3g2hkUZSYzwxyYMcbknCWG3OjRA8qUSW2EbkxjqlLVqpOMMRHNEkNuFCsG\nvXrBRx/Bzp0IQhe6MJvZHORguKMzxpgcscSQW/36pWmEjieeAxzga74Oc2DGGJMzlhhyq0EDaN06\ntRG6Pe0pSUmrTjLGRCxLDIHQrx+sWwfz51OEIlzBFXzGZ3YVtDEmIlliCIQePaB06dRG6Hji2cY2\nlrEszIEZY0z2WWIIhOLFXSP01KmwaxdXcRWFKGRXQRtjIpIlhkDxaYQuT3la09oSgzEmIlliCJSG\nDaFVq9RG6HjiWc5yNrM53JEZY0y2WGIIpH79YO1aWLCALnQBsN5JxpiIY4khkK6/PrURui51OY/z\nrDrJGBNxLDEEUvHicMstMHUqsms38cSTSCLJJIc7MmOMyTJLDIHWrx8cPgxTptCFLhzhCLOZHe6o\njDEmyywxBFqjRnDhhfD667TR1pSlrFUnGWMiiiWGYOjXD9asIXrh91zFVcxkJsc5Hu6ojAm7jWy0\nASYjQEgTg4h0EpF1IrJBRB7wM3+AiPxPRJaLyH9FpH4o4wuY66+HM86A114jnnh2sYtv+TbcURkT\nNgc4QD/6UYta1KAGz/AMe9kb7rBMBnKdGEQkJovLRQHjgCuB+kBPPx/876lqQ1VtAjwLPJ/b+MKi\nRAnXCP3hh1zxVxwxxFi3VTPTOmEAACAASURBVFNgrWY1ccTxBm8wiEE0pjEP8ADVqMZwhrODHeEO\n0aSTrcQgIneJSDef5xOBg95ZQN3TvDwO2KCqG1X1CJAAdPVdQFV9v0KUgAgehc5rhC799qe0o521\nM5gCR1EmMpHmNOdP/uQLvmAMY/iSL1nCEjrSkSd5knM5l7u52y4GzUNENeufvSKyAeijqvNFpC0w\nE7gD6AaUUNWrM3ltd6CTqt7pPe8FtFTVQemW+ydwD1AYuERV1/tZVz+gH0ClSpViExISslyGUGr2\nj38QdeAAD87uwpg6Y3kt8TXqSJ1whxUU+/bto2TJkuEOI2jyc/mCUbb9Uft5vs7zfF3pa5r91YyH\n1jxEuSPlTlnut+K/8X7V95lTaQ4Al22/jJ6/9aTawWoBiSM/7zfIXfk6dOiwVFWb+52pqll+AAeB\nqt7fo4FJ3t/nAztP89ruwASf572AsZksfxPw1uliio2N1Txr4kRV0E2LP1QUvSnpJv1D/9Djejzc\nkQVcYmJiuEMIqvxcvkCXbbEu1lpaSwtpIX1Cn9Bjeuy0r/lVf9W79C4tpsVUVLS7dtelujTXsUTy\nfjugB3SrbtVVukr/q//VGTpDp+gUHaNjdKSO1KE6VMcsHZPj9QNLNIPP1ehsJpm9QEVgM3CZlxwA\njgJFT/ParUBVn+dVvGkZSQBezWZ8ecsNN8CQIVR/eTpN327Ke+e+x3u8RxGKUIUqVKMaValKNe8n\n5e+qVKUUpcIdvTHZoigv8RLDGEYlKvEN33ARF2XptdWoxku8xHCG8xIvMZaxTGUql3M5D/EQbWmL\nIEEuQWCd4AR72ctf3s/f/O33d0bzDnM40/WXoAQDiw8MSuzZTQyzgTdEZBlQG/jcm34BsOk0r10M\nnCciNXAJ4UbcWUEqETlPT1YddQZOqUaKKCmN0JMmMeulH5m45WNKNyzNb/zGZjbzG7/xNV+zjW2c\n4ESal5ahzCkJw/fvszmbGLLU7m9M0O1kJ7dzOzOYQTzxvMmbnMmZ2V5PBSrwBE9wH/cxnvE8z/O0\npz2taMWDPMjVXJ2nEsRxjrOCFSzwfjayMfWDfQ97Mr1ZVxRRlPF+yno/ValKWcqmTvOd5zutDGWI\nIYZ5f8yDeoEvV3YTwz+BUUA1oLuq7vamNwPez+yFqnpMRAYBXwJRuGqoVSIyEndKMx0YJCIdcWcg\nfwG3ZTO+vKdfP3j1Vc6aMoc2jdrQnvanLHKMY2xjW5qE4fv3Ihaxm91pXlOIQlSmcobJozrVc/SP\nadJayUqGMYy2Z7b1u+8MzGc+N3ETO9jBS7zEYAbn+sO7NKW5n/u5i7t4kzcZzWjiiachDXmAB7ie\n64nO9sdX7h3iEItZnJoIFrIwdcib6lTnAu/H34d5+g/6kpTMU0kujYzqmCLlkafbGFK0aKF6wQWa\n+PXXOV5Fsibral2tX+qX+oa+oQ/rw9pbe+sleonW1tpaRIso6X6G6tAs1e8GQiTX5WYkQRO0uBZX\nFI0+Hq2f6CfhDikocrrvjukxfUwf00JaSGtr7YC0CWTkiB7Rt/Vtra/1FUVrak0dr+P1oB7M9HW5\nPS736B79Qr/Qh/QhvVgvTvN/doFeoAN0gL6n7+lm3Zyr7eRUbspHJm0M2W18rg/U9Xl+GfAO8CAQ\nlZ11BeoREYlhwgRV0KVjct5QdDon9IRu1+26WBfrx/qx3ql3Kop21s66V/cGbbsp8lNiOKpH9R69\nR1G0jbbR1bpaz99zvkZrtH6oH4Y7vIDLyb7bqlu1g3ZwnSr0ppAcY6qqx/W4TtNpGqdxiqJn6Vk6\nWkdnuP3slm27bteP9CO9W+/WZtpMC2khRdEojdI4jdN/6b90mk7TnbozAKXJvbySGL4DbvT+rgoc\nAGYBW4CnsrOuQD0iIjEkJ6uWKqW7mzRR/euvkG32FX1FozRKG2pDTdKkoG4rvySG7bpd22t7RdFB\nOkgP62FVVZ0xf4a20TYapVH6vr4f5igDK7v7bpbO0gpaQYtrcZ2kk/SEnghOYJk4oSd0rs7VS/VS\nRdGyWlYf1od1h+5Is9zpypakSfq2vq19ta/W1bqpZwNFtai21/b6iD6ic3SOJmtyEEuTc3klMfwN\n1PH+Hgoken93AJKys65APSIiMaiqvvGGHo+KUq1ZU3Vp8E6505uts7W0ltaKWlEX6aKgbSc/JIbv\n9Ds9R8/RolpU39a308xLTEzUZE3WdtpOC2khnaJTwhRl4GV13x3Ww3qv3qso2lAb6mpdHdzAsuh7\n/V6v0WsURYtrcR2iQ1KrdnzLdkJP6Cpdpa/qq3qT3qRVtWpqIiitpbWzdtan9WldqAtTvxDkdXkl\nMSQD1b2/ZwD3eX9XAw5mZ12BekRMYlDVpS+/rHrOOapFiqiOH696IjTftNboGq2ltbSIFtF39J2g\nbCOSE8MJPaGv6WtaWAtrDa2hP+qPpyyTUr59uk8v0UtUVHSSTgpxpMGRlX23UTemVt8M0AF6QA8E\nP7BsWqWr9Fa9VaM0SmM0Ru/QO/S5H5/T/9P/067aVctpudREcJaepT20h47RMbpcl4esLS7Q8kpi\n+BZ4BrgYd7FbQ296K2BzdtYVqEckJYbExETVP/9UveIK99bfdJOrZgqBnbpT22k7RdGH9eGAX2QX\nqYnhoB7UPtpHUbSTdtJdusvvcr7lO6AH9HK9XFH0NX0tRJEGz+n23Yf6oZbW0nqGnqEf6AehCSoX\nNukm/af+M01DcU2tqb21t07Uibpe14el+isY8kpiaAvsBo7jXfXsTX8K+Cg76wrUI+ISg6rq8eOq\njz+uWqiQar16qitXhmT7h/Vw6odgD+2h+3V/wNYdiYkhSZM0VmNTk2Vm3xrTl++gHtTO2llRdKyO\nDXKkwZXRvjugB3SADlAUjdM43agbQxtYLv2hf+ion0bpVt0a7lCCJliJIVuD6KnqfKACUF5V+/jM\neg0IziV4+VGhQjB8OMyZA7t3Q1wcTJkS9M0WpjATmMBoRjOVqbSjHb/ze9C3mxfNZS6xxLKe9XzK\np4xkJFFEZfn1RSnKR3xEV7oyiEG8yItBjDb01rCGlrRkPOO5l3tZwAJqUCPcYWVLJSrReldrzubs\ncIcScbI97LaqHseNqNpARC4QkaKqmqSqfwYhvvztkktg+XJo0QJuvRX69oWDwb2JiSDcy71MYxpr\nWEMccfzIj0HdZl6iKM/wDFdwBWdxFotZTDzxOVpXEYrwIR/SjW4MZSijU0eIiVyKMolJNKc5v/M7\ns5jFaEZTmMLhDs2EUHaH3Y4WkdG4q5JXAP8D/hKRZ7N6XwaTTuXKMHcuPPQQTJgArVrB+uCPBBJP\nPAtZiCBcxEVMY1rQtxluySTTgx48wAN0pzvf8R11yN1otzHEkEACN3IjwxjGKEYFKNrQSyaZW7iF\nO7iDlrRkBSu4kivDHZYJg+yeMTwL3AIMAOoA5+GqkHrh2hlMTkRHw6hRMHMmbN4MsbEwdWrQN9uY\nxvzADzSgAddxHc/wTKZju0SytawljjimMY3neI4EEihJYIZjjiaaKUzhFm5hOMN5jMci7n1cxjKa\n0YwEEhjJSOYwx6pgCrDsJoabgDtU9S1V/cV7TAbuBG4OeHQFzVVXwY8/Qv360KMH3H03HDkS1E2e\nxVnMYx7Xcz0P8AB96HPaUR0jzcd8TBxx7GIXc5jDv/hXwMeoiSaayUymN70ZwQge5uGISA6K8tE5\nH9GKVhzkIIkk8jAPZ6u9xeQ/2U0MpYFf/Ez/BSiT+3AM1arB/PkwZAi8/DJcfDH8+mtQN1mMYrzP\n+zzKo0xmMpdxGTvZGdRthsJxjvMgD9KNbpzP+SxjGR3oELTtRRHFRCbSl76MYhT3c3+eTg5LWMIV\nXMHY88ZyOZeznOW0pW24wzJ5QHYTwwrgLj/T7/bmmUAoXBheeMFVJ61dC02bwowZQd2kIIxgBO/z\nPj/wAy1pyRrWBHWbwbSTnXSiE0/zNP3pz3zmU4UqQd9uIQoxnvH8g38wmtHcwz15LjmsYAVd6UoL\nWrCUpQxeP5jpTKc85cMdmskjspsYhgG3efd4fst7rMO1O9wb+PAKuG7dYOlSOPdc6NIFHngAjh0L\n6iZv5Ea+4Rv2s59WtGI2s4O6vWBYylJiiWUBC5jABMYzniIUCdn2C1GIsYzlbu7mRV5kMINPud9G\nOKxiFT3oQROa8A3fMJKRbGIT1229Lu8O/2zCIifXMdQBpgIlvceHwBX4P5MwuVW7Nixa5O7r8Mwz\nrovrtm1B3WRLWvIDP3Au53IVVzGOcUHdXiC9yZu0oQ2K8l/+yx3cEZY4BOEFXuA+7mMc4xjIwLAl\nh5/5mZu5mYY05Au+YDjD2cQmHuZhzuCMsMRk8rZs3+lCVbcB//adJiKNgW6BCsqkU6wYvPaaa2/o\n3x+aNIH33oOOHYO2yWpU47/8l5u5mUEMYg1reJEXw3JzlKw4zGGGMITxjOdSLuV93qcCFcIakyA8\nwzPEEMOTPMlRjvIGb4SsYXcjG3mcx3mbtylKUYYxjHu516qMzGll+wI3E0a33AKLF0OFCnD55TBy\nJBw/HrTNlaIUn/AJ93Iv4xhHZzqzhz1B215ObWEL7WjHeMYzjGF8wRdhTwopBOEJnuBRHuVN3uR2\nbuc4wdtnAL/xG/3oR13qkkACd3M3G9nI0zxtScFkSd78+mcyVr8+/PADDBgAjz4KCxfCO++4ZBEE\nUUQxmtHUox4DGEArWvEZn1GLWkHZXnbNYx43cAMHOMCHfEh3uoc7pFOkNOzHEMNwhnOUo0xhSsDP\nvraxjSd5kjd4A4D+9OchHrLrEUy22RlDJCpRAt5+G15/Hb75xvVa+u9/g7rJO7iDOcxhO9tpSUsW\nsCCo2zsdRXmBF+hIR8pSlh/4IU8mBV//5t88wzOpV0of5WhA1vsnf3IP91CLWrzGa/SmN+tZz1jG\nWlIwOZKlrywiMv00i1gLVqiJuLGVWrSA7t2hfXt4+mn417/cvCBoT3u+53uu5mou5VJe53V60zso\n28rMfvZzJ3eSQALXci2TmRwxjajDGEYMMdzDPfSgB//hPznuMbWLXYxmNGMYwyEOcSu38jAPU5Oa\nAY7aFDRZPZfdlYX5m3IZi8mJJk1cl9Y77oD77nMXx731FpQtG5TN1aY23/It13M9t3M7a1nLkzyZ\nq3Ue5jA72ckOdqT+Tvnx93yXdzg+xVPcz/0R19VyKEOJIYbBDKYb3ZjKVIpSNMuv/5u/+T/+jxd5\nkf3s5yZu4hEeyfW4T8akyFJiUNXbgx2IyYXSpeHDD2HMGLj3XmjWDD74wJ1NBEFZyjKLWdzFXTzD\nM6xjHf0K9QNcFc8+9mX6wZ7++V72+t2OIJSjHBW8n/rUpzzlqUAFruAKLuKioJQvFAYxiBhiGMAA\nruEaPuETilEs09fsZS8v8RL/x/+xhz30oAcjGEF96ocoalNQWONzfiECd90FLVvC9dfDRRe5Buoh\nQ6BG4MfRjyGGV3iF8zmfoQxlfsv5FKMYO9mZ4VhLhSmc+iFfgQrUolbqB33Kj+/zspTN12P29Kc/\nMcRwJ3fShS5MZzrFKX7KcvvZz1jG8izPspvddKUrj/EYjWkchqhNQWCJIb9p2RKWLXNnDq+8AmPH\nwnXXubaHCy8M6KYE4S7uoi51GbV3FLUr1M70g74kJSOu2ifY+tCHGGLoTW+u4ipmMCN11NeDHGQ8\n43map/mTP7mSKxnJSJrTPMxRm/zOEkN+VK4cvPkmPPGEq1567TU37lLr1i5hxMdDVOC+iV/BFRRZ\nVYT27dsHbJ0FSS96EU00vehFJzoxjWkkkMCTPMnv/E5HOjKSkbSiVbhDNQWEdVfNz845x/VU2rwZ\nXnoJfv/dnT3UrQvjxsH+/eGO0Hh60pP3eZ/v+Z7KVGYwgzmP85jHPOYwx5KCCSlLDAVByZKu/eHn\nn10jdfnyMGgQVK0K//63Sxgm7HrQg4/4iKu4ijnMYR7zaEe7cIdlCiBLDAVJdLS75uHbb90Fce3b\nw1NPQfXqcPvtsHJluCMs8OKJ51M+pSMdrT3GhI0lhoJIBNq0gY8/dmcRffu67q0NG0KnTjBnDmje\nuoeAMSZ0LDEUdLVru55Lmze7+06vWOEG6GvSxF0oF+Rbixpj8h5LDMY580x46CFISoJJk+DECejd\n21UzPfUU/PVXmAM0xoSKJQaTVpEirr3hp5/giy+gQQOXMKpWdQ3YGzeGO0JjTJBZYjD+icAVV8Ds\n2bB8ubvN6PjxcN55JxuwjTH5kiUGc3qNG7v2hqQkGDYMvvrKXSzXujV89FFQbxZkjAk9Swwm684+\n27U3bN4ML78Mf/zhzh7q1OGcjz+G5ORwR2iMCQBLDCb7SpaEwYNh/Xp3wVzFipw3Zoy70nrwYFi7\nNtwRGmNyIaSJQUQ6icg6EdkgIg/4mX+PiKwWkZ9E5CsROTeU8ZlsiopKbW9Y+sorcM017q5y55/v\nurxOn27VTMZEoJAlBhGJAsYBVwL1gZ4ikn4g+R+B5qraCJgKPBuq+EzuJJ9/vrvd6ObNbvC+1auh\na1d3ncTo0bDrdPd6MsbkFaE8Y4gDNqjqRlU9AiQAXX0XUNVEVT3gPf0OqBLC+EwgVKzoxl9KSnIj\nulav7hqsq1Rxd5n78cdwR2iMOQ3REA19ICLdgU6qeqf3vBfQUlUHZbD8WOAPVX3Cz7x+QD+ASpUq\nxSYkJAQv8ADat28fJUuWDHcYQZFZ2Ups3Mg506ZRac4cog4dYs8FF7D1uuvYcfHFaExMiCPNmYK6\n7yJdfi4b5K58HTp0WKqq/m/uoaoheQDdgQk+z3sBYzNY9hbcGUOR0603NjZWI0ViYmK4QwiaLJXt\nr79UX3hBtXZtVVA96yzVRx9V3bYt2OHlWoHfdxEqP5dNNXflA5ZoBp+roaxK2gpU9XlexZuWhoh0\nBP4NxKuq/3tEmshUpoy71ei6dTBrlrs39WOPQbVq0LMnLFxog/cZkweEMjEsBs4TkRoiUhi4EZju\nu4CINAVewyWFP0MYmwmlQoXgyith5kzX5XXwYPj8c3ef6mbN3FhNBw+GO0pjCqyQJQZVPQYMAr4E\n1gAfqOoqERkpIvHeYqOBksCHIrJcRKZnsDqTX9SuDc8/D1u3uluQHjvmGqmrVHGN1klJ4Y7QmAIn\npNcxqOosVa2jqrVUdZQ37RFVne793VFVK6lqE+8Rn/kaTb5RogT06+cG75s3Dy65xCWMmjVdt1e7\nR4QxIWNXPpu8RQTatXNXVCclua6v333nLpg7/3wYMwb27g13lMbka5YYTN5VpQo8/jj89htMmeIa\nr++6yw29MWCASxh2FmFMwFliMHlfkSJwyy0uEfzwgxsC/O23oVUrqFcPnnzSXXFtjAkISwwmsrRo\nAZMnu5FdJ06Es85y1U3nnguXXQbvvAP794c7SmMimiUGE5nOOAP69IFvvoFffoFHHnG/e/VyyaJP\nH5g/392i1BiTLZYYTOSrWRNGjIANG1yi6NHDNV63a+e6w44YYbckNSYbLDGY/KNQIWjb1l0g98cf\nrsG6Vi0YOdL9btfOzbMbChmTKUsMJn8qUcI1WM+ZA7/+CqNGuWRxxx1QqZKrcpo71+4XYYwflhhM\n/le1Kjz0kLuz3KJFcOut8NlnrrG6enXXeP3zz+GO0pg8wxKDKThEXBfX8ePd2UNCAjRsCE8/DXXr\nnpz311/hjtSYsLLEYAqmokXhhhvcKK9btri7zCUnw8CBULnyyXnHjoU7UmNCzhKDMZUrw733wv/+\nB0uWuDGbvvoKOnd21VD33UdxG8zPFCCWGIxJIQKxsfDyy7BtG3z8MbRsCS++SNztt0ObNvDmm3YB\nncn3LDEY40/hwnDttTBtGmzdyi8DBsCuXe7CucqVoX9/WLzYxmoy+ZIlBmNOp2JFNt9wA6xZAwsW\nwHXXuWsk4uKgSRM34uvu3eGO0piAscRgTFaJuLvMTZ4Mv/8Or74KMTFuxNezz4abb4bERBuGw0Q8\nSwzG5ETp0m7o7yVLYNkyuPNOd6vSSy6BOnXgqadc8jAmAlliMCa3mjaFsWNdIpgyxd1H4qGHXI+m\nrl3dxXTW7dVEEEsMxgRKsWJuGI5582DdOtcF9vvvIT4eqlVzV1j/8ku4ozTmtCwxGBMMdeq4K6o3\nb4ZPPoFmzdzz2rXh0kvdVdeHDoU7SmP8ssRgTDDFxMA118CMGW4wv8cfd0OA9+zpblE6ZIi7sM6Y\nPMQSgzGhUqUKDB/uqpPmzHGD+L36KjRqBBdeCBMm2JDgJk+wxGBMqBUqBB07uuqkrVvhhRdcQujb\n110816cPfP21DQluwsYSgzHhVL68q05auRK+/RZuvNHdfe7SS12D9b/+BUuX2hXWJqQsMRiTF4ic\nrE7avh3+8x9o0cJdVd28OdSrB489BuvXhztSUwBYYjAmryleHK6/3o3TtH07vPGGu7L6scdcb6e4\nOHjxRbuAzgSNJQZj8rKyZd1V1YmJ8Ntv8Nxzru1h6FDXmN2xoxvxdc+ecEdq8hFLDMZEiipVTrY5\nrFnjLphLSnKN1ZUqQbdu8NFHdn2EyTVLDMZEonr1YORI1+bw/fdu3KaFC6F7d5ck+vSBuXOtZ5PJ\nEUsMxkQykZNtDlu2wOzZbljwqVPddRJVqrheT3bvCJMNlhiMyS+io10yePNN12g9dSq0bu0uoouL\ncw3Xjz7qxnEyJhOWGIzJj4oVO9nmsH07TJwI557rhuSoV891gX3+eXeBnTHpWGIwJr8rU+Zkm8OW\nLe5KaxHXkF21Kk0HDXIN2XPmwIED4Y7W5AGWGIwpSM4++2Sbw7p1rmoJ4Nln4fLLXRK5+GJ45BE3\nLMfBg+GN14SFJQZjCiqvzeHHsWPhr7/g88/d9RFHjsCoUW5YjjJloH17GDECvvnGusIWENHhDsAY\nkweULAmdOrkHwN69sGCBu+lQYqJrm3jsMShaFFq1gg4d3CMuDgoXDmvoJvBCesYgIp1EZJ2IbBCR\nB/zMbysiy0TkmIh0D2VsxhgfZ5wBnTvD6NHuvta7dsH06TBwoDu7ePRRV+VUpozrCfXkk7BoERw9\nGu7ITQCE7IxBRKKAccBlwBZgsYhMV9XVPov9BvQG7g1VXMaYLChTBrp0cQ+A3bth/nx3NpGY6Bqv\nAUqUgIsuctVPHTpAbKzrRmsiSij3WBywQVU3AohIAtAVSE0MqprkzTsRwriMMdl15pnuznTXXOOe\n79iRNlE8+KCbXrKkO7NIqXpq2hSiosIXt8kS0RBdDelVDXVS1Tu9572Alqo6yM+yk4EZqjo1g3X1\nA/oBVKpUKTYhISFocQfSvn37KFmyZLjDCIr8XDbI3+ULRtlidu+mzIoVlFm+nDLLl1Pit98AOFai\nBH83acJfsbHsjo3lYNWqrutskOTn/Qa5K1+HDh2Wqmpzf/Mi8hxPVV8HXgdo3ry5tm/fPrwBZdG8\nefOIlFizKz+XDfJ3+YJWtuuuO/n377/DvHlEJyZSfu5cyi9c6KZXrepGiO3Y0fWCqlQpoCHk5/0G\nwStfKBPDVqCqz/Mq3jRjTH5XuTL07OkeABs3ugvq5s5195148003vVEjlyQuu8xVQZUoEb6YC7BQ\n9kpaDJwnIjVEpDBwIzA9hNs3xuQVNWtC//7uNqY7drgL7p56yt3qdOxYuPJK147RoYO7puKHH2yk\n2BAKWWJQ1WPAIOBLYA3wgaquEpGRIhIPICItRGQL0AN4TURWhSo+Y0yYREW5sZseeAC++sp1h/3y\nS7j7bvj7bxg+HFq2dEnjuuvcoIDr19tosUEU0jYGVZ0FzEo37RGfvxfjqpiMMQVV8eJueI7LL3fP\nd+xww3PMmeMen3zipp977slqp0sugQoVwhdzPhORjc/GmAKkQgW44Qb3UIUNG1zbxJw5bmjxiRPd\nck2bnmzIvvhiN8KsyRFLDMaYyCEC553nHgMHwrFjsGzZyYbsF190V2sXKQJt2lCtRg33mhYt3JmI\nyRJLDMaYyBUd7cZriotzV1/v3+/GePLOKGp+/bU7o4iOhsaN3Y2LWrVyj3PPDeo1FJHMEoMxJv8o\nUSLNYID//fRTLoqOhm+/dWM5TZoEY8a4ZStXdgkiJVk0a+YGCTSWGIwx+dex0qXduE2dO3sTjsH/\n/ncyUXz7LXz8sZtXuLBLDimJonVrd/+KAsgSgzGm4IiOdo3UTZvCP/7hpm3fnjZRjBvnbnsKUK1a\n2uqnJk0gJiZ88YeIJQZjTMFWqVLaAQGPHIHly08mioULIWU8tmLF3DUXvsmiYsXwxR4klhiMMcZX\n4cInG7SHDHHTtmxJe1bx/PMn7z1Rq9bJRBEbC+efD6VKhS/+ALDEYIwxp1OlCvTo4R7g7oW9bNnJ\nRDF7NkyZknb5888/9VGhQkT0hLLEYIwx2VWsGLRp4x7gLrxLSoIVK2DNmpOPiRNdF9oUZ57pEkT9\n+mkTRtWqUCikN9TMlCUGY4zJLRGoUcM9UtoqAE6ccNVQvslizRrXE2rXrpPLFS8O9eqlTRb167tq\nqjA0dltiMMaYYClUyPVsqlYNrrgi7bwdO05NGPPnw7vvnlwmOtpd5Z2+SqpevaBeyW2JwRhjwqFC\nBfdo2zbt9ORkWLs2bcJYuRI+/TTt0OPnnkvFXr3cdRoBZonBGGPyklKl3NhOLVqknX74sBtu3Cdh\nHClbNighWGIwxphIUKQINGjgHp6/580LyqbyTjO4McaYPMESgzHGmDQsMRhjjEnDEoMxxpg0LDEY\nY4xJwxKDMcaYNCwxGGOMScMSgzHGmDREVcMdQ66IyA7g13DHkUXlgZ3hDiJI8nPZIH+Xz8oWuXJT\nvnNVtYK/GRGfGCKJiCxR1ebhjiMY8nPZIH+Xz8oWuYJVPqtKMsYYk4YlBmOMMWlYYgit18MdQBDl\n57JB/i6flS1yBaV81sZgjDEmDTtjMMYYk4YlBmOMMWlYYggBEakqIokislpEVonI3eGOKdBEJEpE\nfhSRGeGOJZBEpIyIqHpcEQAACFJJREFUTBWRtSKyRkRahTumQBGRod7xuFJE3heRouGOKTdEZJKI\n/CkiK32mnSkic0Rkvfc7OLc8C7IMyjbaOy5/EpFPRKRMoLZniSE0jgH/UtX6wIXAP0WkfphjCrS7\ngTXhDiIIXgK+UNV6QGPySRlF5BzgLqC5qjYAooAbwxtVrk0GOqWb9gDwlaqeB3zlPY9Ekzm1bHOA\nBqraCPgZeDBQG7PEEAKq+ruqLvP+TsZ9uJwT3qgCR0SqAJ2BCeGOJZBEpDTQFpgIoKpHVPXv8EYV\nUNFAMRGJBooD28IcT66o6nxgd7rJXYG3vL/fAq4JaVAB4q9sqjpbVY95T78DqgRqe5YYQkxEqgNN\nge/DG0lAvQgMA06EO5AAqwHsAN70qskmiEiJcAcVCKq6FXgO+A34HdijqrPDG1VQVFLV372//wAq\nhTOYIOoDfB6olVliCCERKQl8BAxR1b3hjicQRORq4E9VXRruWIIgGmgGvKqqTYH9RG5VRBpeXXtX\nXPI7GyghIreEN6rgUtc3P9/1zxeRf+Oqq98N1DotMYSIiMTgksK7qvpxuOMJoDZAvIgkAQnAJSLy\nTnhDCpgtwBZVTTm7m4pLFPlBR2CTqu5Q1aPAx0DrMMcUDNtFpDLw/+2deehUVRTHP98oy61Vw+BH\nWlK2WeFSRmm/iLDFCqygslwo6p+swLIyLTMMW4mK9sUQM9IobcdQbEHaLJfU0lwyaYUyTf21nf44\nd+LNayZ/yzSjv84HLjPvzn33nHuZeWfuue+dQ3r9rsb6VBRJw4FBwBCr4ENpYRiqgCThfuplZnZP\nrfWpJGZ2g5nVmVk3fPNyjpm1in+eZvYNsE5Sj1R1MrC0hipVki+BfpLape/nybSSjfUcs4Bh6f0w\nYGYNdakokk7FXbhnmdnmSvYdhqE6HA9cjP+b/iSV02utVNAoRgJTJS0CjgZuq7E+FSGtgmYAC4DF\n+LVghw4fIWkaMB/oIekrSZcAk4BTJK3AV0mTaqljcykztgeAjsDsdE15uGLyIiRGEARBkCVWDEEQ\nBEERYRiCIAiCIsIwBEEQBEWEYQiCIAiKCMMQBEEQFBGGIWj1SJq8vUV9lXR2ivj5u6TJtdanHJLq\nJZmkTrXWJageYRiC/5R0UTZJ43L1//cLzhP4k/Bd8ci0QbDdEIYhqAZbgWslda61IpUkhTlpznl7\nAvsAb5jZejPbUFnNgqBlhGEIqsFcYA0wrlyDUisISd1SXZ9cm9MkfSRpi6S3JdVJOlHSQkmbJL0s\naZ8SMsZK+ja1eUpS28xnkjRa0hep38XZoHIZXS6QNEfSFuDyMmPZS9LTkn5Mfb0p6fDCGIAfU9M5\nqc/6Mv20kXR7etJ1s6QPJA0sMWeD0pOvW9O89M71MziNp0HSOkk3pjAYWTm3SVqb2qySdGVOnaMk\nvZf0+FBSr8z5e0iaIk8kszWdf3WpMQU7CGYWJcp/VvAEIy8DpwO/At1TfT0e6bJTqeNU1y3V9cm1\neR/oDxwJLAHexZOwHAv0AVYD9+d02AhMB44ABgLrgfsybSYCn+HJUA4ALsSjqZ6R02UNcG5qU1dm\nzDOB5Xguh554vJ51QFugDXBY6msw0AVoU6afqXic/QHAgcAVaQ6Pys3H8jSmI9IYvwbapTa9gT+A\nW4CDgSHAJmBkRs40PGDgOUnOScDQEnN+EnAI8AYeV6kQOeF+4BPgGNw1Vg+cV+vvXpQW/G5rrUCU\n1l0KhiG9nws8m963xDAMzLS5ItX1ytSNB5bkdPgJ6JCpuwhoANqnsgXon9P9XuDVnC6jtjHeg1K7\nAZm6PYANwKXpuFNqU/8v/XTH81vsn6t/EXgwNx9DMp93SGMtyJqKBzbM9jEejxqb1ffUMnqUmvPj\nU11dOp4FPFnr71qUypWdCYLqcR0wX9KdLexnUeb9t+l1ca5u3/w5ZrYpczwf//feHdgV2A14XVI2\neNgu+Aohy4fb0O1Q/II+v1BhZhskLcZXCo2lFyBgacbrQ9J1Tq5tVtamnKxDgVdy7d8Bbpa0O540\n6k/caP8b2TkvZHrbF19pPATMSC6s2cBLZjZvG/0F2zFhGIKqYWbvS3oeuAO4NfdxIftb9ipYbnP3\nt2y3qe98XVP2zwptz8TDUZeTBe5eai5NiVi5U2rft4QOW1qgQ3P1+ceck+bNzF6T1BU4DQ/f/Yqk\n6WY2ojJqBtUmNp+DajMG3x/IJzb/Pr3ul6k7uoJye6o4LWc/3F//BZ5joQHoamYrc2VtE+Usw39X\nxxUq0j/znjQtl8PHuJHsUkKn9bm2/TKy2uN7DYXcCstw10+WE3BX0kZ8b2AnfP+g2ZjZD2Y2xcyG\nA5cAwyTt2pI+g9oRK4agqpjZSkmP8s9791fiG7TjJV2P+/THVlD0zsCTkibgqSwnAY+Z2S8Aku4C\n7kp367yF++r7AX+aWaPzFJjZCkkzgUckXYb7+ycCPwPPNKGfzyVNBSZLGoXnTdgb9/mvsuIsgGMl\nfY+7eG7CDV5B1t3AB5LGp7q+wCjcQBfkPAc8LumqJKcO6GZmUxqja5rTBcCn+DwPTjo2NHa8wfZF\nrBiCWjABz1H7N8kVdD5+V8xC/C6aMRWUOQ+/cM0FXsD99KMzn4/DN2WvSe1m43fprG6GrBH4XTyz\n0ms7fHO3qS6gEcBTuOttOX531wAgv4q5HjcAC/DN5EEFg2dmC4Dz0liW4AZxEp7kpcBQ3Gjcl+RM\nxjfMG0sDbvwW4neIdcTdcsEOSiTqCYIdlPT8w1ygs5n9UGN1glZErBiCIAiCIsIwBEEQBEWEKykI\ngiAoIlYMQRAEQRFhGIIgCIIiwjAEQRAERYRhCIIgCIoIwxAEQRAU8RdtwdS0HYioegAAAABJRU5E\nrkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f80b8593160>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HyXEScsPY64v",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        },
        "outputId": "2a107bbc-93fa-4b5f-dfc5-ba188c1fb147"
      },
      "source": [
        "from tensorflow.keras.models import load_model\n",
        "\n",
        "def accuracy_score():\n",
        "  best_model = load_model('best_model.h5')\n",
        "  _, train_acc = best_model.evaluate(X_train, y_train_one_hot, verbose=0)\n",
        "  _, val_acc = best_model.evaluate(X_val, y_val_one_hot, verbose=0)\n",
        "  print(\"Training accuracy: {}%\".format(round(train_acc*100, 3)))\n",
        "  print(\"Validation accuracy: {}%\".format(round(val_acc*100, 3)))\n",
        "\n",
        "accuracy_score()"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
            "Training accuracy: 98.222%\n",
            "Validation accuracy: 92.48%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZhgqBjAlb9T8",
        "colab_type": "text"
      },
      "source": [
        "For the following parameters:\n",
        "```\n",
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 20\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "```\n",
        "\n",
        "We get: \n",
        "```\n",
        "Training accuracy: 98.222%\n",
        "Validation accuracy: 92.48%\n",
        "The training stops after 12 epochs.\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SL4WAmAeh8Ns",
        "colab_type": "text"
      },
      "source": [
        "He initialization:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7vrzpIZ-Zv-r",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "33805066-7038-4456-f973-c4dc1d600d6c"
      },
      "source": [
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 30\n",
        "choice_init = 'he_uniform'\n",
        "batch_norm = True\n",
        "\n",
        "train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/30\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.4649 - accuracy: 0.8311\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.75440, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.4643 - accuracy: 0.8312 - val_loss: 1.2451 - val_accuracy: 0.7544\n",
            "Epoch 2/30\n",
            "1092/1100 [============================>.] - ETA: 0s - loss: 0.3114 - accuracy: 0.8881\n",
            "Epoch 00002: val_accuracy improved from 0.75440 to 0.89340, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.3109 - accuracy: 0.8884 - val_loss: 0.3183 - val_accuracy: 0.8934\n",
            "Epoch 3/30\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.2502 - accuracy: 0.9098\n",
            "Epoch 00003: val_accuracy did not improve from 0.89340\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2502 - accuracy: 0.9098 - val_loss: 5.8260 - val_accuracy: 0.6752\n",
            "Epoch 4/30\n",
            "1097/1100 [============================>.] - ETA: 0s - loss: 0.2187 - accuracy: 0.9210\n",
            "Epoch 00004: val_accuracy improved from 0.89340 to 0.91120, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2187 - accuracy: 0.9210 - val_loss: 0.2743 - val_accuracy: 0.9112\n",
            "Epoch 5/30\n",
            "1092/1100 [============================>.] - ETA: 0s - loss: 0.1933 - accuracy: 0.9305\n",
            "Epoch 00005: val_accuracy improved from 0.91120 to 0.91820, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1933 - accuracy: 0.9306 - val_loss: 0.2605 - val_accuracy: 0.9182\n",
            "Epoch 6/30\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.1757 - accuracy: 0.9368\n",
            "Epoch 00006: val_accuracy did not improve from 0.91820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1756 - accuracy: 0.9368 - val_loss: 0.2506 - val_accuracy: 0.9138\n",
            "Epoch 7/30\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.1438 - accuracy: 0.9482\n",
            "Epoch 00007: val_accuracy did not improve from 0.91820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1438 - accuracy: 0.9482 - val_loss: 0.2716 - val_accuracy: 0.9178\n",
            "Epoch 8/30\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.1240 - accuracy: 0.9549\n",
            "Epoch 00008: val_accuracy did not improve from 0.91820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1240 - accuracy: 0.9548 - val_loss: 0.2648 - val_accuracy: 0.9168\n",
            "Epoch 9/30\n",
            "1095/1100 [============================>.] - ETA: 0s - loss: 0.1214 - accuracy: 0.9570\n",
            "Epoch 00009: val_accuracy did not improve from 0.91820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1213 - accuracy: 0.9570 - val_loss: 0.3395 - val_accuracy: 0.9180\n",
            "Epoch 10/30\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.0882 - accuracy: 0.9686\n",
            "Epoch 00010: val_accuracy improved from 0.91820 to 0.91980, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0882 - accuracy: 0.9686 - val_loss: 0.3332 - val_accuracy: 0.9198\n",
            "Epoch 11/30\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.0729 - accuracy: 0.9734\n",
            "Epoch 00011: val_accuracy improved from 0.91980 to 0.92300, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0731 - accuracy: 0.9733 - val_loss: 0.4918 - val_accuracy: 0.9230\n",
            "Epoch 00011: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEeCAYAAACJ266bAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMoQmfKWUyhCa8JVSKkNowldKqQzx/wE8Q3MtjMCJPgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f80b6166cf8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oi3BbdRzcjS3",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        },
        "outputId": "28d2f9c2-37f7-4fc6-d0b5-4152002ef943"
      },
      "source": [
        "accuracy_score()"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
            "Training accuracy: 97.9%\n",
            "Validation accuracy: 92.3%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VJtdli2ddGWe",
        "colab_type": "text"
      },
      "source": [
        "For the following parameters:\n",
        "```\n",
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 30\n",
        "choice_init = 'he_uniform'\n",
        "batch_norm = True\n",
        "```\n",
        "\n",
        "We get: \n",
        "```\n",
        "Training accuracy: 97.9%\n",
        "Validation accuracy: 92.3%\n",
        "The training stops after 11 epochs.\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CJh_TW5EhplB",
        "colab_type": "text"
      },
      "source": [
        "**Conclusion:** Xavier initialization performs marginally better than He initialization as demonstrated above."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rPPMCHj7d-Hh",
        "colab_type": "text"
      },
      "source": [
        "Let us check what happens if we do not use BatchNormalization before the final layer."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Lx3VPeCJdApp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 901
        },
        "outputId": "74888b75-44d9-4060-f723-14a1db97c66f"
      },
      "source": [
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = False\n",
        "\n",
        "train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.4944 - accuracy: 0.8381\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.88680, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.4945 - accuracy: 0.8382 - val_loss: 0.3074 - val_accuracy: 0.8868\n",
            "Epoch 2/50\n",
            "1098/1100 [============================>.] - ETA: 0s - loss: 0.2885 - accuracy: 0.8948\n",
            "Epoch 00002: val_accuracy improved from 0.88680 to 0.89320, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2884 - accuracy: 0.8948 - val_loss: 0.2921 - val_accuracy: 0.8932\n",
            "Epoch 3/50\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.2525 - accuracy: 0.9071\n",
            "Epoch 00003: val_accuracy improved from 0.89320 to 0.90440, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2525 - accuracy: 0.9071 - val_loss: 0.2516 - val_accuracy: 0.9044\n",
            "Epoch 4/50\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.2308 - accuracy: 0.9148\n",
            "Epoch 00004: val_accuracy improved from 0.90440 to 0.90660, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2308 - accuracy: 0.9148 - val_loss: 0.2690 - val_accuracy: 0.9066\n",
            "Epoch 5/50\n",
            "1092/1100 [============================>.] - ETA: 0s - loss: 0.2126 - accuracy: 0.9219\n",
            "Epoch 00005: val_accuracy improved from 0.90660 to 0.90880, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2128 - accuracy: 0.9217 - val_loss: 0.2600 - val_accuracy: 0.9088\n",
            "Epoch 6/50\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.1917 - accuracy: 0.9286\n",
            "Epoch 00006: val_accuracy did not improve from 0.90880\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1917 - accuracy: 0.9286 - val_loss: 0.2539 - val_accuracy: 0.9086\n",
            "Epoch 7/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.1776 - accuracy: 0.9326\n",
            "Epoch 00007: val_accuracy did not improve from 0.90880\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1776 - accuracy: 0.9326 - val_loss: 0.3004 - val_accuracy: 0.9072\n",
            "Epoch 8/50\n",
            "1098/1100 [============================>.] - ETA: 0s - loss: 0.1645 - accuracy: 0.9411\n",
            "Epoch 00008: val_accuracy did not improve from 0.90880\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1645 - accuracy: 0.9411 - val_loss: 0.2751 - val_accuracy: 0.9080\n",
            "Epoch 00008: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEeCAYAAACZlyICAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nFQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f810c071e10>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 28
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eSMZWhOJekyj",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        },
        "outputId": "8fdebf87-8ba6-4329-86f6-d46aaede474d"
      },
      "source": [
        "accuracy_score()"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
            "Training accuracy: 93.04%\n",
            "Validation accuracy: 90.88%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ReDrucrregSc",
        "colab_type": "text"
      },
      "source": [
        "For the following parameters:\n",
        "```\n",
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = False\n",
        "```\n",
        "\n",
        "We get: \n",
        "```\n",
        "Training accuracy: 93.04%\n",
        "Validation accuracy: 90.88%\n",
        "The training stops after 8 epochs.\n",
        "```\n",
        "\n",
        "**Conclusion** : Without BatchNormalization, overfitting starts very early, in epoch 8 itself. That is why the training stopped at epoch 8. The training error will keep decreasing, but the validation error will remain stationary for a few epochs and then start increasing due to overfitting.\n",
        "\n",
        "In order to see this, let us use the train_model() function without the effect of early stopping."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vD0xVdOleLE8",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "cd324569-1a53-421c-b2b6-80d3ff1f3ef7"
      },
      "source": [
        "def train_model_no_earlystop(X_train, y_train, X_val, y_val, learning_rate=0.001, batch_size=50, \n",
        "                num_epochs=20, choice_init='glorot_uniform', batch_norm=True):\n",
        "    # Create the model\n",
        "    fashion_mnist = create_model(choice_init, batch_norm)\n",
        "    \n",
        "    # Compile the model\n",
        "    adam = Adam(learning_rate=learning_rate)\n",
        "    fashion_mnist.compile(optimizer=adam, loss='categorical_crossentropy', metrics=['accuracy'])\n",
        "    \n",
        "    # Fit training data\n",
        "    history = fashion_mnist.fit(X_train, y_train, batch_size=batch_size, epochs=num_epochs, \n",
        "                                verbose=1, validation_data=(X_val, y_val))\n",
        "    \n",
        "    # plot graphs\n",
        "    plot_loss_curves(history)\n",
        "    return history\n",
        "\n",
        "lr = 0.001\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = False\n",
        "\n",
        "history = train_model_no_earlystop(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.4800 - accuracy: 0.8394 - val_loss: 0.3108 - val_accuracy: 0.8856\n",
            "Epoch 2/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.2841 - accuracy: 0.8953 - val_loss: 0.2827 - val_accuracy: 0.9008\n",
            "Epoch 3/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.2496 - accuracy: 0.9097 - val_loss: 0.2643 - val_accuracy: 0.9078\n",
            "Epoch 4/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2252 - accuracy: 0.9159 - val_loss: 0.2505 - val_accuracy: 0.9084\n",
            "Epoch 5/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.2090 - accuracy: 0.9226 - val_loss: 0.2627 - val_accuracy: 0.9096\n",
            "Epoch 6/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1915 - accuracy: 0.9286 - val_loss: 0.2470 - val_accuracy: 0.9122\n",
            "Epoch 7/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1824 - accuracy: 0.9321 - val_loss: 0.2700 - val_accuracy: 0.9136\n",
            "Epoch 8/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1664 - accuracy: 0.9383 - val_loss: 0.3491 - val_accuracy: 0.9052\n",
            "Epoch 9/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1569 - accuracy: 0.9429 - val_loss: 0.2945 - val_accuracy: 0.9096\n",
            "Epoch 10/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1521 - accuracy: 0.9454 - val_loss: 0.2986 - val_accuracy: 0.9176\n",
            "Epoch 11/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1332 - accuracy: 0.9511 - val_loss: 0.2805 - val_accuracy: 0.9164\n",
            "Epoch 12/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1238 - accuracy: 0.9543 - val_loss: 0.2960 - val_accuracy: 0.9160\n",
            "Epoch 13/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1198 - accuracy: 0.9570 - val_loss: 0.2906 - val_accuracy: 0.9156\n",
            "Epoch 14/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1083 - accuracy: 0.9604 - val_loss: 0.3194 - val_accuracy: 0.9190\n",
            "Epoch 15/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1040 - accuracy: 0.9626 - val_loss: 0.3473 - val_accuracy: 0.9176\n",
            "Epoch 16/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1034 - accuracy: 0.9638 - val_loss: 0.3780 - val_accuracy: 0.9144\n",
            "Epoch 17/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0973 - accuracy: 0.9651 - val_loss: 0.4013 - val_accuracy: 0.9154\n",
            "Epoch 18/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0904 - accuracy: 0.9681 - val_loss: 0.3835 - val_accuracy: 0.9190\n",
            "Epoch 19/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0887 - accuracy: 0.9689 - val_loss: 0.3848 - val_accuracy: 0.9170\n",
            "Epoch 20/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0892 - accuracy: 0.9689 - val_loss: 0.3482 - val_accuracy: 0.9164\n",
            "Epoch 21/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0798 - accuracy: 0.9723 - val_loss: 0.4264 - val_accuracy: 0.9198\n",
            "Epoch 22/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0822 - accuracy: 0.9720 - val_loss: 0.4172 - val_accuracy: 0.9130\n",
            "Epoch 23/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0671 - accuracy: 0.9771 - val_loss: 0.4280 - val_accuracy: 0.9206\n",
            "Epoch 24/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0674 - accuracy: 0.9780 - val_loss: 0.4215 - val_accuracy: 0.9146\n",
            "Epoch 25/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0940 - accuracy: 0.9698 - val_loss: 0.4114 - val_accuracy: 0.9100\n",
            "Epoch 26/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0781 - accuracy: 0.9743 - val_loss: 0.3927 - val_accuracy: 0.9086\n",
            "Epoch 27/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0618 - accuracy: 0.9794 - val_loss: 0.3761 - val_accuracy: 0.9160\n",
            "Epoch 28/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0646 - accuracy: 0.9785 - val_loss: 0.4892 - val_accuracy: 0.9102\n",
            "Epoch 29/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0637 - accuracy: 0.9789 - val_loss: 0.4681 - val_accuracy: 0.9106\n",
            "Epoch 30/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0766 - accuracy: 0.9767 - val_loss: 0.5285 - val_accuracy: 0.9162\n",
            "Epoch 31/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0495 - accuracy: 0.9842 - val_loss: 0.4679 - val_accuracy: 0.9162\n",
            "Epoch 32/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0670 - accuracy: 0.9803 - val_loss: 0.4754 - val_accuracy: 0.9144\n",
            "Epoch 33/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0490 - accuracy: 0.9844 - val_loss: 0.5227 - val_accuracy: 0.9182\n",
            "Epoch 34/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0599 - accuracy: 0.9820 - val_loss: 0.5582 - val_accuracy: 0.9094\n",
            "Epoch 35/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0712 - accuracy: 0.9795 - val_loss: 0.4341 - val_accuracy: 0.9024\n",
            "Epoch 36/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0716 - accuracy: 0.9787 - val_loss: 0.5665 - val_accuracy: 0.9118\n",
            "Epoch 37/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0620 - accuracy: 0.9816 - val_loss: 0.5620 - val_accuracy: 0.9120\n",
            "Epoch 38/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0438 - accuracy: 0.9871 - val_loss: 0.5800 - val_accuracy: 0.9130\n",
            "Epoch 39/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0786 - accuracy: 0.9767 - val_loss: 0.6208 - val_accuracy: 0.9158\n",
            "Epoch 40/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0395 - accuracy: 0.9879 - val_loss: 0.6065 - val_accuracy: 0.9162\n",
            "Epoch 41/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0773 - accuracy: 0.9798 - val_loss: 0.6469 - val_accuracy: 0.9120\n",
            "Epoch 42/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0506 - accuracy: 0.9853 - val_loss: 0.5100 - val_accuracy: 0.9142\n",
            "Epoch 43/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0380 - accuracy: 0.9880 - val_loss: 0.6770 - val_accuracy: 0.9092\n",
            "Epoch 44/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.1256 - accuracy: 0.9734 - val_loss: 0.5491 - val_accuracy: 0.9160\n",
            "Epoch 45/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0664 - accuracy: 0.9811 - val_loss: 0.4704 - val_accuracy: 0.9106\n",
            "Epoch 46/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0642 - accuracy: 0.9823 - val_loss: 0.6146 - val_accuracy: 0.9164\n",
            "Epoch 47/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0322 - accuracy: 0.9908 - val_loss: 0.8066 - val_accuracy: 0.9220\n",
            "Epoch 48/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0369 - accuracy: 0.9893 - val_loss: 0.6469 - val_accuracy: 0.9118\n",
            "Epoch 49/50\n",
            "1100/1100 [==============================] - 6s 6ms/step - loss: 0.0530 - accuracy: 0.9855 - val_loss: 0.5660 - val_accuracy: 0.9162\n",
            "Epoch 50/50\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0952 - accuracy: 0.9771 - val_loss: 0.6987 - val_accuracy: 0.9082\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEeCAYAAACOtbLLAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uqtRCzQ2hXD5",
        "colab_type": "text"
      },
      "source": [
        "The above graph shows that overfitting started somewhere around epoch 7 or 8. That is why, when we used early stopping callback, the training stopped at epoch number 8."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W8PVlLJxiwcS",
        "colab_type": "text"
      },
      "source": [
        "Let us now see the effect of changing the batch size."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OVChNO1wf-c0",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 833
        },
        "outputId": "9652d528-1a70-425f-ef85-5ea3712b5c35"
      },
      "source": [
        "lr = 0.001\n",
        "batch_size = 10\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "\n",
        "train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "5497/5500 [============================>.] - ETA: 0s - loss: 0.6392 - accuracy: 0.7720\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.76880, saving model to best_model.h5\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.6391 - accuracy: 0.7720 - val_loss: 2.1075 - val_accuracy: 0.7688\n",
            "Epoch 2/50\n",
            "5487/5500 [============================>.] - ETA: 0s - loss: 0.3815 - accuracy: 0.8672\n",
            "Epoch 00002: val_accuracy improved from 0.76880 to 0.89260, saving model to best_model.h5\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.3813 - accuracy: 0.8673 - val_loss: 0.3580 - val_accuracy: 0.8926\n",
            "Epoch 3/50\n",
            "5489/5500 [============================>.] - ETA: 0s - loss: 0.3300 - accuracy: 0.8842\n",
            "Epoch 00003: val_accuracy did not improve from 0.89260\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.3300 - accuracy: 0.8842 - val_loss: 0.5925 - val_accuracy: 0.8882\n",
            "Epoch 4/50\n",
            "5490/5500 [============================>.] - ETA: 0s - loss: 0.3038 - accuracy: 0.8937\n",
            "Epoch 00004: val_accuracy did not improve from 0.89260\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.3039 - accuracy: 0.8936 - val_loss: 0.7524 - val_accuracy: 0.8824\n",
            "Epoch 5/50\n",
            "5493/5500 [============================>.] - ETA: 0s - loss: 0.2619 - accuracy: 0.9081\n",
            "Epoch 00005: val_accuracy improved from 0.89260 to 0.89940, saving model to best_model.h5\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.2619 - accuracy: 0.9081 - val_loss: 3.7106 - val_accuracy: 0.8994\n",
            "Epoch 6/50\n",
            "5487/5500 [============================>.] - ETA: 0s - loss: 0.2355 - accuracy: 0.9177\n",
            "Epoch 00006: val_accuracy did not improve from 0.89940\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.2357 - accuracy: 0.9177 - val_loss: 37.9547 - val_accuracy: 0.8834\n",
            "Epoch 7/50\n",
            "5487/5500 [============================>.] - ETA: 0s - loss: 0.2150 - accuracy: 0.9248\n",
            "Epoch 00007: val_accuracy improved from 0.89940 to 0.91120, saving model to best_model.h5\n",
            "5500/5500 [==============================] - 21s 4ms/step - loss: 0.2149 - accuracy: 0.9248 - val_loss: 29.7894 - val_accuracy: 0.9112\n",
            "Epoch 00007: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEeCAYAAACHXhKxAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f812606ac18>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 32
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lQ67sHRolbw3",
        "colab_type": "text"
      },
      "source": [
        "Having a smaller batch size makes the validation curve very jumpy."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9pVcqMN7i5tI",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        },
        "outputId": "114e5205-1a99-4539-92b8-117c8ccf2c6f"
      },
      "source": [
        "accuracy_score()"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
            "Training accuracy: 93.567%\n",
            "Validation accuracy: 91.12%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8zb3MCpXmxr_",
        "colab_type": "text"
      },
      "source": [
        "Change in the learning rate"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gXflTFCXm1C8",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 884
        },
        "outputId": "e1db1192-ee32-4725-a0b9-d00ffd7b2b83"
      },
      "source": [
        "lr = 0.01\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "\n",
        "history = train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.6534 - accuracy: 0.7636\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.74260, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.6533 - accuracy: 0.7637 - val_loss: 0.7866 - val_accuracy: 0.7426\n",
            "Epoch 2/50\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.3684 - accuracy: 0.8667\n",
            "Epoch 00002: val_accuracy improved from 0.74260 to 0.87200, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.3684 - accuracy: 0.8667 - val_loss: 0.3619 - val_accuracy: 0.8720\n",
            "Epoch 3/50\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.3026 - accuracy: 0.8913\n",
            "Epoch 00003: val_accuracy improved from 0.87200 to 0.89780, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.3025 - accuracy: 0.8914 - val_loss: 0.2869 - val_accuracy: 0.8978\n",
            "Epoch 4/50\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.2690 - accuracy: 0.9023\n",
            "Epoch 00004: val_accuracy improved from 0.89780 to 0.90180, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2689 - accuracy: 0.9024 - val_loss: 102.2527 - val_accuracy: 0.9018\n",
            "Epoch 5/50\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.2424 - accuracy: 0.9127\n",
            "Epoch 00005: val_accuracy did not improve from 0.90180\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2423 - accuracy: 0.9128 - val_loss: 156.5919 - val_accuracy: 0.8912\n",
            "Epoch 6/50\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.2243 - accuracy: 0.9190\n",
            "Epoch 00006: val_accuracy improved from 0.90180 to 0.90540, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2243 - accuracy: 0.9190 - val_loss: 83.8983 - val_accuracy: 0.9054\n",
            "Epoch 7/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.2117 - accuracy: 0.9223\n",
            "Epoch 00007: val_accuracy did not improve from 0.90540\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2118 - accuracy: 0.9223 - val_loss: 144.7701 - val_accuracy: 0.8978\n",
            "Epoch 8/50\n",
            "1095/1100 [============================>.] - ETA: 0s - loss: 0.2007 - accuracy: 0.9277\n",
            "Epoch 00008: val_accuracy improved from 0.90540 to 0.91600, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2006 - accuracy: 0.9276 - val_loss: 188.1213 - val_accuracy: 0.9160\n",
            "Epoch 00008: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYkAAAEeCAYAAAB/vulGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOzdd3wUdf7H8dcnoUmvRgTugj8FTjqJ\n6FmJCBZQTgUFUQGl2vEEUTnFguKJYDk1VGkKJ3hwiKgognjiKUWkCBYQDiw0EYjUJJ/fHzPBZdlN\ndpPdnd3k8+SxD7LfmfnOe2dn57vzndkZUVWMMcaYQJK8DmCMMSZ+WSNhjDEmKGskjDHGBGWNhDHG\nmKCskTDGGBOUNRLGGGOCskYiTohIqoioiAwrQh2TRMTOaU4Q7vs9ycP5B1znwsklIsPc8VOjkK+n\nW3ebSNdtQmeNRBDuyhnqI9XrvPHEXSbzvM5RHIjITHd5tshnHBGR70XkVxE5KZb5ikpE2rgNTVWv\nswTi05D+w+ssXinldYA4dpPf8wuAvsBY4GO/YTsjML8twElAdhHq6AP0j0AWEz8mAJ2BXsDdQcbJ\nAFKBMap6MALzPAnIiUA9oWgDPAJMAn71GzYVmAEciVEWE4A1EkGo6jTf5yJSCqeR+NR/mD8RqaSq\n+8OcnwKHwg56fB1HgaNFqcPEnQXAVqC7iAxS1UAbzF7u/xMiMUNVLdJ6GCmqmkPsGisThHU3FZGI\nbBaRxSLSUkTeE5G9wGp3WCUReUJEPhORXSJyWES+E5ERIlLer54T+od9y0Sko4gsE5FDIvKTiDzj\nNly+dZxwTCKvTESqiMgrIrLDreMTETk7wOupISITRWS3iGSJyIfua1ssIpsjuOjyXt9UEdnuLpuN\nIvJkgGVTXURGu8MPudlWiMggv/FuFpHP3W6X30Rkk4i8JiK1QshyvYjMFZH/uVl2icgcEWkWYNy8\n97yRiLwtIvtFZK+IzBKRUwKM31hE3nUz/eJmOjmUZaSquTjfsmsAVwWouzJwLbBWVZeFs87lsyxO\nOCYhIkki8oDbrXVIRNaKSPcg0zcSkZdFZJ27bA6471dvv/Em4exFAHwvv3ffDnOHBzwmISI1ReQl\nEdkqIkfc/18SkRp+4+VNf7GI3OeuP4dF5BsR6RHKsgiHiDQTkdnu+nlIRL4SkcEikuw3Xj33M7bF\nzbNDRJb6ZnKX9z0istpdhvtE5GsRmSAipSOdPT+2JxEZfwA+BGYCbwIV3fI6QG+37HWcrqSLgMFA\nS+DSEOu/ArgNyAQmAp2A+4A9wJMh1vEeTrfYYzgbnHuBt0Wkft5ej4iUBT4AWuBsmD4Hmrllv4Q4\nn5CIyB/d+qsALwPf4nQ9PACcJyJtVTWv620mcCHO61+N0x3yJ3f8Z9z6bgIm43QFPgwcBOrhLLuT\nKbhL8A5gN0534s/A/+HsOX4iIq1U9Vu/8esAi4HZwCCgOdAPqAy093md9d1MZYF/4OwVXAm8W0Ae\nX68CQ3H2GGb5DeuKszzy9iIitc75G4XT3bUEGI2zTF8CNgUYtw3O+zUP+B6oAHQBxolILVV9yh1v\nDM7yuhoYCOxyy1cHCyEiVYClwOk4n4WV7usaAFwsIq0D7MU/ibOMxgCH3XEnich3qvpJiK8/XyKS\nDnyEsyf/Es46dCXwNM660d0drxTwPs779DLwDc5noBlOl/Zkt8qHcD6rb+Gs9zlAfZwvCmWJZY+B\nqtojhAfQE1Cgp1/5Zre8d4BpygClA5Q/7k7T2qcs1S0bFqDsNyDVp1yAtcBPfvVOwu258i8DXvYr\n7+KW9/Mpu80te8hv3LzyzSEuKwXmFTDOa+54V/iVP+OW3+o+rxIof4D6/gXsA0oV8v2tEKDsTzgb\nFf9ll/eeX+dX/pJb3tCn7HW3LMPv/Zvtlk8KMd9CnA1+bb/yT92MNSOxzvm8f5N8njcEct0MyT7l\nrdxy9Vs/Ay3LJJxGda9vPmCY//QBPnNtfMqGu2W3+Y17u1v+eIDpvwDK+JTXcZfZ9BCWe94y+kcB\n433ivj/N/N7nN9zp27plzdzngwuobyXwVWHW5Ug/rLspMn7B+bZ3HFU9os5xAkSklIhUE5GaON/M\nAU7o7glijqpu9qlXgUXAKSJSMehUxxvt9/xD9/8zfMquxPnG8rzfuONxPtwRISJJON+IvlDV+X6D\nn8LZ8FztPj+I84E+W/I/i2wvUB7oICISbiZV/c3NJiJS2X2fdgJfE/h9+lFV3/ArO26Zuq/zSmC5\nqi7ymZcCfw8z4gQgGbg5r0BEGgHnAHNVdZdbd6TWOV+dcDZ4o9Q5TpD3OlbifCs+Tt6ydDOUc7uB\nquMcX6kMNCpEhjxX47wvY/3Kx7jlV58whdPIHzuWo6o/4HyDPyPAuGFzuw7PxXkfju0Fue/zcJ/c\n8PvnKKOALse9QB0ROT8SGYvCGonI2Oj74fElIreJyGqcDd0vOCvyYndwtRDrD7RLv9v9v0aAYQXW\noaqBpq+Ps/HL8hv3CE63QaTUwumSW+c/QFV/AX4CTvOZ9z1AE5x+63Ui8qKItPWb9EmcM8TmADtF\n5E0R6S0ilUIJJM5xl3nAfpwP6E730ZTA71Mo78nJ7uvcEGDcr0LJ5eNfOGf/9PIpu8X9f6LviBFa\n53yd5v4f0usQkYoiMlJE/ofTyO9yM+RtMAuTIU994Gv9vSsSAPf5Nz5ZfQV7r0L97ISSCQKsz8B6\nnC89eevzFpzl0B74yT1W83cROctvugdxTmT5WER+cI9j3SAiZSKUOWTWSETGgUCFInIvThfETzj9\n1R2Adji7wRD68s/vDI+QvjUHa8RCnd5LqpqJs9vfB2c3vDPwgYjM8BnnW+BMnGU8GfgjMA7YICL/\nl1/9IvIHnL72ljjdMlfjfIjb4XzwA71PRX5PwqHOGUevAw1F5Fz3YOhNwDac403OjCO3zhXF6zjH\nvObj9MVf5mbI25uN9XYnrtZ9VR2KsxdzD7AR5xjS5yLytM84n+IcF+uM0zXZAqeLdpWIVI9lXjtw\nHV034fRfX67OWSoAiMhlniXK32bgEhGp6Ls34Z5NUZ8Tz2MvrJ0439gb+w8QkWpAbWCVb7mq/oTT\n7TXe3UBOBbqJyLOquswd5zDOhmm+W9cVwNs4G6zb88lzNc43/qt8u4XcOmrgfCMvjJ1AFoG7V84s\nRH0TcI4P9cLpvjkFGO67bhGddS7vm3gjnI2ar+Nehzg/iusITFXV/n7DLglQd7hXCNiE01CW8t2b\ncA8INyDwXkO05e1ln7A+4yyzJE7ck98EvAi8KCLlcBr6we76vMMdJwvnBIQ3wdlDxPkCcCvuCRux\nYHsS0ZWD8yE49o3FXZmHeJYof2/h9Hv7/2irD84B5IhwN15vAS0DbLyG4KyXswFEpLz4nbrp7hXl\n9f1Wd8erGWBWK33HyUfeN83jvlmKSB+cDXGhuDnnAekikuFTr+CcbRRufStxGs/r+f1A7US/0aKx\nzs1167zX93ROEWkF+G/4gy3L2jjfmP3lfRkJ9dvxHJzuSv+6+rjls0OsJ2LcjfpS4EoRaZJX7r7P\nD7hP89bnKv6nsLp7ievdp9Xc8YqyPkeU7UlE1yycA7HviMi/cA7a3UD8/uBtPE4XxRMicjq/nwJ7\nHfAd4a0vp4vI0CDDRuP0ubYD5ojIy279F+JsAJfw+6mADYCPRGQ2zhlde3DOOhqA8w0u79fvC0Tk\nV/f5VqAqv5/dMrWArO/gdBlOFefyC3uA83BOn91I0T4nQ4HLgXki8iJO99CVOBu0wpiA8w30MmCx\n+43UV8TXOVXdICIv4Zwm/KGIvIlzvOUO4Eucbrq8cfeLyALgRhE5CCzD6frrh/N++R8H+K/7/9Mi\n8hpOP/xaVV0bJM7fcc7Me8ltpL5w538rzkkG4Z4QEKr0IOtztqqOwPli9RHOMYS8U2A74pxy/Lqq\nLnTHzwDGusvwa5xGMg2n0ftMVb92x1svIv8FPgN+xNm77ovz6/Nj3awx4fXpVYnyIP9TYBcHmSYZ\n55vEdzhdFltwVuI/Efx013zLfIYN48RTDycR5BTYIPlOOAUTZ+M1CeeA5284Z+y0AJYT4il5br35\nPU5xx6uPswHfgbPyb8I5AF3ep64aOI3KKpzuroPu8nwOn9NBcb5Jvo/z4TyC0yc/H59TTwvIfCHw\nH5xusF9xuqma4Bzw3RzKe47z+4BA60hTnDN7fnOX62s4G9mQT4H1qauauwwUuCnS61w+60USzrn7\nW9x61+Icbwi0HtbE+cLxI85Gf437/vTE75RWd/zB7nt/1DdPPuPXwvmNwTZ3mm043TA1g3xm2wRY\nTie8r0GWd94yCvY45DNuc5w9nV/cZbTefW2+pw3Xx/ndw3qcU7Z/c/9+DKjiM94QnC9LO9y6tuL8\nXqhVOOtLJB7iBjImKLeLYRfON514PZ5ijIkCOyZhjiOBryLaH6f75oRz4o0xxZvtSZjjiMg0oBzO\ngbjDwJ9x+rQ34uzqhnXhQmNMYrNGwhxHRG7GOXOmAc5podtx+vb/pqrbvcxmjIk9aySMMcYEVaxO\nga1Zs6ampqYWevrffvuNChUqRC5QFCVSVkisvJY1ehIpbyJlhaLlXbFixS5VDXxadqxPp4rmIy0t\nTYti0aJFRZo+lhIpq2pi5bWs0ZNIeRMpq2rR8uJchDLgdtXObjLGGBOUNRLGGGOCskbCGGNMUMXq\nwHUgR48eZdu2bRw6VPC93atUqcL69esLHC8eJFJWCD9vuXLlqFu3LqVLx/R2vsYYP8W+kdi2bRuV\nKlUiNTWVgm5Ytn//fipVCukeNZ5LpKwQXl5VZffu3Wzbto369esXPIExJmqKfXfToUOHqFGjRoEN\nhIkfIkKNGjVC2vszxkRXsW8kAGsgEpC9Z8bEhxLRSBhjTHH2Cq+wouqKqNRtjUSU7d69mxYtWtCi\nRQtOOeUU6tSpc+z5kSNH8p12+fLl3HXXXQXO49xzz41I1sWLF9OxY8eI1GWMiY2JTOQ2bmN+7flR\nqb/YH7j2Wo0aNVi1yrld87Bhw6hYsSL33XffseHZ2dmUKhX4bUhPTyc9Pb3AeSxdujQyYY0xCWUG\nM+hNb9rTnr9u+CukRH4etifhgZ49e9K/f3/OPvtsBg8ezOeff86f//xnWrZsybnnnsvXXzt3MPT9\nZj9s2DBuueUW2rRpw2mnncYrr7xyrL6KFSseG79NmzZ07tyZRo0a0b1797y7XDF//nwaNWpEWloa\nd911V1h7DNOnT6dp06Y0adKE+++/H4CcnBx69uxJkyZNaNq0KaNHjwbghRde4Mwzz6RZs2Z07dq1\n6AvLGBPQXOZyEzdxPuczm9mU0TJRmU/J2pO45x5wv9UHclJODiQnBx0eUIsW8NxzYUfZtm0bS5cu\nJTk5mX379vHxxx9TqlQpPvjgAx588EHefPPNE6bZsGEDixYtYv/+/TRo0ICBAwee8DuCL774gnXr\n1nHqqady3nnn8cknn5Cenk6/fv1YsmQJ9evXp1u3biHn/PHHH7n//vtZsWIF1apVo3379syZM4d6\n9erxww8/sHatcyviX3/9FYARI0bw/fffU7Zs2WNlxpjI+oAP6EIXWtGKecyjPOWjNi/bk/BIly5d\nSHYbpL1799KlSxeaNGnCwIEDWbduXcBpOnToQNmyZalZsya1atVi+/YTb+/QunVr6tatS1JSEi1a\ntGDz5s1s2LCB00477dhvDsJpJJYtW0abNm2oVasWpUqVonv37ixZsoTTTjuNTZs2ceedd/Luu+9S\nuXJlAJo1a0b37t2ZNm1a0G40Y0zh/Yf/0IlONKIR7/AOlakc1fmVrE9xAd/4D8bwB2q+l/T929/+\nRkZGBrNnz2bz5s20adMm4DRly5Y99ndycjLZ2dmFGicSqlWrxpdffsl7771HZmYmb7zxBhMnTuTt\nt99myZIlvPXWWwwfPpw1a9ZYY2FMhCxnOR3oQD3qsYAFVKd61OdpexJxYO/evdSpUweASZMmRbz+\nhg0bsmnTJjZv3gzAP//5z5Cnbd26NR999BG7du0iJyeH6dOnc9FFF7Fr1y5yc3O59tpreeKJJ1i5\nciW5ubls3bqVjIwMnn76afbu3UtWVlbEX48xJdEa1nApl1Kd6nzAB6RE4yh1APYVLw4MHjyYHj16\n8MQTT9ChQ4eI13/SSSfx8ssvc9lll1GhQgXOOuusoOMuXLiQunXrHns+c+ZMRowYQUZGBqpKhw4d\n6NSpE19++SW9evUiNzcXgKeeeoqcnBxuvPFG9u7di6py1113UbVq1Yi/HmNKmm/4hna0oxzlWMhC\n6lK34IkiJdiNJhLxEeimQ1999VXIN97Yt29fyON6Ldys+/fvV1XV3NxcHTBggI4aNSoasYIqzLIN\n572LpES62UwiZVVNrLzxkvV7/V7ral2tpbV0va4POp7ddMgUybhx42jRogWNGzdm79699OvXz+tI\nxpgC/MiPXMIlZJHFAhbQiEYxz2DdTSXEwIEDGThwoNcxjDEh2slO2tGO7Wznfd6nBS08yRGzPQkR\nmSgiO0RkrU/ZP0VklfvYLCKr3PJUETnoMywzVjmNMcZrv/Irl3Ipm9jEW7zFOZzjWZZY7klMAv4B\nTMkrUNXr8/4WkWeBvT7jb1RVb5pOY4zxSBZZXMEVrGUtc5lLG9p4midmjYSqLhGR1EDDxLku9HXA\nxbHKY4wx8eYgB7mKq/icz5nJTC7jMq8jIepe2ycmM3MaiXmq2sSv/EJglKqm+4y3DvgG2AcMVdWP\ng9TZF+gLkJKSkjZjxozjhlepUoXTTz89pHw5OTnHfgUd7xIpKxQu73fffcfevXsLHjHCsrKyjl0P\nK94lUlZIrLyxznpUjvJwk4f5rPpnPLDhAdptbxfW9EXJm5GRsSJv+3uCYKc9ReMBpAJrA5S/AvzV\n53lZoIb7dxqwFahcUP3xeApsmzZt9N133z2ubPTo0dq/f/+g01x00UW6bNkyVVW9/PLLdc+ePSeM\nM2TIEH3mmWfynffs2bN13bp1x57/7W9/0/fffz+c+AEtWrRIO3ToENY0dgpsdCRSVtXEyhvLrEf1\nqHbRLoqiY3RMoeootqfAikgp4Brg2M+AVfWwqu52/14BbAQaeJOwaLp164b/3s2MGTNCvn7S/Pnz\nC/2DtDlz5vDVV18de/7YY49xySWXFKouY0x05JLLrdzKTGYyilH0dTpG4obnjQRwCbBBVbflFYhI\nLRFJdv8+DTgD2ORRviLp3Lkzb7/99rEbDG3evJkff/yRCy64gAEDBpCenk7jxo155JFHAk6fmprK\nrl27ABg+fDgNGjTg/PPP59tvvz02zrhx4zjrrLNo3rw51157LQcOHGDp0qXMnTuXQYMG0aJFCzZu\n3EjPnj2ZNWsW4PyyumXLljRt2pRbbrmFw4cPH5vfI488QqtWrWjatCkbNmwI+bXmd0nxs88+2y4p\nbowfRbmDO5jCFB7jMQYSf6epx+zAtYhMB9oANUVkG/CIqk4AugLT/Ua/EHhMRI4CuUB/Vf2lqBnu\n4R5WEfxS4Tkn5ZBMeP3mLWjBcwS/cGD16tVp3bo177zzDp06dWLGjBlcd911iAjDhw+nevXq5OTk\n0LZtW1avXk2zZs0C1rNixQpmzJjBqlWryM7OpkWLFpxzjnNa3DXXXEOfPn0AGDp0KBMmTODOO+/k\nqquuomPHjnTu3Pm4ug4dOkTPnj1ZuHAhDRo04Oabb+aVV17hnnvuAaBmzZqsXLmSl19+mZEjRzJ+\n/PgCl0NBlxT/7LPPqFSpkl1S3BiXogxmMK/wCoMZzFCGeh0poJjtSahqN1WtraqlVbWu20Cgqj1V\nNdNv3DdVtbGqtlDVVqr6VqxyRoNvl5NvV9Mbb7xBq1ataNmyJevWrTuua8jfxx9/zNVXX0358uWp\nXLkyV1xxxbFha9eu5YILLqBp06a89tprQS81nufrr7+mfv36NGjg9OD16NGDJUuWHBt+zTXXAJCW\nlnbsooAFKeiS4vfdd59dUtwYH4/zOCMZyW3cxghGIIjXkQIqUZ/O/L7xA+w/GJ1LhXfq1ImBAwey\ncuVKDhw4QFpaGt9//z0jR45k2bJlVKtWjZ49e3Lo0KFC1d+zZ0/mzJlD8+bNmTRpEosXLy5S3rzL\njUfiUuN5lxSfM2eOXVLcGNezPMsjPEIPevAiL8ZtAwHxcUyi2KtYsSIZGRnccsstx/Yi9u3bR4UK\nFahSpQrbt2/nnXfeybeOCy+8kDlz5nDw4EH2799/3Pj79++ndu3aHD16lNdee+1YeaVKldi/f/8J\ndTVs2JDNmzfz3XffATB16lQuuuiiIr3Ggi4p3qlTJ7ukuDFAJpncx310oQvjGU9SnG+G7atbjHTr\n1o2rr776WLdT8+bNadmyJY0aNaJevXqcd955+U7fqlUrrr/+epo3b87JJ59Mq1atjg17/PHHOfvs\ns6lVqxZnn332sYaha9eu9OnThxdeeOHYAWuAcuXK8eqrr9KlSxeys7M566yz6N+/f1ivJ9xLimdn\nZ5OUlGSXFDcl2lSmchu30ZGOTGMapRJhExzs3NhEfMTj7ySiJZGyqtrvJKIlkbKqJlbeSGedpbM0\nSZP0Yr1YD+rBiNatWox/J2GMMcXdO7xDN7pxDufwb/5NOcp5HSlk1kgYY0wULWYx13ANTWnK27xN\nRRLjsiR5SkQjoTG8PpWJDHvPTHHwX/5LRzpyGqfxHu9RlcQ79lbsG4ly5cqxe/du2+gkEFVl9+7d\nlCuXOLvkxvj7gi+4jMs4hVP4gA+oSU2vIxVKAhxaL5q6deuybds2du7cWeC4hw4dSpgNUyJlhfDz\nlitX7rizp4xJJOtZT3vaU5nKLGQhtantdaRCK/aNROnSpalfv35I4y5evJiWLVtGOVFkJFJWSLy8\nxhTWRjbSlrYkk8xCFvJH/uh1pCIp9o2EMcbEyla20pa2HOEIi1nMGZzhdaQis0bCGGMiYDvbuYRL\n2MMePuRDmtCk4IkSQLE/cG1McZVDDhsqbUCxkzK89gu/0I52bGMb85lPGmleR4oYaySMSVCDGcyA\ntAFMYpLXUUq0fezjMi7jG75hLnM5j/wvsZNorJEwJgHNZjajGEWZnDI8yINkYRdI9MIBDtCRjnzB\nF8xkJm1p63WkiLNGwpgEs4lN9KIXZ3EWz6x+hp/5mRGM8DpWiXOYw1zN1XzCJ0xjGldypdeRosIa\nCWMSyCEO0ZnOJJHEG7xBs73N6E53RjKSLWzxOl6JcZSjXM/1LGAB4xnP9VzvdaSoiVkjISITRWSH\niKz1KRsmIj+IyCr3cYXPsAdE5DsR+VpELo1VTmPi2T3cwxd8wRSmkEoqAE/xFEkkMYQh3oYrIXLI\noQc9+Df/5kVepBe9vI4UVbHck5gEXBagfLQ6tyltoarzAUTkTJx7Xzd2p3lZRMK7+bQxxcxrvMYY\nxjCYwXSk47HyetRjEIOYwQyWstTDhMVfLrn0ox/Tmc4IRnAHd3gdKepieY/rJcAvIY7eCZihqodV\n9XvgO6B11MIZE+fWs55+9OMCLmA4w08YPpjBnMqpDGQgueR6kLD4U5SBDGQCExjKUO7nfq8jxYTE\n8sJ3IpIKzFPVJu7zYUBPYB+wHPirqu4RkX8A/1XVae54E4B3VHVWgDr7An0BUlJS0vLu/FYYWVlZ\nVKyYGJfxTaSskFh54y3rwaSDDEgbwL7S+xi7fCw1j/x+oTjfrO+lvMeIP43gwfUP0m57O6/i5ive\nlm1+/LNOqD+BaX+cRuetnblt421xd1/qoizbjIyMFaqaHnBgsLsRReMBpAJrfZ6nAMk4ezTDgYlu\n+T+AG33GmwB0Lqj+QHemC0dJvmtWtCVS3njKmqu5eqPeqKKiH+gHJwz3zZqjOZqu6VpH62iWZsUw\nZejiadkWxDfrk/qkomhf7au5mutdqHwUyzvTqep2Vc1R1VxgHL93Kf0A1PMZta5bZkyJMp7xTGMa\nwxhW4Dn4SSQxmtH8wA+MZGSMEhZ/L/IiD/Ig3enOy7wcd3sQ0eZpIyEivtfPvRrIO/NpLtBVRMqK\nSH3gDODzWOczxktf8AV3ciftac9QhoY0zfmcz3Vcx9M8zTa2RTlh8TeRidzFXVzN1UxiEsmUvPNn\nYnkK7HTgU6ChiGwTkVuBv4vIGhFZDWQAAwFUdR3wBvAV8C5wu6rmxCqrMV7by1660IWa1GQa00gK\n46P6NE+TSy4P8mAUExZ/H578Ib3pzaVcynSmU6qEXg81Zq9aVbsFKJ6Qz/jDIcBpHMYUc4rSi15s\nYQsf8RG1qBXW9Kmkci/38hRPcQd30NpODAzbYhbzZKMnuYAL+Bf/oixlvY7kGfvFtTFx5nmeZzaz\nGcEIzuXcQtXxAA+QQgoDGWhXiQ3TPvbRgx7UPlSbt3iL8pT3OpKnrJEwJo58yqcMYhB/4S/cy72F\nrqcSlRjOcJaylDd4I4IJi797uZdtbGPIhiFUprLXcTxnjYQxcWIXu7iO6/gDf+BVXi3yWTQ96UkL\nWnA/93OQgxFKWby9zdtMYAKDGUzjfY29jhMXrJEwJg7kksuN3MgOdjCTmVSlapHrTCaZUYxiC1sY\nzegIpCzefuEX+tCHJjRhGMO8jhM3rJEwJg48xVO8x3s8z/O0olXE6s0gg6u5mid5kp/4KWL1Fkd3\ncic72ckUppToA9X+rJEwxmOLWMTDPMwN3EA/+kW8/r/zd45wJOTfWpREs5jF67zOwzxMS1p6HSeu\nWCNhjId+4ie60Y2GNGQMY6Lya97TOZ27uZtXeZWVrIx4/YluO9sZwADSSLPLrQdgjYQxHskmm650\nZT/7mcUsKhK9C98NZSg1qMG93GunxPpQlP70Zz/7mcxkSlPa60hxxxoJYzzyMA+zhCVkksmZnBnV\neVWhCo/zOB/xEbOZHdV5JZJpTGMOc3iCJ2iMnc0UiDUSxnjgbd7mKZ6iD324iZtiMs/e9KYxjRnE\nIA5zOCbzjGfb2Mad3Ml5nMdA54pAJgBrJIyJsS1s4SZuogUteIEXYjbfUpRiNKPZxKaYzjceKUpv\nenOUoyX2wn2hskbCmBg6whGu4zpyyGEmMylHuZjOvx3t6EhHHudxdrAjpvOOJ+MYx3u8xzM8w+mc\n7nWcuGaNhDExNIhBfM7nvEf5QSIAACAASURBVMqrnm2cRjKSgxzkYR72ZP5e28Qm7uVe2tKW/vT3\nOk7cs0bCmBiZyUxe4AXu4R6u4RrPcjSkIbdzO+MYxxrWeJbDC7nk0oteJJHERCaGdQn2ksqWkDEx\n8A3fcCu3cg7n8DRPex2Hh3mYKlQpcVeJfYEXWMISnud5/sAfvI6TEKyRMCbKDnKQLnShDGV4gzco\nQxmvI1Gd6jzKoyxkIfOY53WcmPiar3mAB+hIR3rS0+s4CcMaCWOi7E7uZDWrmcpU6h1363Zv9ac/\njWjEfdzHEY54HSeqssmmBz0oT3nGMrbE3ae6KGJ5+9KJIrJDRNb6lD0jIhtEZLWIzBaRqm55qogc\nFJFV7iMzVjmNiaTJTGYCE3iIh7icy72Oc5zSlOZZnuUbvuFlXvY6TlQ9wzN8xme8zMvUprbXcRJK\nLPckJgGX+ZW9DzRR1WbAN8ADPsM2qmoL92GnIJiEs4Y1DGAAGWTwKI96HSegy7mc9rTnUR5lN7u9\njhMVq1nNIzxCF7pwPdd7HSfhxKyRUNUlwC9+ZQtUNdt9+l+gbqzyGBNN+9lPF7pQhSq8zutx+2Mt\nQRjFKPaxr1jeQ+EIR7iZm6lGtWK/txQt8XRM4hbgHZ/n9UXkCxH5SEQu8CqUMeFSlL705Vu+ZTrT\nOYVTvI6Ur8Y0pj/9eYVX+IqvvI4TUU/wBF/yJeMYR01qeh0nIYlq7E5/E5FUYJ6qNvErfwhIB65R\nVRWRskBFVd0tImnAHKCxqu4LUGdfoC9ASkpK2owZMwqdLysri4oVo3clzkhKpKyQWHmLmnXOqXN4\nvsHz9N7Um+7/6x7BZCeK1HLdW3ov3c/uTpO9TRixZkQEkgUWy/VgQ6UN3N7qdtptb8eQDeFfAjyR\n1lkoWt6MjIwVqpoecKCqxuwBpAJr/cp6Ap8C5fOZbjGQXlD9aWlpWhSLFi0q0vSxlEhZVRMrb1Gy\nLtNlWkbL6BV6heZoTuRCBRHJ5fqsPqsoOl/nR6xOf7FaDw7oAf2T/knral3do3sKVUcirbOqRcsL\nLNcg21VPu5tE5DJgMHCVqh7wKa8lIsnu36cBZwCbvElpTGj2sIcudOEUTmEKUxLu17x3cAenczp/\n5a8c5ajXcYrkb/yN9axnAhMicr/wkiyWp8BOx9ljaCgi20TkVuAfQCXgfb9TXS8EVovIKmAW0F9V\nfwlYsTFxQFF60IMf+IE3eIMa1PA6UtjKUIaRjGQ96xnLWK/jFNrHfMwoRtGf/rSnvddxEl6pWM1I\nVbsFKJ4QZNw3gTejm8iYyBnJSN7iLZ7nec7mbK/jFNpVXEUGGcfuuV2Nal5HCksWWfSkJ6mk8gzP\neB2nWEis/WFj4tDHfMwDPEBnOnMnd3odp0gEYTSj2cMeHudxr+OE7X7u53u+ZxKTono72JLEGglj\nimAHO+hKV+pTn/GMLxaXe2hOc27lVl7kRb7hG6/jhOx93udlXmYgA7mQC72OU2xYI2FMIeWQQ3e6\n8wu/MItZVKGK15Ei5gme4CROYhCDvI4Skr3s5RZuoRGNeIInvI5TrFgjYUwhPc7jfMAH/IN/0Jzm\nXseJqBRSeIiHmMtcPuADr+MU6B7u4Sd+YjKTOYmTvI5TrFgjYUwhLGABj/EYPejBLdzidZyouJu7\nSSWVe7mXHHK8jhPUXOYyiUkMYQitae11nGLHGgljwvQDP9Cd7pzJmbzES8XiOEQg5SjHMzzDGtYw\nIfCJiJ7bzW760pdmNCuxt2ONNmskjAnDUY5yPddziEPMYhYVqOB1pKi6lmu5gAsYylD2stfrOCe4\nndv5hV+YwpS4uJlTcWSNhDFheJAH+YRPGMc4GtHI6zhRl3dK7C528SRPeh3nOP90/w1jWLE7JhRP\nrJEwJkT/5t+MZCS3cRtd6ep1nJhJI42buZnneI5NcXJ1nJ/5mdu4jda0ZjCDvY5TrFkjYUwINrGJ\nHvQgjTRGMcrrODH3JE9SilJxsUFWlD704QAHmMxkSsXuwhElkjUSxhTgEIe4jusQhJnMpCxlvY4U\nc6dyKg/wAG/yJh/xkadZJjOZeczjSZ4sEV1+XrNGwpgC3Mu9rGAFk5lMfep7Hcczf+Wv1KMeAxno\n2SmxW9nK3dzNhVzI3dztSYaSxhoJY/LxOq/zCq8wiEFcxVVex/HUSZzE0zzNF3zBFKbEfP6Kcgu3\nkEMOr/Jqwl2KPVHZUjYmiA1soC99OZ/zGc5wr+PEha505RzO4UEeZD/7YzrvTDL5gA94lmc5jdNi\nOu+SzBoJYwL4jd/oTGfKU54ZzKA0pb2OFBfyTon9mZ95mqdjNt+NbOQ+7qM97enr3K3YxEiRGwkR\nsU+PKVYU5TZu4yu+4nVepw51vI4UV87hHG7gBkYyki1sifr8csihJz0pTWkmMKHY/sI9XoXVSIjI\nXSJyrc/zCcBBEflaRBpGPJ0xHphfez5TmMIjPMIlXOJ1nLg0ghEkkcQQhkR9Xs/xHP/hP7zAC9Sl\nbtTnZ44X7p7EXcBOABG5ELgOuAFYBTwb2WjGxN4qVvH8Gc9zCZcwlKFex4lb9ajHfdzHDGawlKVR\nm8961vMQD9GJTtzETVGbjwku3EaiDvC9+/eVwExVfQMYBpxT0MQiMlFEdojIWp+y6iLyvoh86/5f\nzS0XEXlBRL4TkdUi0irMrMaEJe9HWlWOVuE1XiOZZK8jxbXBDOZUTuUe7iGX3IjXn002N3MzFanI\nGMZYN5NHwm0k9gEnu3+3Axa6fx8FyoUw/STgMr+yIcBCVT3DrS9v//Vy4Az30Rd4JcysxoRlufuv\n+5bunHxsNTfBVKQiT/EUy1jG67we8fpHMILlLCeTTFJIiXj9JjThNhILgHEiMh44HXjHLW/M73sY\nQanqEuAXv+JOwGT378nAX3zKp6jjv0BVEakdZl5jQjaGMVSgAu22t/M6SsK4kRtJJ50hDOE3fotY\nvatYxaM8Sje60ZnOEavXhE9UNfSRRSoDw4E/AK+o6rtu+aPAYVUt8DKRIpIKzFPVJu7zX1W1qvu3\nAHtUtaqIzANGqOp/3GELgftVdblffX1x9jRISUlJmzFjRsivx19WVhYVKybGzdMTKSvEf96sUll0\n+XMX2m5vS/+V/eM6q694WK5rqqzhrpZ30fP7nvTY0iPfcUPJe0SOMCBtAL+W/pVXl71K5ezKkYwb\nsnhYtuEoSt6MjIwVqpoecKCqxvQBpAJrfZ7/6jd8j/v/POB8n/KFQHp+daelpWlRLFq0qEjTx1Ii\nZVWN/7wv6ouKost1edxn9RUvWbtoFz1JT9KtujXf8ULJ+4A+oCg6T+dFKF3hxMuyDVVR8gLLNch2\nNdxTYM/0PdVVRNqJyDQReUBECnuUb3teN5L7/w63/Aegns94dd0yYyJKUTLJJJ100kjzOk5Cepqn\nySWXB3mwSPV8xmc8zdPcwi10oEOE0pmiCPeYxESgJYCI1AP+DVQHbgeeKGSGuUDePmoPt8688pvd\ns5zOAfaq6k+FnIcxQX3CJ6xjHf3p73WUhFWf+gxkIFOZyud8Xqg6DnCAm7mZutRlNKMjnNAUVriN\nRCNgpft3Z+AzVb0CuAnoVtDEIjId+BRoKCLbRORWYATQTkS+BS5xnwPMBzYB3wHjgNvCzGpMSMYw\nhspULlE3EoqGB3iAFFIYyECU0I915nmIh/iGb5jIRCrjzXEIc6Jw79aRDBxx/26LsyEH2AgFn6Om\nqsEakrYBxlWcPRRjomYXu5jJTPrQp9jfrzraKlOZ4QynN715gze4nutDnnYxi3mO57iDO2h74ubA\neCjcPYm1wAARuQBnw/6uW14H2BXJYMbEwmQmc5jD9KOf11GKhZ70pAUtGMxgDnIwpGn2s59e9OJ0\nTmfEsY4EEy/CbSTuB/oAi4HpqrrGLb8KCtkRaYxHFGUMYziP82hCE6/jFAvJJDOKUfyP/4V8XGEQ\ng9jCFiYxyfbm4lBYjYQ6P4arBdRU1Vt8Bo0BBkQymDHRtohFfMu3dsA6wjLI4C/8hSd5kp/I/1yT\n93iPMYzhPu7jPM6LUUITjrAvFa6qOThXfm0iIo1FpJyqblbVHQVObEwcySST6lS3X/RGwTM8wxGO\n8BAPBR1nD3u4lVs5kzN5jMdimM6EI9zfSZQSkWeAPcCXwBpgj4j83e4rYRLJz/zMbGbTi16UC+my\nYyYcp3M6d3EXk5jEymMnRB7vbu7mZ35mClPsPYhj4e5J/B24EegPNMC5+N4AnFNgn4psNGOiZyIT\nySbb7nIWRUMZSg1qBDwldjazmcpUhjLUfsAY58JtJG4AblXVyaq60X1MAnoD3SOezpgoyCGHsYzl\nYi6mAQ28jlNsVaUqj/EYS1jCbGYfK9/JTvrRj5a0zLc7ysSHcBuJKji/ifC3Eaha9DjGRN8CFrCF\nLXbAOgb60IfGNGYQgzjMYRRlAAPYy16mMMXuHZ4Awm0kvsS5O52/u91hxsS9vPsTdKKT11GKvVKU\nYjSj2cQmXuAFPjz5Q97kTR7jMTvtOEGE+4vrwcB8EbkE+K9bdg5wKs5NgoyJa1vZyjzmMYQhlKGM\n13FKhHa0owMdeJzH4Qw4h3O4j/u8jmVCVJjfSTQAZgEV3cdM4FIC72EYE1fGM/7YbUpN7IxkJAc5\nyJGkI0xmst0aNoGEuyeBqv4Ixx9tEpHmwLWRCmVMNGSTzXjGcxmXkUqq13FKlEY0YhrT2LJuCw2a\n2ckCiSTsH9MZk6jmMY8f+dEOWHvkeq6n9S+tvY5hwmSNhCkxMsmkLnW5giu8jmJMwrBGwpQIm9jE\ne7xHb3pTKvxeVmNKrJA+LSIyt4BR7A4hJq6NZSzJJNOb3l5HMSahhPqVancIw78vYhZjouIIR5jI\nRK7kSupQx+s4xiSUkBoJVe0VrQAi0hD4p0/RacDDOL/g7gPsdMsfVNX5GBOm2cxmJzvtgLUxheB5\n56yqfg20ABCRZOAHYDbQCxitqiM9jGeKgUwyqU992tHO6yjGJJx4O3DdFtioqlu8DmKKhw1sYDGL\n6UtfkuJudTcm/omqFjxWjIjIRGClqv5DRIYBPYF9wHLgr6q6J8A0fcG53nNKSkrajBkzCj3/rKws\nKlasWOjpYymRsoJ3eV/6v5eYU2cO//z0n1Q/Wj2kaRJp2SZSVkisvImUFYqWNyMjY4WqpgccqKpx\n8QDKALuAFPd5CpCMs7czHJhYUB1paWlaFIsWLSrS9LGUSFlVvcl7QA9oNa2m1+v1YU2XSMs2kbKq\nJlbeRMqqWrS8wHINsl2Np/3vy3H2IrYDqOp2Vc1R1VxgHGA/1TRhmclM9rDHDlgbUwTx1Eh0A6bn\nPRGR2j7DrgbWxjyRSWiZZNKQhlzERV5HMSZheX52E4CIVADaAf18iv8uIi0ABTb7DTMmX6tZzad8\nyihGIYjXcYxJWHHRSKjqb0ANv7KbPIpjioExjKEsZelBD6+jGJPQ4qm7yZiIyCKLqUzleq6nOqGd\n0WSMCcwaCVPsTGc6+9lPP+uhNKbIrJEwxU4mmTSlKX/mz15HMSbhWSNhipXlLGclK+lPfztgbUwE\nWCNhipVMMqlABW7kRq+jGFMsWCNhio1f+ZXpTOcGbqCy3eLEmIiwRsIUG9OYxgEO2AFrYyLIGglT\nLCjKGMaQTjpppHkdx5hiIy5+TGdMUS1lKWtZy3jGex3FmGLF9iRMsZBJJpWpTFe6eh3FmGLFGgmT\n8Haxi5nM5CZuogIVvI5jTLFijYRJeJOZzGEO2wFrY6LAGgmT0PIOWJ/HeTSlqddxjCl2rJEwCW0R\ni/iWb+3GQsZEiTUSJqFlkkl1qtOZzl5HMaZYskbCJKyf+ZnZzKYnPSlHOa/jGFMsWSNhEtarvEo2\n2fSlr9dRjCm24ubHdCKyGdgP5ADZqpouItWBfwKpOLcwvU5V93iV0cSPHHIYy1gu5mIa0tDrOMYU\nW/G2J5Ghqi1UNd19PgRYqKpnAAvd58awgAVsZrMdsDYmyuKtkfDXCZjs/j0Z+IuHWUwcySSTkzmZ\nTnTyOooxxVo8NRIKLBCRFSKS18mcoqo/uX//DKR4E83Ek61sZR7zuJVbKUMZr+MYU6yJqnqdAQAR\nqaOqP4jIycD7wJ3AXFWt6jPOHlWt5jddX3COXKakpKTNmDGj0BmysrKoWLFioaePpUTKCpHNOyl1\nElP+OIXXPnuN2odqR6ROX4m0bBMpKyRW3kTKCkXLm5GRscKnm/94qhp3D2AYcB/wNVDbLasNfJ3f\ndGlpaVoUixYtKtL0sZRIWVUjl/eoHtVT9VS9XC+PSH2BJNKyTaSsqomVN5GyqhYtL7Bcg2xX46K7\nSUQqiEilvL+B9sBaYC7Qwx2tB/BvbxKaeDGPefzIj3bA2pgYiZdTYFOA2SICTqbXVfVdEVkGvCEi\ntwJbgOs8zGjiQCaZ1KEOV3CF11GMKRHiopFQ1U1A8wDlu4G2sU9k4tEmNrGABTzCI5SKj1XXmGIv\nLrqbjAnFOMaRRBK96e11FGNKDGskTEI4whEmMIEruZI61PE6jjElhjUSJiHMZjY72Wk3FjImxqyR\nMAlhDGNIJZX2tPc6ijElijUSJu5tYAOLWEQ/+pFkq6wxMWWfOBP3xjKW0pSmF728jmJMiWONhIlr\nBznIJCZxDdeQYpfuMibmrJEwcW0mM9nDHjtgbYxHrJEwcW0MY2hAA9rQxusoxpRI1kiYuLWa1Sxl\nKf3pjyBexzGmRLJGwsStMYyhLGXpcewaj8aYWLNGwsSlLLKYylSu4zqqU93rOMaUWNZImLg0gxns\nZ79dEtwYj1kjYeJSJpk0pSl/5s9eRzGmRLNGwsSd5SxnBSvsgLUxccAaCRN3MsmkAhW4kRu9jmJM\niWeNhIkre9nLdKbTjW5UprLXcYwp8ayRMHFlGtM4wAE7YG1MnPC8kRCReiKySES+EpF1InK3Wz5M\nRH4QkVXuw25qXMwpSiaZpJNOGmlexzHGEB/3uM4G/qqqK0WkErBCRN53h41W1ZEeZjMxtJSlrGUt\n4xnvdRRjjMvzRkJVfwJ+cv/eLyLrwe5PWRJlkkllKtOVrl5HMca4RFW9znCMiKQCS4AmwL1AT2Af\nsBxnb2NPgGn6An0BUlJS0mbMmFHo+WdlZVGxYsVCTx9LiZQVCs67t9ReupzbhQ4/deDub++OYbIT\nJdKyTaSskFh5EykrFC1vRkbGClVNDzhQVePiAVQEVgDXuM9TgGSc4ybDgYkF1ZGWlqZFsWjRoiJN\nH0uJlFW14LzP6rOKoqt1dWwC5SORlm0iZVVNrLyJlFW1aHmB5Rpku+r5gWsAESkNvAm8pqr/AlDV\n7aqao6q5wDigtZcZTfTkHbA+j/NoSlOv4xhjfHjeSIiIABOA9ao6yqe8ts9oVwNrY53NxMYiFvEt\n39qNhYyJQ54fuAbOA24C1ojIKrfsQaCbiLQAFNgMtgUprjLJpDrV6Uxnr6MYY/x43kio6n8g4AV6\n5sc6i4m97WxnNrO5i7s4iZO8jmOM8eN5d5Mp2SYykWyy6eucoGaMiTPWSBjP5JDDWMZyMRfTkIZe\nxzHGBGCNhPHMAhawmc12wNqYOGaNhPHMGMZwMifzF/7idRRjTBDWSBhPbGMbb/EWt3IrZSjjdRxj\nTBDWSBhPjGc8itKHPl5HMcbkwxoJE3PZZDOOcVzKpdSnvtdxjDH5sEbCxNw85vEjP9qNhYxJANZI\nmJgbwxjqUIcOdPA6ijGmANZImJjaxCbe4z360IdS3v/g3xhTAGskTEyNYxxJJNGb3l5HMcaEwBoJ\nEzNHOMJEJtKRjtSxmw8akxCskTAxM4c57GCHHbA2JoFYI2FiJpNMUkmlPe29jmKMCZE1EiYm/lf+\nfyxiEf3oR5KtdsYkDPu0mph4q/ZblKIUvejldRRjTBiskTBRd5CDLDhlAddwDSmkeB3HGBOGuG8k\nROQyEflaRL4TkSFe5zHhm8Us9pXeZwesjUlAcf1rJhFJBl4C2gHbgGUiMldVv4rkfH7+dQPTto1g\nx5EdfLn2NZJJJllKkUQSySSTJMnk/cv7O0mccYL9HXB4UmmnTilFUlIppyxv3KTSv5eT7PyfV5Y3\nPKn0sek05yiak+0up6S8Bea78CK5iIokk0zqHahHm/JtvI5ijAlTXDcSQGvgO1XdBCAiM4BOQEQb\nif/98CmDmkyGJpGsNcraBh8kuYCCqPvc7/9AZaH+H/Y0AnuqwTOPlkEyq/oFDdCQ+ZeFMk6Ey849\nehTKBLl8ebDGN1LlYU5zzuHDUK5c8LpCmV8owyM0ztmHDsFJId7LPJwvOlEYt/XBg1C+fOj1FnI+\nkZru/5o0gTZtCjfPfMR7I1EH2OrzfBtwtu8IItIXnBskp6SksHjx4rBnoodOZvHyF/ntt/2ULV8W\nJYdczXH+J5dcySVXs1FyySWHHHJAlJy8MnEeqjnkijPGsXLNQZP09zJyyBV1hpGDouSIW+YOU3LJ\nSXKn961flBzJRVGO5h4lOTkZxdki63HrlTrPVX8v9/3bdxznr+P+R44by/0/SL1+daCg4jsNlDuS\nRPstp7O1/e8fOFGfVufYzDT/50GEVFeI9YkqR48epXTp0qFPH2JO33kEFeY8so8epVSgrKHOL795\nhjtOAUSVo9nZlC4VwmYnnPlFIFsg2UePkhVK1kAKm6kIr2VflSqF2v4VJN4biQKp6lhgLEB6erq2\nKUJLunjxYooyfSwlUlaAxVUX0yxB8ibSsk2krJBYeRMpK8BXUcob7weufwDq+Tyv65YZY4yJgXhv\nJJYBZ4hIfREpA3QF5nqcyRhjSoy47m5S1WwRuQN4D0gGJqrqOo9jGWNMiRHXjQSAqs4H5nudwxhj\nSqJ4724yxhjjIWskjDHGBGWNhDHGmKCskTDGGBOUaJR+regFEdkJbClCFTWBXRGKE22JlBUSK69l\njZ5EyptIWaFoef+oqrUCDShWjURRichyVU33OkcoEikrJFZeyxo9iZQ3kbJC9PJad5MxxpigrJEw\nxhgTlDUSxxvrdYAwJFJWSKy8ljV6EilvImWFKOW1YxLGGGOCsj0JY4wxQVkjYYwxJqgS30iIyEQR\n2SEia73OEgoRqScii0TkKxFZJyJ3e50pGBEpJyKfi8iXbtZHvc5UEBFJFpEvRGSe11kKIiKbRWSN\niKwSkeVe5ymIiFQVkVkiskFE1ovIn73OFIiINHSXad5jn4jc43WuYERkoPv5Wisi00UkhPvZhlF/\nST8mISIXAlnAFFWN+7tci0htoLaqrhSRSsAK4C+qGtH7fkeCiAhQQVWzRKQ08B/gblX9r8fRghKR\ne4F0oLKqdvQ6T35EZDOQrqoJ8YMvEZkMfKyq4937w5RX1V+9zpUfEUnGudHZ2apalB/qRoWI1MH5\nXJ2pqgdF5A1gvqpOitQ8SvyehKouAX7xOkeoVPUnVV3p/r0fWI9zL/C4o44s92lp9xG330pEpC7Q\nARjvdZbiRkSqABcCEwBU9Ui8NxCutsDGeGwgfJQCThKRUkB54MdIVl7iG4lEJiKpQEvgM2+TBOd2\n36wCdgDvq2rcZgWeAwYDuV4HCZECC0RkhYj09TpMAeoDO4FX3e688SJSwetQIegKTPc6RDCq+gMw\nEvgf8BOwV1UXRHIe1kgkKBGpCLwJ3KOq+7zOE4yq5qhqC5z7k7cWkbjs0hORjsAOVV3hdZYwnK+q\nrYDLgdvdrtN4VQpoBbyiqi2B34Ah3kbKn9sldhUw0+sswYhINaATTiN8KlBBRG6M5DyskUhAbv/+\nm8Brqvovr/OEwu1aWARc5nWWIM4DrnL7+WcAF4vING8j5c/9Fomq7gBmA629TZSvbcA2nz3JWTiN\nRjy7HFipqtu9DpKPS4DvVXWnqh4F/gWcG8kZWCORYNyDwROA9ao6yus8+RGRWiJS1f37JKAdsMHb\nVIGp6gOqWldVU3G6GD5U1Yh+I4skEangnriA223THojbM/RU9Wdgq4g0dIvaAnF3soWfbsRxV5Pr\nf8A5IlLe3Ta0xTlOGTElvpEQkenAp0BDEdkmIrd6nakA5wE34XzTzTtF7wqvQwVRG1gkIquBZTjH\nJOL+1NIEkQL8R0S+BD4H3lbVdz3OVJA7gdfc9aEF8KTHeYJyG952ON/M45a7ZzYLWAmswdmmR/Ty\nHCX+FFhjjDHBlfg9CWOMMcFZI2GMMSYoaySMMcYEZY2EMcaYoKyRMMYYE5Q1EqZEEZFJ8XaFVxHp\nJCLfiki2iEzyOk8wItJGRFREanqdxcSONRImZtwNtIrI3/zKS/rGZwLOL+j/CMTtpd9NyWSNhIm1\nQ8AgEanldZBIci+VUpjpqgI1gPdU9QdV3RvZZMYUjTUSJtYWAZuBvwUbIdCehYikumXpfuNc7l4F\n9aCIfCwidUXkIvdGR1kiMk9EagSYx1AR2e6O86p72ZC8YSIig0Vko1vvGt+Lpvlk6SYiH4rIQaBf\nkNdSTUQmi8get64PRKRx3msA9rijfujW2SZIPWVE5Gn3qgAHRGSZiFwaYJl1dH+Ff8hdLml+9Vzj\nvp7DIrJVRB5yL+fgO58nRWSLO84mEbnLL05zEfnMzbFcRFr5TF9FRKaKcyOvQ+70cXvDHhMCVbWH\nPWLyACYB84ArgCPA/7nlbXAue10z0HO3LNUtS/cb53PgAqAZzrWLPgEWAmfj3Dzoe+BFvwz7ca7s\n2QS4FOemMi/4jDMc+BrnYoT1gRtwrlrawS/LZqCzO07dIK/53zjXq7oQaArMBbYCJwFlgDPduq4B\nTgHKBKnnNeC/bj2nAXe4y7C53/LY4L6mJu5r/Ann5j4AaUAO8CjQAOiOc8OtO33mMx3nYnzXuvPJ\nAG4OsMwzgEbAezjXCsq7esOLwCqciw3+0Z2mi9frnj2K8Ln1OoA9Ss4jr5Fw/14EzHD/LkojcanP\nOHe4Za18yoYBa/0y/ApU9Cm7ETj8/+2dT2hcVRSHvxPaRppqQWqwEFpBXGRRKIVAiloEN120m4Ig\nCGrQvYuIhpBqiJvQFZ5dtwAAA1RJREFURhARRS1pIcSFVbCRolBoaSkUFFLSVCs2rStX6cL0DyUo\nOS7OfXjzmpfMJEOnlt8Hj5k57849511m3pl7zp17gLZ03AWeL9n+EVHxK7eld4XrfSa125PJNgNz\nwJvp9ZbU5oVl+nmaqHGxrST/Dvi0NB6vZOc3pWstdI0TGxfmfQwSu7Pm9u6tsGOpMX82yTrS6wlg\ntNmfNR2NO9YhRHN4F7hgZofX2M+l7HmxpfN0SdZefo//VzEPYoPHDcTNuBV4BPjRzPKNzdYTM4ec\nlepKdxI39wuFwN3nzGyamEHUyi7AgF+zyBDJ1tOltrmu2yVdncDJUvvzwPtm9hhRwGqBcODLkY95\nUQWtnZiBfAZ8k8Jcp4Dv3f3sCv2JBxg5CdEU3P0nM/sWOAR8UDpdVIbL74hVieG/825T32VZPbm3\nou1+YhvmKl0QIajVUs/Omi2pfdcSNtxdgw2rteeeMSeNm7v/YGbbiVoMLwInzey4u/c0xkxxv1Hi\nWjSTfiKfUC5ENJset2aynQ3Uu8MWl87sJuL714gaB/PAdnefKR311jm+QnzHdheC9It9B/XVUrhI\nOMwnl7Dpz1Lb7kxXG5GbKOoLXCHCQznPEeGmW0QuoYXIN6wad7/h7mPu/jrwBvCambWupU/RPDST\nEE3D3WfM7Avu/W/ADJHcHTSzPiIHMNBA1euAUTMbIko+DgNfuvsdADMbAUbSqp9zRGy/G1hw95r3\n6nf3q2Z2Avjcogb1X0RS/CbwVR39/G5m48AxM+slagc8TuQIrvvi6oQDZjZLhIHeI5xfoetD4Gcz\nG0yyLqCXcNaFnq+BI2b2VtLTATzl7mO12JrGdBL4hRjnA8nG+VqvVzxYaCYhms0Q8E8uSOGil4nV\nNVPEapz+Buo8S9zEzhBlP08D72TnDxIJ3bdTu1PEap8/VqGrh1gNNJEeNxKJ4XrDRD3AUSI89xux\nSmwPUJ7d9BHOYJJIRO8rnJ+7TwIvpWu5TDjHYeCT7P2vEg7k46TnGJFsr5V5whFOESvNHiVCd+J/\niooOCfEQkP5fcQZ4wt1vNNkc8RChmYQQQohK5CSEEEJUonCTEEKISjSTEEIIUYmchBBCiErkJIQQ\nQlQiJyGEEKISOQkhhBCV/AsA3SgTtbdWcAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VQtZVmxHoHJi",
        "colab_type": "text"
      },
      "source": [
        "For a learning rate of 0.01, the validation loss curve is very wavy, due to the large updates in the paramters at every gradient descent step. The model overshoots the minima and possibly oscillates around the minima also, which is why we get very large loss values at some epochs."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QZDbp_mEoeSg",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "f54f3d7c-308c-4078-8809-3746a7ca5ba7"
      },
      "source": [
        "lr = 0.0005\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "\n",
        "history = train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1097/1100 [============================>.] - ETA: 0s - loss: 0.4034 - accuracy: 0.8563\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.86240, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.4032 - accuracy: 0.8563 - val_loss: 0.4118 - val_accuracy: 0.8624\n",
            "Epoch 2/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.2709 - accuracy: 0.9021\n",
            "Epoch 00002: val_accuracy improved from 0.86240 to 0.87760, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2706 - accuracy: 0.9023 - val_loss: 0.3451 - val_accuracy: 0.8776\n",
            "Epoch 3/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.2264 - accuracy: 0.9176\n",
            "Epoch 00003: val_accuracy improved from 0.87760 to 0.91020, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2263 - accuracy: 0.9176 - val_loss: 0.2746 - val_accuracy: 0.9102\n",
            "Epoch 4/50\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.1956 - accuracy: 0.9296\n",
            "Epoch 00004: val_accuracy did not improve from 0.91020\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1956 - accuracy: 0.9297 - val_loss: 0.2954 - val_accuracy: 0.9048\n",
            "Epoch 5/50\n",
            "1097/1100 [============================>.] - ETA: 0s - loss: 0.1708 - accuracy: 0.9391\n",
            "Epoch 00005: val_accuracy improved from 0.91020 to 0.92820, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1709 - accuracy: 0.9390 - val_loss: 0.2158 - val_accuracy: 0.9282\n",
            "Epoch 6/50\n",
            "1097/1100 [============================>.] - ETA: 0s - loss: 0.1416 - accuracy: 0.9481\n",
            "Epoch 00006: val_accuracy did not improve from 0.92820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1418 - accuracy: 0.9481 - val_loss: 0.2388 - val_accuracy: 0.9232\n",
            "Epoch 7/50\n",
            "1092/1100 [============================>.] - ETA: 0s - loss: 0.1217 - accuracy: 0.9562\n",
            "Epoch 00007: val_accuracy did not improve from 0.92820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1216 - accuracy: 0.9562 - val_loss: 0.2438 - val_accuracy: 0.9264\n",
            "Epoch 8/50\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.1035 - accuracy: 0.9618\n",
            "Epoch 00008: val_accuracy did not improve from 0.92820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1035 - accuracy: 0.9618 - val_loss: 0.2543 - val_accuracy: 0.9214\n",
            "Epoch 9/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.0852 - accuracy: 0.9690\n",
            "Epoch 00009: val_accuracy did not improve from 0.92820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0852 - accuracy: 0.9690 - val_loss: 0.3072 - val_accuracy: 0.9056\n",
            "Epoch 10/50\n",
            "1095/1100 [============================>.] - ETA: 0s - loss: 0.0710 - accuracy: 0.9743\n",
            "Epoch 00010: val_accuracy did not improve from 0.92820\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0711 - accuracy: 0.9743 - val_loss: 0.2608 - val_accuracy: 0.9236\n",
            "Epoch 00010: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEeCAYAAACZlyICAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PCb1IYbBo3pd",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        },
        "outputId": "761c6521-4196-401c-f06f-9441876eb3bc"
      },
      "source": [
        "accuracy_score()"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
            "Training accuracy: 95.709%\n",
            "Validation accuracy: 92.82%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E9fqfBYppXGb",
        "colab_type": "text"
      },
      "source": [
        "For the following parameters:\n",
        "```\n",
        "lr = 0.0005\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "```\n",
        "\n",
        "We get: \n",
        "```\n",
        "Training accuracy: 95.709%\n",
        "Validation accuracy: 92.82%\n",
        "The training stops after 10 epochs.\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZVlbxo5ZpJll",
        "colab_type": "text"
      },
      "source": [
        "The accuracy is a slight improvement over the case where we used learning rate of 0.001."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GA8aIfItp1Ve",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "2e614792-3a2e-4d30-fe57-f05a22861a55"
      },
      "source": [
        "lr = 0.0001\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "\n",
        "history = train_model(X_train, y_train_one_hot, X_val, y_val_one_hot, learning_rate=lr, batch_size=batch_size, \n",
        "            num_epochs=max_epochs, choice_init=choice_init, batch_norm=batch_norm)"
      ],
      "execution_count": 41,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.4366 - accuracy: 0.8469\n",
            "Epoch 00001: val_accuracy improved from -inf to 0.84720, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.4364 - accuracy: 0.8469 - val_loss: 0.4466 - val_accuracy: 0.8472\n",
            "Epoch 2/50\n",
            "1099/1100 [============================>.] - ETA: 0s - loss: 0.2989 - accuracy: 0.8922\n",
            "Epoch 00002: val_accuracy improved from 0.84720 to 0.86200, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2990 - accuracy: 0.8921 - val_loss: 0.3847 - val_accuracy: 0.8620\n",
            "Epoch 3/50\n",
            "1094/1100 [============================>.] - ETA: 0s - loss: 0.2480 - accuracy: 0.9113\n",
            "Epoch 00003: val_accuracy improved from 0.86200 to 0.88840, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2480 - accuracy: 0.9113 - val_loss: 0.3270 - val_accuracy: 0.8884\n",
            "Epoch 4/50\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.2148 - accuracy: 0.9225\n",
            "Epoch 00004: val_accuracy did not improve from 0.88840\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.2152 - accuracy: 0.9224 - val_loss: 0.3923 - val_accuracy: 0.8728\n",
            "Epoch 5/50\n",
            "1092/1100 [============================>.] - ETA: 0s - loss: 0.1839 - accuracy: 0.9345\n",
            "Epoch 00005: val_accuracy improved from 0.88840 to 0.89600, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1840 - accuracy: 0.9345 - val_loss: 0.3492 - val_accuracy: 0.8960\n",
            "Epoch 6/50\n",
            "1100/1100 [==============================] - ETA: 0s - loss: 0.1574 - accuracy: 0.9420\n",
            "Epoch 00006: val_accuracy improved from 0.89600 to 0.90420, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1574 - accuracy: 0.9420 - val_loss: 0.2876 - val_accuracy: 0.9042\n",
            "Epoch 7/50\n",
            "1095/1100 [============================>.] - ETA: 0s - loss: 0.1332 - accuracy: 0.9519\n",
            "Epoch 00007: val_accuracy did not improve from 0.90420\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1329 - accuracy: 0.9519 - val_loss: 0.4013 - val_accuracy: 0.8838\n",
            "Epoch 8/50\n",
            "1096/1100 [============================>.] - ETA: 0s - loss: 0.1132 - accuracy: 0.9585\n",
            "Epoch 00008: val_accuracy did not improve from 0.90420\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.1134 - accuracy: 0.9584 - val_loss: 0.3825 - val_accuracy: 0.8880\n",
            "Epoch 9/50\n",
            "1097/1100 [============================>.] - ETA: 0s - loss: 0.0900 - accuracy: 0.9678\n",
            "Epoch 00009: val_accuracy improved from 0.90420 to 0.92340, saving model to best_model.h5\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0899 - accuracy: 0.9678 - val_loss: 0.3005 - val_accuracy: 0.9234\n",
            "Epoch 10/50\n",
            "1093/1100 [============================>.] - ETA: 0s - loss: 0.0767 - accuracy: 0.9726\n",
            "Epoch 00010: val_accuracy did not improve from 0.92340\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0765 - accuracy: 0.9727 - val_loss: 0.5608 - val_accuracy: 0.8676\n",
            "Epoch 11/50\n",
            "1095/1100 [============================>.] - ETA: 0s - loss: 0.0659 - accuracy: 0.9761\n",
            "Epoch 00011: val_accuracy did not improve from 0.92340\n",
            "1100/1100 [==============================] - 7s 6ms/step - loss: 0.0659 - accuracy: 0.9761 - val_loss: 0.3585 - val_accuracy: 0.9098\n",
            "Epoch 00011: early stopping\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEeCAYAAACOtbLLAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nq+LEYGZmVf4fEdL5kqY6zjUAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HaFmcsVuqMN5",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 88
        },
        "outputId": "9be92e15-d222-4245-bff2-8c93c7e7734c"
      },
      "source": [
        "accuracy_score()"
      ],
      "execution_count": 42,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.\n",
            "Training accuracy: 97.438%\n",
            "Validation accuracy: 92.34%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LQX0V8UgqUq4",
        "colab_type": "text"
      },
      "source": [
        "For the following parameters:\n",
        "```\n",
        "lr = 0.0001\n",
        "batch_size = 50\n",
        "max_epochs = 50\n",
        "choice_init = 'glorot_uniform'\n",
        "batch_norm = True\n",
        "```\n",
        "\n",
        "We get: \n",
        "```\n",
        "Training accuracy: 97.438%\n",
        "Validation accuracy: 92.34%\n",
        "The training stops after 10 epochs.\n",
        "```"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "08qteS0gmMR7",
        "colab_type": "text"
      },
      "source": [
        "**CONCLUSION:** We can see that the best results are obtained by using Xaiver initialization, a batch size of 50, early stopping to avoid overfitting, learning rate = 0.0005. The use of batch normalization also helps mitigate the problem of overfitting."
      ]
    }
  ]
}