training_computervision_mnist.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "id": "6e64d027",
      "metadata": {
        "id": "6e64d027"
      },
      "source": [
        "## Coding problem : classifying handwritten digits (MNIST dataset)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "8da42603",
      "metadata": {
        "id": "8da42603"
      },
      "source": [
        "The objective of this task is to train a simple neural network on the mnist dataset in order to classify the handwritten digits into numbers ranging from zero to 9.\n",
        "The problem is a multi class classification problem on image data.\n",
        "\n",
        "Number of classes = 10\n",
        "\n",
        "Task = multi class classification"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "6cf3228c",
      "metadata": {
        "id": "6cf3228c"
      },
      "source": [
        "### Reading the data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 160,
      "id": "b113b315",
      "metadata": {
        "id": "b113b315"
      },
      "outputs": [],
      "source": [
        "from keras.datasets import mnist\n",
        "from matplotlib import pyplot\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "import numpy as np"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 161,
      "id": "ecJurJOWK8af",
      "metadata": {
        "id": "ecJurJOWK8af"
      },
      "outputs": [],
      "source": [
        "from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay\n",
        "from sklearn.metrics import accuracy_score"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 183,
      "id": "0bdd4c64",
      "metadata": {
        "id": "0bdd4c64"
      },
      "outputs": [],
      "source": [
        "(train_X, train_y), (test_X, test_y) = mnist.load_data()"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "78d3bbe3",
      "metadata": {
        "id": "78d3bbe3"
      },
      "source": [
        "### Data visualisation "
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 184,
      "id": "fa7b44f8",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fa7b44f8",
        "outputId": "096ab6ab-9182-47a8-ecf4-e10829491721"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_train:(60000, 28, 28)\n",
            "Y_train:(60000,)\n",
            "X_test:(10000, 28, 28)\n",
            "Y_test:(10000,)\n"
          ]
        }
      ],
      "source": [
        "print('X_train:' + str(train_X.shape))\n",
        "print('Y_train:' + str(train_y.shape))\n",
        "print('X_test:'  + str(test_X.shape))\n",
        "print('Y_test:'  + str(test_y.shape))"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "WjT0HqHiaF9s",
      "metadata": {
        "id": "WjT0HqHiaF9s"
      },
      "source": [
        "Preparing for deployment : extracting one image to be tested later in the webservice."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 185,
      "id": "CrT1tNv9ZuQR",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "CrT1tNv9ZuQR",
        "outputId": "88509077-9239-4b5d-ba6c-26d69b364b07"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "execution_count": 185,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "import cv2\n",
        "cv2.imwrite(\"sample.png\",train_X[59980])"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "FvQAAQ36ajMq",
      "metadata": {
        "id": "FvQAAQ36ajMq"
      },
      "source": [
        "Data vizualisation"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 186,
      "id": "87b09fc0",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "87b09fc0",
        "outputId": "35d87cc9-ee31-4e8f-860f-67cb5319f7fb"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "for i in range(59995,60000):  \n",
        "    pyplot.imshow(train_X[i])\n",
        "    pyplot.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "e2cc9ba8",
      "metadata": {
        "id": "e2cc9ba8"
      },
      "source": [
        "### Preprocessing"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 187,
      "id": "7c5f4e83",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "7c5f4e83",
        "outputId": "147b71c6-ca57-4497-a5c2-452ec1cc4645"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "28\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "784"
            ]
          },
          "execution_count": 187,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "image_size = train_X.shape[1]\n",
        "print(image_size)\n",
        "input_size = image_size * image_size\n",
        "input_size"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "eedd87df",
      "metadata": {
        "id": "eedd87df"
      },
      "source": [
        "Resize the NN input fot training purposes"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 188,
      "id": "4313262e",
      "metadata": {
        "id": "4313262e"
      },
      "outputs": [],
      "source": [
        "train_X = np.reshape(train_X, [-1, input_size])\n",
        "test_X = np.reshape(test_X, [-1, input_size])"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "75f0d6c0",
      "metadata": {
        "id": "75f0d6c0"
      },
      "source": [
        "In order to train a neural network the data should be converted into floats in the training set.\n",
        "The values of the numbers in the training set should be scaled between 0 and 255 (the data is already standarized) we divide by 255 to normalise the data and have values between 0 and 1."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 189,
      "id": "2606395d",
      "metadata": {
        "id": "2606395d"
      },
      "outputs": [],
      "source": [
        "train_X=tf.cast(train_X, tf.float32) / 255.0"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 190,
      "id": "a14f10b7",
      "metadata": {
        "id": "a14f10b7"
      },
      "outputs": [],
      "source": [
        "train_y=tf.cast(train_y, tf.int32) "
      ]
    },
    {
      "cell_type": "markdown",
      "id": "18117b44",
      "metadata": {
        "id": "18117b44"
      },
      "source": [
        "One hot encoding of the classes (y), the hyperparameter depth corresponds to the number of classes"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 191,
      "id": "69cba9ba",
      "metadata": {
        "id": "69cba9ba"
      },
      "outputs": [],
      "source": [
        "train_y = tf.one_hot(train_y, depth=10)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "93036d0f",
      "metadata": {
        "id": "93036d0f"
      },
      "source": [
        "Building training set using the tensorflow api"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 192,
      "id": "d68b5e10",
      "metadata": {
        "id": "d68b5e10"
      },
      "outputs": [],
      "source": [
        "\n",
        "train = tf.data.Dataset.from_tensor_slices((train_X, train_y))"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "8b3c3bc7",
      "metadata": {
        "id": "8b3c3bc7"
      },
      "source": [
        "Similarily to the previous steps we build a test set scaled and compatible with the tensorflow api."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 193,
      "id": "59879a88",
      "metadata": {
        "id": "59879a88"
      },
      "outputs": [],
      "source": [
        "test_X=tf.cast(test_X, tf.float32) / 255.0\n",
        "test_y=tf.cast(test_y, tf.int32) \n",
        "test_y = tf.one_hot(test_y, depth=10)\n",
        "test = tf.data.Dataset.from_tensor_slices((test_X, test_y))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 194,
      "id": "11f533d0",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "11f533d0",
        "outputId": "20def530-de28-4939-ab02-dc5f49961498"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "tensors: [0 0 0 ... 0 0 0]\n"
          ]
        }
      ],
      "source": [
        "import sys\n",
        "\n",
        "tf.print(\"tensors:\", train_X[0], output_stream=sys.stdout)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 195,
      "id": "Sppg7A055KSx",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Sppg7A055KSx",
        "outputId": "8066bf08-9f33-4c25-8e44-40307817d5f0"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "<tf.Tensor: shape=(60000, 784), dtype=float32, numpy=\n",
              "array([[0., 0., 0., ..., 0., 0., 0.],\n",
              "       [0., 0., 0., ..., 0., 0., 0.],\n",
              "       [0., 0., 0., ..., 0., 0., 0.],\n",
              "       ...,\n",
              "       [0., 0., 0., ..., 0., 0., 0.],\n",
              "       [0., 0., 0., ..., 0., 0., 0.],\n",
              "       [0., 0., 0., ..., 0., 0., 0.]], dtype=float32)>"
            ]
          },
          "execution_count": 195,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "train_X"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 196,
      "id": "iON0n2iV3r6Z",
      "metadata": {
        "id": "iON0n2iV3r6Z"
      },
      "outputs": [],
      "source": [
        "\n",
        "val_X=train_X[55000:]\n",
        "train_X=train_X[:55000]\n",
        "val_y=train_y[55000:]\n",
        "train_y=train_y[:55000]\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "0443c34f",
      "metadata": {
        "id": "0443c34f"
      },
      "source": [
        "Designing the Neural network (initial architecture)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 175,
      "id": "ad4619d9",
      "metadata": {
        "id": "ad4619d9"
      },
      "outputs": [],
      "source": [
        "img_classifier = keras.Sequential([\n",
        "    #input_shape=(28, 28)), #number of pixels are 28x28\n",
        "    #hidden layers choosing a number of layer as a power of 8 with a relu activation function. Relu is ideal for representing complex relationships between the input and the output because of its non linearity\n",
        "    keras.layers.Dense(units=256, activation='relu', input_dim=input_size),\n",
        "    keras.layers.Dense(units=192, activation='relu'),\n",
        "    keras.layers.Dense(units=128, activation='relu'),\n",
        "    keras.layers.Dense(units=10, activation='softmax') #number of classes 10  \n",
        "])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "1079e322",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1079e322",
        "outputId": "64492bec-db07-40dc-b290-de6d5929dcf3"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Help on package keras.api._v2.keras.losses in keras.api._v2.keras:\n",
            "\n",
            "NAME\n",
            "    keras.api._v2.keras.losses - Built-in loss functions.\n",
            "\n",
            "PACKAGE CONTENTS\n",
            "\n",
            "\n",
            "FILE\n",
            "    /usr/local/lib/python3.9/dist-packages/keras/api/_v2/keras/losses/__init__.py\n",
            "\n",
            "\n"
          ]
        }
      ],
      "source": [
        "help(tf.losses)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "a50e5c2f",
      "metadata": {
        "id": "a50e5c2f"
      },
      "outputs": [],
      "source": [
        "img_classifier.compile(optimizer='adam', \n",
        "              #loss function categorical_crossentropy because we re in a multi class classification problem, other loss functions like binary crossentropy wdnt be a good fit\n",
        "              loss=\"categorical_crossentropy\",\n",
        "              metrics=['accuracy'])\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "iHxp7PF1ymPw",
      "metadata": {
        "id": "iHxp7PF1ymPw"
      },
      "outputs": [],
      "source": [
        "img_classifier.build()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "sO2e7175y93F",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sO2e7175y93F",
        "outputId": "874f676c-bd7f-4cff-805f-c2471670b60e"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"sequential_3\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " dense_12 (Dense)            (None, 256)               200960    \n",
            "                                                                 \n",
            " dense_13 (Dense)            (None, 192)               49344     \n",
            "                                                                 \n",
            " dense_14 (Dense)            (None, 128)               24704     \n",
            "                                                                 \n",
            " dense_15 (Dense)            (None, 10)                1290      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 276,298\n",
            "Trainable params: 276,298\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "img_classifier.summary()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "PoLpJyg0QURZ",
      "metadata": {
        "id": "PoLpJyg0QURZ"
      },
      "outputs": [],
      "source": [
        "early_stopping_cb=keras.callbacks.EarlyStopping(patience=10)\n",
        "early_stopping_mc=keras.callbacks.ModelCheckpoint(\"minst.h5\",save_best_only=False)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "LXDDNXMD5C_4",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LXDDNXMD5C_4",
        "outputId": "aaa97add-1b39-4f66-b073-ec8cc7d173d9"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "<tf.Tensor: shape=(0, 784), dtype=float32, numpy=array([], shape=(0, 784), dtype=float32)>"
            ]
          },
          "execution_count": 49,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "val_X"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "KYcQ5KxGzGXl",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KYcQ5KxGzGXl",
        "outputId": "342c1cf1-8d8f-4ed5-ec11-aa9c83e7b0f3"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 29ms/step - loss: 0.4472 - accuracy: 0.8783 - val_loss: 0.1408 - val_accuracy: 0.9622\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.1519 - accuracy: 0.9553 - val_loss: 0.1055 - val_accuracy: 0.9704\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 16ms/step - loss: 0.0998 - accuracy: 0.9702 - val_loss: 0.0978 - val_accuracy: 0.9702\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0754 - accuracy: 0.9766 - val_loss: 0.0819 - val_accuracy: 0.9776\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0536 - accuracy: 0.9838 - val_loss: 0.0708 - val_accuracy: 0.9788\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0410 - accuracy: 0.9875 - val_loss: 0.0749 - val_accuracy: 0.9792\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.0316 - accuracy: 0.9905 - val_loss: 0.0733 - val_accuracy: 0.9788\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0254 - accuracy: 0.9927 - val_loss: 0.0681 - val_accuracy: 0.9826\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0206 - accuracy: 0.9940 - val_loss: 0.0823 - val_accuracy: 0.9774\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0151 - accuracy: 0.9959 - val_loss: 0.0871 - val_accuracy: 0.9764\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 16ms/step - loss: 0.0126 - accuracy: 0.9964 - val_loss: 0.0787 - val_accuracy: 0.9820\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0092 - accuracy: 0.9975 - val_loss: 0.0837 - val_accuracy: 0.9810\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 0.0100 - accuracy: 0.9971 - val_loss: 0.0813 - val_accuracy: 0.9796\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.0088 - accuracy: 0.9973 - val_loss: 0.0866 - val_accuracy: 0.9804\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0102 - accuracy: 0.9967 - val_loss: 0.0798 - val_accuracy: 0.9826\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 16ms/step - loss: 0.0055 - accuracy: 0.9986 - val_loss: 0.0934 - val_accuracy: 0.9804\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 17ms/step - loss: 0.0033 - accuracy: 0.9992 - val_loss: 0.0907 - val_accuracy: 0.9832\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 16ms/step - loss: 0.0022 - accuracy: 0.9995 - val_loss: 0.0939 - val_accuracy: 0.9828\n"
          ]
        }
      ],
      "source": [
        "history = img_classifier.fit(train_X, train_y, epochs=20, batch_size=500, verbose=1,\n",
        "                                    validation_data=(val_X, val_y), callbacks=[early_stopping_cb,early_stopping_mc])"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "_djWBjrtaOl4",
      "metadata": {
        "id": "_djWBjrtaOl4"
      },
      "source": [
        "### Base Model Evaluation"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "pQAGGKvzGZqU",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pQAGGKvzGZqU",
        "outputId": "d0690e08-2d57-49bc-897f-0a2aef7722a5"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "20/20 [==============================] - 0s 6ms/step - loss: 0.0843 - accuracy: 0.9809\n",
            "\n",
            "Test accuracy: 98.1%\n"
          ]
        }
      ],
      "source": [
        "loss, acc = img_classifier.evaluate(test_X, test_y, batch_size=500)\n",
        "print(\"\\nTest accuracy: %.1f%%\" % (100.0 * acc))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "MKULnd2dF5eC",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 795
        },
        "id": "MKULnd2dF5eC",
        "outputId": "6a3e2743-8ff9-4451-dc61-43205695b490"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "# Plot the loss function\n",
        "fig, ax = pyplot.subplots(1, 1, figsize=(10,6))\n",
        "ax.plot(np.sqrt(history.history['loss']), 'r', label='train')\n",
        "ax.plot(np.sqrt(history.history['val_loss']), 'b' ,label='val')\n",
        "ax.set_xlabel(r'Epoch', fontsize=20)\n",
        "ax.set_ylabel(r'Loss', fontsize=20)\n",
        "ax.legend()\n",
        "ax.tick_params(labelsize=20)\n",
        "\n",
        "# Plot the accuracy\n",
        "fig, ax = pyplot.subplots(1, 1, figsize=(10,6))\n",
        "ax.plot(np.sqrt(history.history['accuracy']), 'r', label='train')\n",
        "ax.plot(np.sqrt(history.history['val_accuracy']), 'b' ,label='val')\n",
        "ax.set_xlabel(r'Epoch', fontsize=20)\n",
        "ax.set_ylabel(r'Accuracy', fontsize=20)\n",
        "ax.legend()\n",
        "ax.tick_params(labelsize=20)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "mRDqoltEK5Fe",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "mRDqoltEK5Fe",
        "outputId": "6c29b080-3eab-41f4-dab0-6bb983e29830"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "313/313 [==============================] - 1s 2ms/step\n"
          ]
        }
      ],
      "source": [
        "pred_y=img_classifier.predict(test_X)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "xxjEfMlpM59K",
      "metadata": {
        "id": "xxjEfMlpM59K"
      },
      "outputs": [],
      "source": [
        "pred_y=np.argmax(pred_y, axis=1)\n",
        "\n",
        "test_y=np.argmax(test_y, axis=1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "gdM1aZ5qLBXb",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 279
        },
        "id": "gdM1aZ5qLBXb",
        "outputId": "3c41ea75-2bcf-4701-dfcf-94ae758ece37"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "cm = confusion_matrix(test_y, pred_y)\n",
        "disp = ConfusionMatrixDisplay(confusion_matrix=cm)\n",
        "disp.plot()\n",
        "\n",
        "pyplot.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "HuH_ScJnAHpr",
      "metadata": {
        "id": "HuH_ScJnAHpr"
      },
      "source": [
        "As we can see the base classifier is doing well. we have a Test accuracy equal to 98% and the confusion matrix shows very limited misclassification. We need though to optimise further the hyper parameters in order to improve the loss witnessed in the above figure."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "YSARwsYu6xzg",
      "metadata": {
        "id": "YSARwsYu6xzg"
      },
      "outputs": [],
      "source": [
        "# Import datasets, classifiers and performance metrics\n",
        "from sklearn import metrics"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "vcncZZLD6iTz",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vcncZZLD6iTz",
        "outputId": "33356dd5-2c0e-4f2f-c1ae-aa73348c0e60"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Classification report for classifier <keras.engine.sequential.Sequential object at 0x7f32d3d82340>:\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.99      0.99      0.99       980\n",
            "           1       0.99      0.99      0.99      1135\n",
            "           2       0.98      0.98      0.98      1032\n",
            "           3       0.98      0.98      0.98      1010\n",
            "           4       0.98      0.98      0.98       982\n",
            "           5       0.99      0.98      0.98       892\n",
            "           6       0.99      0.98      0.98       958\n",
            "           7       0.98      0.97      0.98      1028\n",
            "           8       0.98      0.98      0.98       974\n",
            "           9       0.97      0.98      0.97      1009\n",
            "\n",
            "    accuracy                           0.98     10000\n",
            "   macro avg       0.98      0.98      0.98     10000\n",
            "weighted avg       0.98      0.98      0.98     10000\n",
            "\n",
            "\n"
          ]
        }
      ],
      "source": [
        "print(\n",
        "    f\"Classification report for classifier {img_classifier}:\\n\"\n",
        "    f\"{metrics.classification_report(test_y, pred_y)}\\n\"\n",
        ")"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "584f6e45",
      "metadata": {
        "id": "584f6e45"
      },
      "source": [
        "### Hyper parameters tuning\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "0b565500",
      "metadata": {
        "id": "0b565500"
      },
      "outputs": [],
      "source": [
        "\n",
        "from tensorboard.plugins.hparams import api as hp\n",
        "\n",
        "from tensorflow.keras import backend as k\n",
        "HP_DROPOUT = hp.HParam('dropout', hp.Discrete([0.1,0.2, 0.3,0.4,0.45,0.5]))\n",
        "HP_OPTIMIZER = hp.HParam('optimizer', hp.Discrete(['adam', 'sgd']))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "YKdpYXk-TQ3l",
      "metadata": {
        "id": "YKdpYXk-TQ3l"
      },
      "outputs": [],
      "source": [
        "METRIC_acc = 'accuracy'\n",
        "with tf.summary.create_file_writer('logs/mlp/hparam_tuning').as_default():\n",
        "  hp.hparams_config(\n",
        "    hparams=[HP_DROPOUT, HP_OPTIMIZER],\n",
        "    metrics=[hp.Metric(METRIC_acc, display_name='accuracy')],\n",
        "  )"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "_OMWl3odSfYQ",
      "metadata": {
        "id": "_OMWl3odSfYQ"
      },
      "outputs": [],
      "source": [
        "early_stopping_cb=keras.callbacks.EarlyStopping(patience=10)\n",
        "early_stopping_mc=keras.callbacks.ModelCheckpoint(\"minst_training.h5\",save_best_only=False)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "i97t94kIdrAW",
      "metadata": {
        "id": "i97t94kIdrAW"
      },
      "source": [
        "Adding dropout to the architecture and configuring regulation rate!"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "y3JOJ5grVTmj",
      "metadata": {
        "id": "y3JOJ5grVTmj"
      },
      "outputs": [],
      "source": [
        "def train_test_model(run_dir,hparams,eps):\n",
        "  model = tf.keras.models.Sequential([\n",
        "    keras.layers.Dense(units=256, activation='relu', input_dim=input_size),\n",
        "    keras.layers.Dropout(hparams[HP_DROPOUT]),\n",
        "    keras.layers.Dense(units=192, activation='relu'), \n",
        "    keras.layers.Dropout(hparams[HP_DROPOUT]),\n",
        "    keras.layers.Dense(units=128, activation='relu'),\n",
        "    keras.layers.Dropout(hparams[HP_DROPOUT]),\n",
        "    keras.layers.Dense(units=10, activation='softmax') #number of classes 10  \n",
        "  ])\n",
        "  model.compile(\n",
        "      optimizer=hparams[HP_OPTIMIZER],\n",
        "      loss='categorical_crossentropy',\n",
        "      metrics=['accuracy'],\n",
        "  )\n",
        "  #model.build()\n",
        "  #model.summary()\n",
        "  model.fit(train_X, train_y, batch_size=500, verbose=1,\n",
        "                                    validation_data=(val_X, val_y), epochs=eps, callbacks=[\n",
        "        tf.keras.callbacks.TensorBoard(run_dir),  # log metrics\n",
        "        hp.KerasCallback(run_dir, hparams),  # log hparams\n",
        "        early_stopping_cb,\n",
        "        early_stopping_mc\n",
        "    ]) # Run with 1 epoch to speed things up for demo purposes\n",
        "  \n",
        "  acc = model.evaluate(test_X, test_y,batch_size=500,)\n",
        "  return acc , model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "EMWDoW-cVDS6",
      "metadata": {
        "id": "EMWDoW-cVDS6"
      },
      "outputs": [],
      "source": [
        "def run(run_dir, hparams):\n",
        "  with tf.summary.create_file_writer(run_dir).as_default():\n",
        "    hp.hparams(hparams)  # record the values used in this trial\n",
        "    accuracy , model = train_test_model(run_dir,hparams,20)\n",
        "    #tf.summary.scalar(\"accuracy\", accuracy, step=500)\n",
        "    return accuracy , model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "qxjGbCa_Ufw6",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qxjGbCa_Ufw6",
        "outputId": "3acf6429-06ab-4b33-eb02-987468a1d6c9"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "--- Starting trial: run-0\n",
            "{'dropout': 0.1, 'optimizer': 'adam'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 29ms/step - loss: 0.5180 - accuracy: 0.8464 - val_loss: 0.1390 - val_accuracy: 0.9576\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.1759 - accuracy: 0.9477 - val_loss: 0.0986 - val_accuracy: 0.9710\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1195 - accuracy: 0.9638 - val_loss: 0.0839 - val_accuracy: 0.9728\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0908 - accuracy: 0.9722 - val_loss: 0.0773 - val_accuracy: 0.9766\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0700 - accuracy: 0.9793 - val_loss: 0.0674 - val_accuracy: 0.9790\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 3s 32ms/step - loss: 0.0587 - accuracy: 0.9814 - val_loss: 0.0647 - val_accuracy: 0.9810\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 0.0483 - accuracy: 0.9848 - val_loss: 0.0694 - val_accuracy: 0.9792\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0428 - accuracy: 0.9863 - val_loss: 0.0661 - val_accuracy: 0.9810\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0357 - accuracy: 0.9884 - val_loss: 0.0729 - val_accuracy: 0.9800\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0307 - accuracy: 0.9899 - val_loss: 0.0634 - val_accuracy: 0.9820\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.0279 - accuracy: 0.9907 - val_loss: 0.0647 - val_accuracy: 0.9842\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0232 - accuracy: 0.9922 - val_loss: 0.0684 - val_accuracy: 0.9848\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0209 - accuracy: 0.9930 - val_loss: 0.0702 - val_accuracy: 0.9822\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0198 - accuracy: 0.9934 - val_loss: 0.0671 - val_accuracy: 0.9842\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.0177 - accuracy: 0.9944 - val_loss: 0.0658 - val_accuracy: 0.9840\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.0143 - accuracy: 0.9954 - val_loss: 0.0706 - val_accuracy: 0.9822\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0139 - accuracy: 0.9956 - val_loss: 0.0716 - val_accuracy: 0.9840\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0151 - accuracy: 0.9949 - val_loss: 0.0752 - val_accuracy: 0.9832\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0138 - accuracy: 0.9953 - val_loss: 0.0797 - val_accuracy: 0.9830\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.0147 - accuracy: 0.9947 - val_loss: 0.0822 - val_accuracy: 0.9838\n",
            "20/20 [==============================] - 0s 8ms/step - loss: 0.0767 - accuracy: 0.9820\n",
            "--- Starting trial: run-1\n",
            "{'dropout': 0.1, 'optimizer': 'sgd'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 2.1563 - accuracy: 0.2841 - val_loss: 1.8904 - val_accuracy: 0.6228\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 1.6557 - accuracy: 0.5775 - val_loss: 1.1986 - val_accuracy: 0.7820\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 1.1226 - accuracy: 0.6937 - val_loss: 0.7570 - val_accuracy: 0.8498\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.8531 - accuracy: 0.7524 - val_loss: 0.5657 - val_accuracy: 0.8796\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.7160 - accuracy: 0.7887 - val_loss: 0.4643 - val_accuracy: 0.8944\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.6334 - accuracy: 0.8127 - val_loss: 0.4032 - val_accuracy: 0.9048\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.5745 - accuracy: 0.8289 - val_loss: 0.3621 - val_accuracy: 0.9120\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 3s 32ms/step - loss: 0.5338 - accuracy: 0.8427 - val_loss: 0.3325 - val_accuracy: 0.9154\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.5000 - accuracy: 0.8540 - val_loss: 0.3104 - val_accuracy: 0.9214\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.4720 - accuracy: 0.8613 - val_loss: 0.2929 - val_accuracy: 0.9248\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.4531 - accuracy: 0.8660 - val_loss: 0.2794 - val_accuracy: 0.9266\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.4338 - accuracy: 0.8716 - val_loss: 0.2678 - val_accuracy: 0.9290\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.4158 - accuracy: 0.8769 - val_loss: 0.2585 - val_accuracy: 0.9298\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.4026 - accuracy: 0.8805 - val_loss: 0.2498 - val_accuracy: 0.9312\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.3925 - accuracy: 0.8849 - val_loss: 0.2422 - val_accuracy: 0.9338\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.3804 - accuracy: 0.8861 - val_loss: 0.2355 - val_accuracy: 0.9354\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.3685 - accuracy: 0.8904 - val_loss: 0.2288 - val_accuracy: 0.9364\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.3633 - accuracy: 0.8933 - val_loss: 0.2241 - val_accuracy: 0.9370\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.3533 - accuracy: 0.8972 - val_loss: 0.2181 - val_accuracy: 0.9382\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.3426 - accuracy: 0.9006 - val_loss: 0.2133 - val_accuracy: 0.9382\n",
            "20/20 [==============================] - 0s 7ms/step - loss: 0.2658 - accuracy: 0.9247\n",
            "--- Starting trial: run-2\n",
            "{'dropout': 0.2, 'optimizer': 'adam'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 22ms/step - loss: 0.5674 - accuracy: 0.8270 - val_loss: 0.1387 - val_accuracy: 0.9626\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1989 - accuracy: 0.9399 - val_loss: 0.0999 - val_accuracy: 0.9702\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.1438 - accuracy: 0.9573 - val_loss: 0.0838 - val_accuracy: 0.9752\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1123 - accuracy: 0.9659 - val_loss: 0.0782 - val_accuracy: 0.9776\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0927 - accuracy: 0.9713 - val_loss: 0.0697 - val_accuracy: 0.9800\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0766 - accuracy: 0.9763 - val_loss: 0.0718 - val_accuracy: 0.9792\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.0689 - accuracy: 0.9790 - val_loss: 0.0707 - val_accuracy: 0.9802\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 3s 29ms/step - loss: 0.0592 - accuracy: 0.9819 - val_loss: 0.0583 - val_accuracy: 0.9836\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0532 - accuracy: 0.9828 - val_loss: 0.0627 - val_accuracy: 0.9830\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0469 - accuracy: 0.9852 - val_loss: 0.0674 - val_accuracy: 0.9822\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0434 - accuracy: 0.9862 - val_loss: 0.0655 - val_accuracy: 0.9818\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.0373 - accuracy: 0.9877 - val_loss: 0.0654 - val_accuracy: 0.9828\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.0356 - accuracy: 0.9888 - val_loss: 0.0703 - val_accuracy: 0.9804\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0326 - accuracy: 0.9897 - val_loss: 0.0639 - val_accuracy: 0.9818\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0308 - accuracy: 0.9901 - val_loss: 0.0597 - val_accuracy: 0.9832\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0277 - accuracy: 0.9910 - val_loss: 0.0645 - val_accuracy: 0.9852\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.0260 - accuracy: 0.9915 - val_loss: 0.0695 - val_accuracy: 0.9810\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.0248 - accuracy: 0.9920 - val_loss: 0.0739 - val_accuracy: 0.9826\n",
            "20/20 [==============================] - 0s 7ms/step - loss: 0.0751 - accuracy: 0.9816\n",
            "--- Starting trial: run-3\n",
            "{'dropout': 0.2, 'optimizer': 'sgd'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 3s 22ms/step - loss: 2.2175 - accuracy: 0.1918 - val_loss: 2.0194 - val_accuracy: 0.5792\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.8891 - accuracy: 0.4562 - val_loss: 1.4958 - val_accuracy: 0.7100\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 1.4324 - accuracy: 0.5969 - val_loss: 0.9641 - val_accuracy: 0.8032\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 4s 32ms/step - loss: 1.0838 - accuracy: 0.6744 - val_loss: 0.6781 - val_accuracy: 0.8592\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.8919 - accuracy: 0.7240 - val_loss: 0.5349 - val_accuracy: 0.8828\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.7775 - accuracy: 0.7573 - val_loss: 0.4525 - val_accuracy: 0.8936\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.7023 - accuracy: 0.7823 - val_loss: 0.4005 - val_accuracy: 0.9012\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.6465 - accuracy: 0.8012 - val_loss: 0.3628 - val_accuracy: 0.9070\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.6018 - accuracy: 0.8167 - val_loss: 0.3355 - val_accuracy: 0.9122\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.5678 - accuracy: 0.8280 - val_loss: 0.3140 - val_accuracy: 0.9168\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.5401 - accuracy: 0.8360 - val_loss: 0.2982 - val_accuracy: 0.9210\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.5162 - accuracy: 0.8443 - val_loss: 0.2840 - val_accuracy: 0.9234\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.4965 - accuracy: 0.8508 - val_loss: 0.2727 - val_accuracy: 0.9256\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.4792 - accuracy: 0.8567 - val_loss: 0.2628 - val_accuracy: 0.9278\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 0.4639 - accuracy: 0.8605 - val_loss: 0.2542 - val_accuracy: 0.9298\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.4466 - accuracy: 0.8671 - val_loss: 0.2456 - val_accuracy: 0.9310\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.4368 - accuracy: 0.8696 - val_loss: 0.2394 - val_accuracy: 0.9314\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.4250 - accuracy: 0.8731 - val_loss: 0.2341 - val_accuracy: 0.9336\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.4123 - accuracy: 0.8779 - val_loss: 0.2273 - val_accuracy: 0.9360\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.4053 - accuracy: 0.8787 - val_loss: 0.2220 - val_accuracy: 0.9364\n",
            "20/20 [==============================] - 0s 7ms/step - loss: 0.2713 - accuracy: 0.9227\n",
            "--- Starting trial: run-4\n",
            "{'dropout': 0.3, 'optimizer': 'adam'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 3s 22ms/step - loss: 0.6744 - accuracy: 0.7879 - val_loss: 0.1603 - val_accuracy: 0.9520\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.2461 - accuracy: 0.9267 - val_loss: 0.1139 - val_accuracy: 0.9674\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1756 - accuracy: 0.9479 - val_loss: 0.0900 - val_accuracy: 0.9734\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.1415 - accuracy: 0.9580 - val_loss: 0.0777 - val_accuracy: 0.9758\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1205 - accuracy: 0.9640 - val_loss: 0.0713 - val_accuracy: 0.9776\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.1045 - accuracy: 0.9683 - val_loss: 0.0662 - val_accuracy: 0.9798\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0928 - accuracy: 0.9723 - val_loss: 0.0672 - val_accuracy: 0.9790\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0839 - accuracy: 0.9744 - val_loss: 0.0663 - val_accuracy: 0.9810\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.0745 - accuracy: 0.9765 - val_loss: 0.0633 - val_accuracy: 0.9814\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0662 - accuracy: 0.9798 - val_loss: 0.0589 - val_accuracy: 0.9846\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0650 - accuracy: 0.9800 - val_loss: 0.0605 - val_accuracy: 0.9828\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0580 - accuracy: 0.9822 - val_loss: 0.0577 - val_accuracy: 0.9848\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0517 - accuracy: 0.9833 - val_loss: 0.0585 - val_accuracy: 0.9840\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.0503 - accuracy: 0.9846 - val_loss: 0.0627 - val_accuracy: 0.9828\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0481 - accuracy: 0.9852 - val_loss: 0.0613 - val_accuracy: 0.9832\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0444 - accuracy: 0.9857 - val_loss: 0.0682 - val_accuracy: 0.9832\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0420 - accuracy: 0.9869 - val_loss: 0.0645 - val_accuracy: 0.9824\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.0411 - accuracy: 0.9872 - val_loss: 0.0565 - val_accuracy: 0.9862\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.0382 - accuracy: 0.9877 - val_loss: 0.0598 - val_accuracy: 0.9838\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0354 - accuracy: 0.9891 - val_loss: 0.0587 - val_accuracy: 0.9844\n",
            "20/20 [==============================] - 0s 12ms/step - loss: 0.0650 - accuracy: 0.9834\n",
            "--- Starting trial: run-5\n",
            "{'dropout': 0.3, 'optimizer': 'sgd'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 26ms/step - loss: 2.2442 - accuracy: 0.1728 - val_loss: 2.0403 - val_accuracy: 0.5322\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 1.9767 - accuracy: 0.3610 - val_loss: 1.6231 - val_accuracy: 0.6948\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.6266 - accuracy: 0.5056 - val_loss: 1.1558 - val_accuracy: 0.7808\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.3116 - accuracy: 0.5879 - val_loss: 0.8377 - val_accuracy: 0.8252\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 1.0943 - accuracy: 0.6484 - val_loss: 0.6538 - val_accuracy: 0.8606\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.9489 - accuracy: 0.6958 - val_loss: 0.5418 - val_accuracy: 0.8778\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.8550 - accuracy: 0.7261 - val_loss: 0.4698 - val_accuracy: 0.8924\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.7792 - accuracy: 0.7521 - val_loss: 0.4192 - val_accuracy: 0.9012\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.7203 - accuracy: 0.7734 - val_loss: 0.3818 - val_accuracy: 0.9056\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.6688 - accuracy: 0.7913 - val_loss: 0.3520 - val_accuracy: 0.9096\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.6320 - accuracy: 0.8046 - val_loss: 0.3290 - val_accuracy: 0.9160\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.6044 - accuracy: 0.8148 - val_loss: 0.3102 - val_accuracy: 0.9180\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.5747 - accuracy: 0.8258 - val_loss: 0.2950 - val_accuracy: 0.9214\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.5518 - accuracy: 0.8333 - val_loss: 0.2826 - val_accuracy: 0.9242\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.5349 - accuracy: 0.8366 - val_loss: 0.2725 - val_accuracy: 0.9254\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.5139 - accuracy: 0.8438 - val_loss: 0.2620 - val_accuracy: 0.9260\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.5017 - accuracy: 0.8498 - val_loss: 0.2538 - val_accuracy: 0.9286\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.4870 - accuracy: 0.8551 - val_loss: 0.2466 - val_accuracy: 0.9282\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.4717 - accuracy: 0.8591 - val_loss: 0.2393 - val_accuracy: 0.9306\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.4592 - accuracy: 0.8637 - val_loss: 0.2339 - val_accuracy: 0.9318\n",
            "20/20 [==============================] - 0s 11ms/step - loss: 0.2870 - accuracy: 0.9167\n",
            "--- Starting trial: run-6\n",
            "{'dropout': 0.4, 'optimizer': 'adam'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 25ms/step - loss: 0.7946 - accuracy: 0.7425 - val_loss: 0.1836 - val_accuracy: 0.9482\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.3009 - accuracy: 0.9121 - val_loss: 0.1201 - val_accuracy: 0.9646\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.2241 - accuracy: 0.9361 - val_loss: 0.1007 - val_accuracy: 0.9696\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.1785 - accuracy: 0.9485 - val_loss: 0.0848 - val_accuracy: 0.9742\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.1552 - accuracy: 0.9557 - val_loss: 0.0811 - val_accuracy: 0.9774\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 0.1349 - accuracy: 0.9609 - val_loss: 0.0766 - val_accuracy: 0.9776\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1222 - accuracy: 0.9645 - val_loss: 0.0761 - val_accuracy: 0.9774\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.1094 - accuracy: 0.9686 - val_loss: 0.0682 - val_accuracy: 0.9800\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1025 - accuracy: 0.9699 - val_loss: 0.0651 - val_accuracy: 0.9798\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 4s 32ms/step - loss: 0.0982 - accuracy: 0.9716 - val_loss: 0.0646 - val_accuracy: 0.9800\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0879 - accuracy: 0.9741 - val_loss: 0.0685 - val_accuracy: 0.9818\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0828 - accuracy: 0.9758 - val_loss: 0.0616 - val_accuracy: 0.9824\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0793 - accuracy: 0.9766 - val_loss: 0.0629 - val_accuracy: 0.9820\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0722 - accuracy: 0.9780 - val_loss: 0.0625 - val_accuracy: 0.9814\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 4s 34ms/step - loss: 0.0712 - accuracy: 0.9792 - val_loss: 0.0624 - val_accuracy: 0.9830\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0670 - accuracy: 0.9799 - val_loss: 0.0618 - val_accuracy: 0.9820\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0636 - accuracy: 0.9811 - val_loss: 0.0580 - val_accuracy: 0.9838\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0604 - accuracy: 0.9817 - val_loss: 0.0645 - val_accuracy: 0.9816\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0610 - accuracy: 0.9815 - val_loss: 0.0587 - val_accuracy: 0.9836\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.0582 - accuracy: 0.9820 - val_loss: 0.0617 - val_accuracy: 0.9820\n",
            "20/20 [==============================] - 0s 7ms/step - loss: 0.0712 - accuracy: 0.9807\n",
            "--- Starting trial: run-7\n",
            "{'dropout': 0.4, 'optimizer': 'sgd'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 3s 22ms/step - loss: 2.2712 - accuracy: 0.1551 - val_loss: 2.0932 - val_accuracy: 0.5008\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 4s 32ms/step - loss: 2.0988 - accuracy: 0.2737 - val_loss: 1.8186 - val_accuracy: 0.6686\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 1.8703 - accuracy: 0.3961 - val_loss: 1.4256 - val_accuracy: 0.7518\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 1.5858 - accuracy: 0.4960 - val_loss: 1.0394 - val_accuracy: 0.8038\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.3385 - accuracy: 0.5646 - val_loss: 0.7927 - val_accuracy: 0.8348\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.1782 - accuracy: 0.6109 - val_loss: 0.6511 - val_accuracy: 0.8600\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 4s 32ms/step - loss: 1.0506 - accuracy: 0.6549 - val_loss: 0.5555 - val_accuracy: 0.8760\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.9554 - accuracy: 0.6877 - val_loss: 0.4895 - val_accuracy: 0.8872\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.8827 - accuracy: 0.7118 - val_loss: 0.4415 - val_accuracy: 0.8950\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.8200 - accuracy: 0.7331 - val_loss: 0.4050 - val_accuracy: 0.8992\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.7716 - accuracy: 0.7535 - val_loss: 0.3745 - val_accuracy: 0.9066\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.7334 - accuracy: 0.7676 - val_loss: 0.3508 - val_accuracy: 0.9100\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.6970 - accuracy: 0.7813 - val_loss: 0.3326 - val_accuracy: 0.9146\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 20ms/step - loss: 0.6682 - accuracy: 0.7909 - val_loss: 0.3154 - val_accuracy: 0.9184\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.6385 - accuracy: 0.8011 - val_loss: 0.3013 - val_accuracy: 0.9198\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.6234 - accuracy: 0.8099 - val_loss: 0.2902 - val_accuracy: 0.9212\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.5985 - accuracy: 0.8151 - val_loss: 0.2801 - val_accuracy: 0.9226\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.5710 - accuracy: 0.8236 - val_loss: 0.2689 - val_accuracy: 0.9236\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.5652 - accuracy: 0.8277 - val_loss: 0.2618 - val_accuracy: 0.9252\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.5496 - accuracy: 0.8332 - val_loss: 0.2544 - val_accuracy: 0.9270\n",
            "20/20 [==============================] - 0s 8ms/step - loss: 0.3098 - accuracy: 0.9108\n",
            "--- Starting trial: run-8\n",
            "{'dropout': 0.45, 'optimizer': 'adam'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 31ms/step - loss: 0.8952 - accuracy: 0.7087 - val_loss: 0.1913 - val_accuracy: 0.9440\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.3337 - accuracy: 0.9053 - val_loss: 0.1225 - val_accuracy: 0.9634\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.2484 - accuracy: 0.9299 - val_loss: 0.1038 - val_accuracy: 0.9698\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.2058 - accuracy: 0.9431 - val_loss: 0.0915 - val_accuracy: 0.9744\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1771 - accuracy: 0.9510 - val_loss: 0.0873 - val_accuracy: 0.9746\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 4s 34ms/step - loss: 0.1589 - accuracy: 0.9562 - val_loss: 0.0800 - val_accuracy: 0.9758\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1427 - accuracy: 0.9604 - val_loss: 0.0748 - val_accuracy: 0.9766\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1329 - accuracy: 0.9627 - val_loss: 0.0714 - val_accuracy: 0.9794\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1207 - accuracy: 0.9653 - val_loss: 0.0676 - val_accuracy: 0.9804\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.1112 - accuracy: 0.9681 - val_loss: 0.0682 - val_accuracy: 0.9812\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.1067 - accuracy: 0.9696 - val_loss: 0.0698 - val_accuracy: 0.9814\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0998 - accuracy: 0.9711 - val_loss: 0.0683 - val_accuracy: 0.9816\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0940 - accuracy: 0.9725 - val_loss: 0.0658 - val_accuracy: 0.9812\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0898 - accuracy: 0.9736 - val_loss: 0.0670 - val_accuracy: 0.9822\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.0874 - accuracy: 0.9748 - val_loss: 0.0657 - val_accuracy: 0.9822\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.0825 - accuracy: 0.9761 - val_loss: 0.0636 - val_accuracy: 0.9824\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.0817 - accuracy: 0.9757 - val_loss: 0.0620 - val_accuracy: 0.9816\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.0743 - accuracy: 0.9782 - val_loss: 0.0592 - val_accuracy: 0.9828\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0765 - accuracy: 0.9776 - val_loss: 0.0631 - val_accuracy: 0.9826\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.0742 - accuracy: 0.9779 - val_loss: 0.0607 - val_accuracy: 0.9836\n",
            "20/20 [==============================] - 0s 7ms/step - loss: 0.0721 - accuracy: 0.9807\n",
            "--- Starting trial: run-9\n",
            "{'dropout': 0.45, 'optimizer': 'sgd'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 3s 22ms/step - loss: 2.3016 - accuracy: 0.1381 - val_loss: 2.1205 - val_accuracy: 0.5474\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 2.1301 - accuracy: 0.2383 - val_loss: 1.8785 - val_accuracy: 0.6672\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 3s 29ms/step - loss: 1.9311 - accuracy: 0.3447 - val_loss: 1.5333 - val_accuracy: 0.7136\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 1.6910 - accuracy: 0.4378 - val_loss: 1.1872 - val_accuracy: 0.7480\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.4681 - accuracy: 0.5105 - val_loss: 0.9346 - val_accuracy: 0.7864\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.3029 - accuracy: 0.5623 - val_loss: 0.7712 - val_accuracy: 0.8232\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 1.1717 - accuracy: 0.6051 - val_loss: 0.6598 - val_accuracy: 0.8484\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 3s 29ms/step - loss: 1.0740 - accuracy: 0.6422 - val_loss: 0.5776 - val_accuracy: 0.8644\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 0.9887 - accuracy: 0.6729 - val_loss: 0.5182 - val_accuracy: 0.8778\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.9275 - accuracy: 0.6930 - val_loss: 0.4731 - val_accuracy: 0.8884\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.8721 - accuracy: 0.7159 - val_loss: 0.4393 - val_accuracy: 0.8934\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.8264 - accuracy: 0.7319 - val_loss: 0.4100 - val_accuracy: 0.8994\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.7904 - accuracy: 0.7449 - val_loss: 0.3880 - val_accuracy: 0.9022\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.7578 - accuracy: 0.7572 - val_loss: 0.3677 - val_accuracy: 0.9074\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.7283 - accuracy: 0.7687 - val_loss: 0.3501 - val_accuracy: 0.9100\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.6993 - accuracy: 0.7807 - val_loss: 0.3343 - val_accuracy: 0.9136\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.6743 - accuracy: 0.7890 - val_loss: 0.3202 - val_accuracy: 0.9146\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 3s 30ms/step - loss: 0.6531 - accuracy: 0.7953 - val_loss: 0.3093 - val_accuracy: 0.9164\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.6384 - accuracy: 0.8021 - val_loss: 0.2998 - val_accuracy: 0.9192\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.6247 - accuracy: 0.8061 - val_loss: 0.2907 - val_accuracy: 0.9200\n",
            "20/20 [==============================] - 0s 11ms/step - loss: 0.3491 - accuracy: 0.9012\n",
            "--- Starting trial: run-10\n",
            "{'dropout': 0.5, 'optimizer': 'adam'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 4s 24ms/step - loss: 0.9892 - accuracy: 0.6709 - val_loss: 0.2014 - val_accuracy: 0.9458\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.3733 - accuracy: 0.8930 - val_loss: 0.1384 - val_accuracy: 0.9630\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.2806 - accuracy: 0.9213 - val_loss: 0.1139 - val_accuracy: 0.9686\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 3s 29ms/step - loss: 0.2352 - accuracy: 0.9343 - val_loss: 0.0984 - val_accuracy: 0.9714\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.2057 - accuracy: 0.9424 - val_loss: 0.0949 - val_accuracy: 0.9728\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.1816 - accuracy: 0.9493 - val_loss: 0.0846 - val_accuracy: 0.9746\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1694 - accuracy: 0.9531 - val_loss: 0.0827 - val_accuracy: 0.9776\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1544 - accuracy: 0.9577 - val_loss: 0.0787 - val_accuracy: 0.9768\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.1452 - accuracy: 0.9597 - val_loss: 0.0749 - val_accuracy: 0.9808\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1334 - accuracy: 0.9627 - val_loss: 0.0737 - val_accuracy: 0.9788\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1293 - accuracy: 0.9632 - val_loss: 0.0723 - val_accuracy: 0.9810\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.1187 - accuracy: 0.9664 - val_loss: 0.0702 - val_accuracy: 0.9808\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.1141 - accuracy: 0.9671 - val_loss: 0.0715 - val_accuracy: 0.9806\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.1137 - accuracy: 0.9680 - val_loss: 0.0684 - val_accuracy: 0.9802\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1041 - accuracy: 0.9705 - val_loss: 0.0680 - val_accuracy: 0.9806\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.1036 - accuracy: 0.9705 - val_loss: 0.0670 - val_accuracy: 0.9798\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0982 - accuracy: 0.9723 - val_loss: 0.0684 - val_accuracy: 0.9804\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0953 - accuracy: 0.9723 - val_loss: 0.0677 - val_accuracy: 0.9810\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 4s 34ms/step - loss: 0.0923 - accuracy: 0.9729 - val_loss: 0.0663 - val_accuracy: 0.9804\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.0878 - accuracy: 0.9745 - val_loss: 0.0649 - val_accuracy: 0.9812\n",
            "20/20 [==============================] - 0s 7ms/step - loss: 0.0766 - accuracy: 0.9788\n",
            "--- Starting trial: run-11\n",
            "{'dropout': 0.5, 'optimizer': 'sgd'}\n",
            "Epoch 1/20\n",
            "110/110 [==============================] - 3s 22ms/step - loss: 2.3492 - accuracy: 0.1246 - val_loss: 2.1552 - val_accuracy: 0.4854\n",
            "Epoch 2/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 2.2030 - accuracy: 0.1899 - val_loss: 1.9957 - val_accuracy: 0.6398\n",
            "Epoch 3/20\n",
            "110/110 [==============================] - 3s 29ms/step - loss: 2.0621 - accuracy: 0.2750 - val_loss: 1.7481 - val_accuracy: 0.6934\n",
            "Epoch 4/20\n",
            "110/110 [==============================] - 3s 24ms/step - loss: 1.8767 - accuracy: 0.3548 - val_loss: 1.4386 - val_accuracy: 0.7316\n",
            "Epoch 5/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.6848 - accuracy: 0.4223 - val_loss: 1.1720 - val_accuracy: 0.7660\n",
            "Epoch 6/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 1.5158 - accuracy: 0.4761 - val_loss: 0.9751 - val_accuracy: 0.7998\n",
            "Epoch 7/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.3738 - accuracy: 0.5254 - val_loss: 0.8286 - val_accuracy: 0.8222\n",
            "Epoch 8/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 1.2581 - accuracy: 0.5684 - val_loss: 0.7220 - val_accuracy: 0.8428\n",
            "Epoch 9/20\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 1.1638 - accuracy: 0.6021 - val_loss: 0.6422 - val_accuracy: 0.8570\n",
            "Epoch 10/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.0871 - accuracy: 0.6284 - val_loss: 0.5790 - val_accuracy: 0.8702\n",
            "Epoch 11/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 1.0226 - accuracy: 0.6556 - val_loss: 0.5290 - val_accuracy: 0.8812\n",
            "Epoch 12/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.9663 - accuracy: 0.6777 - val_loss: 0.4876 - val_accuracy: 0.8884\n",
            "Epoch 13/20\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.9188 - accuracy: 0.6951 - val_loss: 0.4550 - val_accuracy: 0.8952\n",
            "Epoch 14/20\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.8791 - accuracy: 0.7094 - val_loss: 0.4283 - val_accuracy: 0.9006\n",
            "Epoch 15/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.8359 - accuracy: 0.7261 - val_loss: 0.4037 - val_accuracy: 0.9038\n",
            "Epoch 16/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.8113 - accuracy: 0.7370 - val_loss: 0.3817 - val_accuracy: 0.9076\n",
            "Epoch 17/20\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.7774 - accuracy: 0.7505 - val_loss: 0.3640 - val_accuracy: 0.9094\n",
            "Epoch 18/20\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.7549 - accuracy: 0.7584 - val_loss: 0.3496 - val_accuracy: 0.9120\n",
            "Epoch 19/20\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.7268 - accuracy: 0.7688 - val_loss: 0.3354 - val_accuracy: 0.9136\n",
            "Epoch 20/20\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.7040 - accuracy: 0.7778 - val_loss: 0.3219 - val_accuracy: 0.9164\n",
            "20/20 [==============================] - 0s 11ms/step - loss: 0.3814 - accuracy: 0.8945\n"
          ]
        }
      ],
      "source": [
        "session_num = 0\n",
        "models=[]\n",
        "for dropout_rate in HP_DROPOUT.domain.values:\n",
        "    for optimizer in HP_OPTIMIZER.domain.values:\n",
        "      hparams = {\n",
        "          HP_DROPOUT: dropout_rate,\n",
        "          HP_OPTIMIZER: optimizer,\n",
        "      }\n",
        "      run_name = \"run-%d\" % session_num\n",
        "      print('--- Starting trial: %s' % run_name)\n",
        "      print({h.name: hparams[h] for h in hparams})\n",
        "      models.append(run('./logs/hparam_tuning_2_3/' + run_name, hparams))\n",
        "      session_num += 1"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "GES5lMa9c5-k",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "GES5lMa9c5-k",
        "outputId": "9449fd6a-a405-4d4a-bfeb-d774e98ed5ca"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "2023-03-15 22:46:54.457387: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/nvidia/lib:/usr/local/nvidia/lib64\n",
            "2023-03-15 22:46:54.457501: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/nvidia/lib:/usr/local/nvidia/lib64\n",
            "2023-03-15 22:46:54.457516: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.\n",
            "2023-03-15 22:46:56.098255: E tensorflow/compiler/xla/stream_executor/cuda/cuda_driver.cc:267] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected\n",
            "\n",
            "NOTE: Using experimental fast data loading logic. To disable, pass\n",
            "    \"--load_fast=false\" and report issues on GitHub. More details:\n",
            "    https://github.com/tensorflow/tensorboard/issues/4784\n",
            "\n",
            "Serving TensorBoard on localhost; to expose to the network, use a proxy or pass --bind_all\n",
            "TensorBoard 2.11.2 at http://localhost:6006/ (Press CTRL+C to quit)\n"
          ]
        }
      ],
      "source": [
        "!tensorboard --logdir logs/mlp/hparam_tuning"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "GdKm1QXOb1TE",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "GdKm1QXOb1TE",
        "outputId": "2a8ae18f-f81a-491e-b3c3-1e289de85e40"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "([0.07673513144254684, 0.9819999933242798], <keras.engine.sequential.Sequential object at 0x7f32c9c5f160>)\n",
            "([0.2658335566520691, 0.9247000217437744], <keras.engine.sequential.Sequential object at 0x7f32c8d10250>)\n",
            "([0.07511485368013382, 0.9815999865531921], <keras.engine.sequential.Sequential object at 0x7f32c82ee130>)\n",
            "([0.27128732204437256, 0.9226999878883362], <keras.engine.sequential.Sequential object at 0x7f32d3cd7790>)\n",
            "([0.06496474891901016, 0.9833999872207642], <keras.engine.sequential.Sequential object at 0x7f32e40d2700>)\n",
            "([0.28695136308670044, 0.916700005531311], <keras.engine.sequential.Sequential object at 0x7f32d0146f70>)\n",
            "([0.07120592892169952, 0.9807000160217285], <keras.engine.sequential.Sequential object at 0x7f32c812f2b0>)\n",
            "([0.30975988507270813, 0.9107999801635742], <keras.engine.sequential.Sequential object at 0x7f32c073f9a0>)\n",
            "([0.07213472574949265, 0.9807000160217285], <keras.engine.sequential.Sequential object at 0x7f32e358dbb0>)\n",
            "([0.3490550220012665, 0.901199996471405], <keras.engine.sequential.Sequential object at 0x7f32e3442a90>)\n",
            "([0.07659175992012024, 0.9787999987602234], <keras.engine.sequential.Sequential object at 0x7f32e31d2640>)\n",
            "([0.381436824798584, 0.8945000171661377], <keras.engine.sequential.Sequential object at 0x7f32e3000430>)\n"
          ]
        }
      ],
      "source": [
        "for m in models:\n",
        "  print(m)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "0oN7aP1rGNZZ",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "0oN7aP1rGNZZ",
        "outputId": "6a30474b-bbcd-439b-8e26-09a67ce96ecc"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "([0.06496474891901016, 0.9833999872207642],\n",
              " <keras.engine.sequential.Sequential at 0x7f32e40d2700>)"
            ]
          },
          "execution_count": 117,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "models[4]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "YBE3AdAjdcEn",
      "metadata": {
        "id": "YBE3AdAjdcEn"
      },
      "outputs": [],
      "source": [
        "best_model=models[0][1]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "dTD1vQXhVdSQ",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dTD1vQXhVdSQ",
        "outputId": "29980304-cf45-44c0-e1a1-cb2d18d4f2d9"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "<keras.engine.sequential.Sequential at 0x7f32c9c5f160>"
            ]
          },
          "execution_count": 103,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "best_model\n"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "Z2lsrZUQHaMx",
      "metadata": {
        "id": "Z2lsrZUQHaMx"
      },
      "source": [
        "with the previous method we identified that the best hyperparameter for the optimizer and the dropout are : for trial run-4\n",
        "{'dropout': 0.3, 'optimizer': 'adam'}"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "q-RniQ6IHqZX",
      "metadata": {
        "id": "q-RniQ6IHqZX"
      },
      "source": [
        "we need now to find the best learning rate and the regularization"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 176,
      "id": "ZNZAIJZnNgDA",
      "metadata": {
        "id": "ZNZAIJZnNgDA"
      },
      "outputs": [],
      "source": [
        "from keras import regularizers"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 177,
      "id": "ijUaHjmGHZcd",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "ijUaHjmGHZcd",
        "outputId": "20a327d8-8b07-4e3d-a937-fed671e60c1a"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"sequential_25\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " dense_100 (Dense)           (None, 256)               200960    \n",
            "                                                                 \n",
            " dropout_60 (Dropout)        (None, 256)               0         \n",
            "                                                                 \n",
            " dense_101 (Dense)           (None, 192)               49344     \n",
            "                                                                 \n",
            " dropout_61 (Dropout)        (None, 192)               0         \n",
            "                                                                 \n",
            " dense_102 (Dense)           (None, 128)               24704     \n",
            "                                                                 \n",
            " dropout_62 (Dropout)        (None, 128)               0         \n",
            "                                                                 \n",
            " dense_103 (Dense)           (None, 10)                1290      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 276,298\n",
            "Trainable params: 276,298\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "Epoch 1/90\n",
            "110/110 [==============================] - 4s 25ms/step - loss: 3.6608 - accuracy: 0.7417 - val_loss: 1.1918 - val_accuracy: 0.9294\n",
            "Epoch 2/90\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 1.0034 - accuracy: 0.8859 - val_loss: 0.6640 - val_accuracy: 0.9430\n",
            "Epoch 3/90\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.7496 - accuracy: 0.9016 - val_loss: 0.5687 - val_accuracy: 0.9502\n",
            "Epoch 4/90\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.6751 - accuracy: 0.9105 - val_loss: 0.5258 - val_accuracy: 0.9520\n",
            "Epoch 5/90\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.6383 - accuracy: 0.9159 - val_loss: 0.5015 - val_accuracy: 0.9566\n",
            "Epoch 6/90\n",
            "110/110 [==============================] - 3s 32ms/step - loss: 0.6139 - accuracy: 0.9200 - val_loss: 0.4864 - val_accuracy: 0.9552\n",
            "Epoch 7/90\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.5953 - accuracy: 0.9227 - val_loss: 0.4699 - val_accuracy: 0.9596\n",
            "Epoch 8/90\n",
            "110/110 [==============================] - 4s 36ms/step - loss: 0.5734 - accuracy: 0.9253 - val_loss: 0.4553 - val_accuracy: 0.9600\n",
            "Epoch 9/90\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.5667 - accuracy: 0.9253 - val_loss: 0.4507 - val_accuracy: 0.9608\n",
            "Epoch 10/90\n",
            "110/110 [==============================] - 4s 34ms/step - loss: 0.5483 - accuracy: 0.9292 - val_loss: 0.4350 - val_accuracy: 0.9642\n",
            "Epoch 11/90\n",
            " 95/110 [========================>.....] - ETA: 0s - loss: 0.5373 - accuracy: 0.9301"
          ]
        },
        {
          "ename": "KeyboardInterrupt",
          "evalue": "ignored",
          "output_type": "error",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-177-09648a28055d>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     18\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbuild\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     19\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msummary\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 20\u001b[0;31m model.fit(train_X, train_y, batch_size=500, verbose=1,\n\u001b[0m\u001b[1;32m     21\u001b[0m                                     validation_data=(val_X, val_y), epochs=90, callbacks=[\n\u001b[1;32m     22\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py\u001b[0m in \u001b[0;36merror_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m     63\u001b[0m         \u001b[0mfiltered_tb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     64\u001b[0m         \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 65\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     66\u001b[0m         \u001b[0;32mexcept\u001b[0m \u001b[0mException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     67\u001b[0m             \u001b[0mfiltered_tb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_process_traceback_frames\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__traceback__\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)\u001b[0m\n\u001b[1;32m   1648\u001b[0m                         ):\n\u001b[1;32m   1649\u001b[0m                             \u001b[0mcallbacks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mon_train_batch_begin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mstep\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1650\u001b[0;31m                             \u001b[0mtmp_logs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain_function\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0miterator\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1651\u001b[0m                             \u001b[0;32mif\u001b[0m \u001b[0mdata_handler\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshould_sync\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1652\u001b[0m                                 \u001b[0mcontext\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0masync_wait\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/util/traceback_utils.py\u001b[0m in \u001b[0;36merror_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    148\u001b[0m     \u001b[0mfiltered_tb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    149\u001b[0m     \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 150\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mfn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    151\u001b[0m     \u001b[0;32mexcept\u001b[0m \u001b[0mException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    152\u001b[0m       \u001b[0mfiltered_tb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_process_traceback_frames\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0me\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__traceback__\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/polymorphic_function/polymorphic_function.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwds)\u001b[0m\n\u001b[1;32m    878\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    879\u001b[0m       \u001b[0;32mwith\u001b[0m \u001b[0mOptionalXlaContext\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_jit_compile\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 880\u001b[0;31m         \u001b[0mresult\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwds\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    881\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    882\u001b[0m       \u001b[0mnew_tracing_count\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexperimental_get_tracing_count\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/polymorphic_function/polymorphic_function.py\u001b[0m in \u001b[0;36m_call\u001b[0;34m(self, *args, **kwds)\u001b[0m\n\u001b[1;32m    910\u001b[0m       \u001b[0;31m# In this case we have created variables on the first call, so we run the\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    911\u001b[0m       \u001b[0;31m# defunned version which is guaranteed to never create variables.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 912\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_no_variable_creation_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwds\u001b[0m\u001b[0;34m)\u001b[0m  \u001b[0;31m# pylint: disable=not-callable\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    913\u001b[0m     \u001b[0;32melif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_variable_creation_fn\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    914\u001b[0m       \u001b[0;31m# Release the lock early so that multiple threads can perform the call\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/polymorphic_function/tracing_compiler.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    132\u001b[0m       (concrete_function,\n\u001b[1;32m    133\u001b[0m        filtered_flat_args) = self._maybe_define_function(args, kwargs)\n\u001b[0;32m--> 134\u001b[0;31m     return concrete_function._call_flat(\n\u001b[0m\u001b[1;32m    135\u001b[0m         filtered_flat_args, captured_inputs=concrete_function.captured_inputs)  # pylint: disable=protected-access\n\u001b[1;32m    136\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/polymorphic_function/monomorphic_function.py\u001b[0m in \u001b[0;36m_call_flat\u001b[0;34m(self, args, captured_inputs, cancellation_manager)\u001b[0m\n\u001b[1;32m   1743\u001b[0m         and executing_eagerly):\n\u001b[1;32m   1744\u001b[0m       \u001b[0;31m# No tape is watching; skip to running the function.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1745\u001b[0;31m       return self._build_call_outputs(self._inference_function.call(\n\u001b[0m\u001b[1;32m   1746\u001b[0m           ctx, args, cancellation_manager=cancellation_manager))\n\u001b[1;32m   1747\u001b[0m     forward_backward = self._select_forward_and_backward_functions(\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/polymorphic_function/monomorphic_function.py\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, ctx, args, cancellation_manager)\u001b[0m\n\u001b[1;32m    376\u001b[0m       \u001b[0;32mwith\u001b[0m \u001b[0m_InterpolateFunctionError\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    377\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mcancellation_manager\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 378\u001b[0;31m           outputs = execute.execute(\n\u001b[0m\u001b[1;32m    379\u001b[0m               \u001b[0mstr\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msignature\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    380\u001b[0m               \u001b[0mnum_outputs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_num_outputs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/execute.py\u001b[0m in \u001b[0;36mquick_execute\u001b[0;34m(op_name, num_outputs, inputs, attrs, ctx, name)\u001b[0m\n\u001b[1;32m     50\u001b[0m   \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     51\u001b[0m     \u001b[0mctx\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mensure_initialized\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 52\u001b[0;31m     tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,\n\u001b[0m\u001b[1;32m     53\u001b[0m                                         inputs, attrs, num_outputs)\n\u001b[1;32m     54\u001b[0m   \u001b[0;32mexcept\u001b[0m \u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_NotOkStatusException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ],
      "source": [
        "model = tf.keras.models.Sequential([\n",
        "    keras.layers.Dense(units=256, activation='relu', input_dim=input_size,\n",
        "                kernel_regularizer=regularizers.l2(0.01)),\n",
        "    keras.layers.Dropout(0.3),\n",
        "    keras.layers.Dense(units=192, activation='relu',\n",
        "                kernel_regularizer=regularizers.l2(0.01)), \n",
        "    keras.layers.Dropout(0.3),\n",
        "    keras.layers.Dense(units=128, activation='relu',\n",
        "                kernel_regularizer=regularizers.l2(0.01)),\n",
        "    keras.layers.Dropout(0.3),\n",
        "    keras.layers.Dense(units=10, activation='softmax') #number of classes 10  \n",
        "  ])\n",
        "model.compile(\n",
        "      optimizer=keras.optimizers.Adam(learning_rate=0.001),\n",
        "      loss='categorical_crossentropy',\n",
        "      metrics=['accuracy'],\n",
        "  )\n",
        "model.build()\n",
        "model.summary()\n",
        "model.fit(train_X, train_y, batch_size=500, verbose=1,\n",
        "                                    validation_data=(val_X, val_y), epochs=90, callbacks=[\n",
        "        \n",
        "        early_stopping_cb,\n",
        "        early_stopping_mc\n",
        "    ]) \n",
        "  \n",
        "acc = model.evaluate(test_X, test_y,batch_size=500,)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "dhj3TKOuKN0T",
      "metadata": {
        "id": "dhj3TKOuKN0T"
      },
      "outputs": [],
      "source": [
        "acc"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "Bx5psxisOygy",
      "metadata": {
        "id": "Bx5psxisOygy"
      },
      "source": [
        "with regularisation the model performs poorly (under fitting). further optimisations should be done to have better parameters. the final model for now will not have regularisation."
      ]
    },
    {
      "cell_type": "markdown",
      "id": "ax2hlLyfLmBr",
      "metadata": {
        "id": "ax2hlLyfLmBr"
      },
      "source": [
        "after many trainings we find the best learning rate for this problem is the default value 0.001"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "ezPD11mDPaEi",
      "metadata": {
        "id": "ezPD11mDPaEi"
      },
      "source": [
        "#### Final model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 197,
      "id": "F4lFz9hMPXbh",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "F4lFz9hMPXbh",
        "outputId": "10d27d6c-af02-4aac-aee7-0d66bd4aef1f"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"sequential_28\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " dense_112 (Dense)           (None, 256)               200960    \n",
            "                                                                 \n",
            " dropout_69 (Dropout)        (None, 256)               0         \n",
            "                                                                 \n",
            " dense_113 (Dense)           (None, 192)               49344     \n",
            "                                                                 \n",
            " dropout_70 (Dropout)        (None, 192)               0         \n",
            "                                                                 \n",
            " dense_114 (Dense)           (None, 128)               24704     \n",
            "                                                                 \n",
            " dropout_71 (Dropout)        (None, 128)               0         \n",
            "                                                                 \n",
            " dense_115 (Dense)           (None, 10)                1290      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 276,298\n",
            "Trainable params: 276,298\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "Epoch 1/90\n",
            "110/110 [==============================] - 5s 30ms/step - loss: 0.6861 - accuracy: 0.7854 - val_loss: 0.1626 - val_accuracy: 0.9522\n",
            "Epoch 2/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.2486 - accuracy: 0.9275 - val_loss: 0.1097 - val_accuracy: 0.9680\n",
            "Epoch 3/90\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.1794 - accuracy: 0.9482 - val_loss: 0.0932 - val_accuracy: 0.9740\n",
            "Epoch 4/90\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.1424 - accuracy: 0.9578 - val_loss: 0.0865 - val_accuracy: 0.9744\n",
            "Epoch 5/90\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.1240 - accuracy: 0.9631 - val_loss: 0.0799 - val_accuracy: 0.9786\n",
            "Epoch 6/90\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.1060 - accuracy: 0.9690 - val_loss: 0.0707 - val_accuracy: 0.9782\n",
            "Epoch 7/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0930 - accuracy: 0.9719 - val_loss: 0.0681 - val_accuracy: 0.9810\n",
            "Epoch 8/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0851 - accuracy: 0.9743 - val_loss: 0.0671 - val_accuracy: 0.9798\n",
            "Epoch 9/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0747 - accuracy: 0.9779 - val_loss: 0.0666 - val_accuracy: 0.9812\n",
            "Epoch 10/90\n",
            "110/110 [==============================] - 4s 39ms/step - loss: 0.0708 - accuracy: 0.9785 - val_loss: 0.0641 - val_accuracy: 0.9810\n",
            "Epoch 11/90\n",
            "110/110 [==============================] - 3s 28ms/step - loss: 0.0650 - accuracy: 0.9797 - val_loss: 0.0635 - val_accuracy: 0.9838\n",
            "Epoch 12/90\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0596 - accuracy: 0.9817 - val_loss: 0.0615 - val_accuracy: 0.9836\n",
            "Epoch 13/90\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0509 - accuracy: 0.9837 - val_loss: 0.0589 - val_accuracy: 0.9824\n",
            "Epoch 14/90\n",
            "110/110 [==============================] - 3s 27ms/step - loss: 0.0523 - accuracy: 0.9838 - val_loss: 0.0622 - val_accuracy: 0.9820\n",
            "Epoch 15/90\n",
            "110/110 [==============================] - 3s 26ms/step - loss: 0.0500 - accuracy: 0.9845 - val_loss: 0.0544 - val_accuracy: 0.9838\n",
            "Epoch 16/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0460 - accuracy: 0.9854 - val_loss: 0.0592 - val_accuracy: 0.9830\n",
            "Epoch 17/90\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0410 - accuracy: 0.9869 - val_loss: 0.0544 - val_accuracy: 0.9842\n",
            "Epoch 18/90\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0397 - accuracy: 0.9869 - val_loss: 0.0589 - val_accuracy: 0.9824\n",
            "Epoch 19/90\n",
            "110/110 [==============================] - 3s 31ms/step - loss: 0.0395 - accuracy: 0.9876 - val_loss: 0.0620 - val_accuracy: 0.9838\n",
            "Epoch 20/90\n",
            "110/110 [==============================] - 3s 25ms/step - loss: 0.0361 - accuracy: 0.9883 - val_loss: 0.0604 - val_accuracy: 0.9836\n",
            "Epoch 21/90\n",
            "110/110 [==============================] - 3s 23ms/step - loss: 0.0357 - accuracy: 0.9886 - val_loss: 0.0590 - val_accuracy: 0.9844\n",
            "Epoch 22/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0364 - accuracy: 0.9887 - val_loss: 0.0623 - val_accuracy: 0.9832\n",
            "Epoch 23/90\n",
            "110/110 [==============================] - 2s 23ms/step - loss: 0.0322 - accuracy: 0.9894 - val_loss: 0.0581 - val_accuracy: 0.9836\n",
            "Epoch 24/90\n",
            "110/110 [==============================] - 4s 33ms/step - loss: 0.0329 - accuracy: 0.9897 - val_loss: 0.0643 - val_accuracy: 0.9850\n",
            "Epoch 25/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0310 - accuracy: 0.9901 - val_loss: 0.0637 - val_accuracy: 0.9838\n",
            "Epoch 26/90\n",
            "110/110 [==============================] - 2s 22ms/step - loss: 0.0293 - accuracy: 0.9905 - val_loss: 0.0633 - val_accuracy: 0.9834\n",
            "Epoch 27/90\n",
            "110/110 [==============================] - 2s 21ms/step - loss: 0.0274 - accuracy: 0.9911 - val_loss: 0.0611 - val_accuracy: 0.9844\n",
            "20/20 [==============================] - 0s 8ms/step - loss: 0.0643 - accuracy: 0.9848\n"
          ]
        }
      ],
      "source": [
        "model = tf.keras.models.Sequential([\n",
        "    keras.layers.Dense(units=256, activation='relu', input_dim=input_size\n",
        "                ),\n",
        "    keras.layers.Dropout(0.3),\n",
        "    keras.layers.Dense(units=192, activation='relu'), \n",
        "    keras.layers.Dropout(0.3),\n",
        "    keras.layers.Dense(units=128, activation='relu'),\n",
        "    keras.layers.Dropout(0.3),\n",
        "    keras.layers.Dense(units=10, activation='softmax') #number of classes 10  \n",
        "  ])\n",
        "model.compile(\n",
        "      optimizer=keras.optimizers.Adam(learning_rate=0.001),\n",
        "      loss='categorical_crossentropy',\n",
        "      metrics=['accuracy'],\n",
        "  )\n",
        "model.build()\n",
        "model.summary()\n",
        "history=model.fit(train_X, train_y, batch_size=500, verbose=1,\n",
        "                                    validation_data=(val_X, val_y), epochs=90, callbacks=[\n",
        "        \n",
        "        early_stopping_cb,\n",
        "        early_stopping_mc\n",
        "    ]) \n",
        "  \n",
        "acc = model.evaluate(test_X, test_y,batch_size=500,)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 198,
      "id": "OBuKm3dUQLJT",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OBuKm3dUQLJT",
        "outputId": "000caf4c-f245-49a7-df55-fb0dd3d17622"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "[0.0642763078212738, 0.9847999811172485]"
            ]
          },
          "execution_count": 198,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "acc"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "lZT-tOuTQbV9",
      "metadata": {
        "id": "lZT-tOuTQbV9"
      },
      "source": [
        "Final accuracy : 98.5 % and 0.06 loss , close to SOA"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "mset8SSzz1SN",
      "metadata": {
        "id": "mset8SSzz1SN"
      },
      "source": [
        "### Model validation"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "bSxlfKBeBHT9",
      "metadata": {
        "id": "bSxlfKBeBHT9"
      },
      "source": [
        "\n",
        "\n",
        "> accuracy of the best model on the test set is 98,3%\n",
        "\n",
        "\n",
        "\n",
        "> the below figure is the confusion matrix\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 199,
      "id": "ocqSj3m301uI",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 297
        },
        "id": "ocqSj3m301uI",
        "outputId": "ff653a21-9edf-4690-c749-98a807e1bf06"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "313/313 [==============================] - 1s 3ms/step\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "best_model=model\n",
        "pred_y=best_model.predict(test_X)\n",
        "pred_y=np.argmax(pred_y, axis=1)\n",
        "test_y=np.argmax(test_y, axis=1)\n",
        "cm = confusion_matrix(test_y, pred_y)\n",
        "disp = ConfusionMatrixDisplay(confusion_matrix=cm)\n",
        "disp.plot()\n",
        "pyplot.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 200,
      "id": "fMC4hI4iCBrD",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fMC4hI4iCBrD",
        "outputId": "bf6d6336-bc3e-4b1d-d70d-1ceaa4e30281"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Classification report for classifier <keras.engine.sequential.Sequential object at 0x7f32d64cbf10>:\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.99      0.99      0.99       980\n",
            "           1       1.00      0.99      0.99      1135\n",
            "           2       0.99      0.99      0.99      1032\n",
            "           3       0.98      0.98      0.98      1010\n",
            "           4       0.98      0.99      0.98       982\n",
            "           5       0.98      0.98      0.98       892\n",
            "           6       0.98      0.99      0.98       958\n",
            "           7       0.98      0.99      0.98      1028\n",
            "           8       0.99      0.97      0.98       974\n",
            "           9       0.99      0.98      0.98      1009\n",
            "\n",
            "    accuracy                           0.98     10000\n",
            "   macro avg       0.98      0.98      0.98     10000\n",
            "weighted avg       0.98      0.98      0.98     10000\n",
            "\n",
            "\n"
          ]
        }
      ],
      "source": [
        "print(\n",
        "    f\"Classification report for classifier {model}:\\n\"\n",
        "    f\"{metrics.classification_report(test_y, pred_y)}\\n\"\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 201,
      "id": "BGNpHRnx2Dci",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 800
        },
        "id": "BGNpHRnx2Dci",
        "outputId": "783a1da0-0b58-4398-feb2-a2a9899e0f35"
      },
      "outputs": [
        {
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAnoAAAGFCAYAAABqhl5dAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/NK7nSAAAACXBIWXMAAAsTAAALEwEAmpwYAABHF0lEQVR4nO3deZgU1dn+8e/Dvo2A7ArIDCLgCjggERdAxS0ak2g0iWjMYoy7r75XTEyiJvHVJL8oatwTl2iMURMlJi64gLjgAoqKsomALCKI7JvAPL8/TrfTDN0z0z09U9U99+e66qrpqurqp6dpuDlV5xxzd0RERESk+DSJugARERERqR8KeiIiIiJFSkFPREREpEgp6ImIiIgUKQU9ERERkSKloCciIiJSpJpFXUAcde7c2fv06RN1GSIiIiI1mjZt2mfu3iXdPgW9NPr06cPUqVOjLkNERESkRma2MNM+XboVERERKVIKeiIiIiJFSkFPREREpEjpHj0REREpaFu3bmXx4sVs3rw56lLqVatWrejZsyfNmzev9XMU9ERERKSgLV68mJKSEvr06YOZRV1OvXB3Vq5cyeLFiyktLa3183TpVkRERAra5s2b6dSpU9GGPAAzo1OnTlm3WiroiYiISMEr5pCXlMt7VNATERERqYPVq1dz6623Zv284447jtWrV+e/oBQKeiIiIiJ1kCnobdu2rdrnPfnkk3To0KGeqgrUGUNERESkDi6//HLmzZvHoEGDaN68Oa1ataJjx47MmjWLOXPmcNJJJ7Fo0SI2b97MRRddxNlnnw1UzsS1fv16jj32WA455BBeffVVdt99d8aPH0/r1q3rXJuCnoiIiBSPiy+G6dPze85Bg2DcuIy7r7vuOmbMmMH06dOZNGkSxx9/PDNmzPiyd+zdd9/NrrvuyqZNmxg6dCjf/OY36dSp0w7nmDt3Ln//+9+56667+Na3vsU///lPTj/99DqXrku3UVixAp54AjZsiLoSERERybNhw4btMATKTTfdxAEHHMDw4cNZtGgRc+fO3ek5paWlDBo0CIADDzyQBQsW5KUWtehF4eWX4RvfgGnTYMiQqKsREREpHtW0vDWUtm3bfvnzpEmTeO6555gyZQpt2rRh5MiRaYdIadmy5Zc/N23alE2bNuWlFrXoRaGsLKw/+ijaOkRERKTOSkpKWLduXdp9a9asoWPHjrRp04ZZs2bx2muvNWhtatGLQrI5V0FPRESk4HXq1IkRI0aw77770rp1a7p16/blvmOOOYbbb7+dgQMH0r9/f4YPH96gtSnoRWGXXaBzZwU9ERGRIvHggw+m3d6yZUueeuqptPuS9+F17tyZGTNmfLn9sssuy1tdunQblbIyBT0RERGpVwp6UVHQExERkXqmoBeVsjJYuBC2b4+6EhERESlSsQl6ZtbTzO42s6VmtsXMFpjZODPrmOV5DjGz8Ynnbzazj83sSTM7pr5qz0lZGWzbBosXR12JiIiIFKlYBD0z6wtMA84C3gBuAD4CLgKmmFmnap6eep6fAC8BRyTWNwAvAocDT5nZFfmvPkfqeSsiIiL1LBZBD7gV6Apc6O4nufvl7j6aENT6A9fUdAIzaw5cC2wGDnT3se7+M3cfC5QDW4ArzKxldedpMBpLT0REROpZ5EEv0Zo3BlgA3FJl95XABmCsmbWlersC7YE57j47dYe7zwTmAK2Bdnkou+569oRmzRT0REREGpl27RouikQe9IBRifUEd69I3eHu64BXgDZATSMMLgdWAHuZWb/UHWa2F9APmO7uK/NSdV01awZ77KGgJyIiIvUmDgMm90+s52TYP5fQ4rcX8Hymk7i7m9l5wAPANDN7DFgK7A58HXgfOC1fReeFhlgREREpeJdffjm9evXivPPOA+Cqq66iWbNmTJw4kVWrVrF161Z++9vf8rWvfa3Ba4tD0GufWK/JsD+5vUNNJ3L3R8xsKfB34IyUXZ8C9xA6eKRlZmcDZwP07t27ppfKj7Iy+Ne/Gua1REREGoGLL4bp0/N7zkGDYNy4zPtPPfVULr744i+D3sMPP8wzzzzDhRdeyC677MJnn33G8OHDOfHEEzGz/BZXgzhcus0bMzsdeI7Q43Yg4ZLvQEJL4J+AhzI9193vdPdydy/v0qVLQ5Qbgt6KFZBhImQRERGJv8GDB7N8+XKWLl3KO++8Q8eOHenevTs///nP2X///TnyyCNZsmQJn376aYPXFocWvWSLXfsM+5PbV1d3ksR9eHcD7wJjU+73m2VmYwmXiE8xs5HuPqlOFedLcoiV+fNh//2jrUVERKQIVNfyVp9OOeUUHn30UZYtW8app57K3/72N1asWMG0adNo3rw5ffr0YfPmzQ1eVxxa9JI9ZPfKsD/ZsSLTPXxJY4DmwItpOnVUAJMTDw/Mpch6oSFWREREisKpp57KQw89xKOPPsopp5zCmjVr6Nq1K82bN2fixIksXLgwkrri0KI3MbEeY2ZNUkOamZUAI4CNwGs1nCc5Pl6m667J7V/kWmjeKeiJiIgUhX322Yd169ax++6706NHD7773e9ywgknsN9++1FeXs6AAQMiqSvyoOfu88xsAqFF7jzg5pTdVwNtgTvcfUNyo5kNSDx3VsqxLyXWJ5vZ/3P3d1OOHwScDDjwQn28j5x07AgdOijoiYiIFIH33nvvy587d+7MlClT0h63fv36hiop+qCXcC7wKnCTmR0BzAQOIoyxNweoOnXZzMT6y64r7v6Gmd1DmEbtzcTwKguBPsBJQAtgnLu/X39vIwcaYkVERETqSSyCXqJVrxz4NXAMcBzwCXAjcLW7r6rlqX5AuBfve8DRQAmwFngZuMvdM/a6jUxZGaT8D0BEREQkX2IR9ADcfRGhNa42x6YdhMbdHbg3sRSGsjJ44gmoqIAmcegbIyIiIsVCySJqZWWwZQt88knUlYiIiBSs0NZT3HJ5jwp6UUuOpaf79ERERHLSqlUrVq5cWdRhz91ZuXIlrVq1yup5sbl022ilDrFy6KHR1iIiIlKAevbsyeLFi1mxYkXUpdSrVq1a0bNnz6yeo6AXtd69w715atETERHJSfPmzSlNXiGTHejSbdRatIBevRT0REREJO8U9OJAY+mJiIhIPVDQi4OyMpg/P+oqREREpMgo6MVBWVkYXmXjxqgrERERkSKioBcHyRtIFyyItAwREREpLgp6cZA6xIqIiIhInijoxYGCnoiIiNQDBb046NwZ2rVT0BMREZG8UtCLAzMNsSIiIiJ5p6AXFwp6IiIikmcKenGRHEuviCdkFhERkYaloBcXZWVhHL3ly6OuRERERIqEgl5cJMfS0+VbERERyRMFvbjQECsiIiKSZwp6cdGnT1gr6ImIiEieKOjFRatWsPvuCnoiIiKSNwp6caIhVkRERCSPFPTiJDnEioiIiEgeKOjFSVkZLF4MW7ZEXYmIiIgUAQW9OCktDQMmL1wYdSUiIiJSBBT04kRDrIiIiEgeKejFiYKeiIiI5JGCXpx07x6GWVHQExERkTxQ0IsTMw2xIiIiInmjoBc3CnoiIiKSJwp6cZMcS8896kpERESkwMUm6JlZTzO728yWmtkWM1tgZuPMrGMtnz/SzLwWS6/6fi91UloKa9fC559HXYmIiIgUuGZRFwBgZn2BV4GuwHhgFjAMuAg4xsxGuPvKGk6zALg6w779gG8AM9x9UV6Kri+pPW87dYq2FhERESlosQh6wK2EkHehu9+c3Ghm1wOXANcA51R3AndfAFyVbp+Z/T3x4115qLV+pQa9oUOjrUVEREQKWuSXbhOteWMILXK3VNl9JbABGGtmbXM8f2fg68Am4K+5V9pASkvDWh0yREREpI4iD3rAqMR6grtXpO5w93XAK0AbYHiO5z8TaAk84u6rcy2ywbRtC926KeiJiIhIncUh6PVPrOdk2D83sd4rx/P/KLG+I8fnNzwNsSIiIiJ5EIeg1z6xXpNhf3J7h2xPbGaHE4LkDHd/tYZjzzazqWY2dcWKFdm+VH4p6ImIiEgexCHo1aezE+s7azrQ3e9093J3L+/SpUs9l1WDsjJYtAi2bo22DhERESlocQh6yRa79hn2J7evzuakZrYr8E1CJ4z7c6osKqWlsH17CHsiIiIiOYpD0JudWGe6B69fYp3pHr5Mkp0wHi6IThipUodYEREREclRHILexMR6jJntUI+ZlQAjgI3Aa1meN9kJo8bLtrGjoCciIiJ5EHnQc/d5wASgD3Beld1XA22B+919Q3KjmQ0wswGZzmlmhwIDqUUnjFjabTdo0UJBT0REROokLjNjnEuYAu0mMzsCmAkcRBhjbw5wRZXjZybWluF8te6EEUtNm0KfPgp6IiIiUieRt+jBl6165cC9hIB3KdAXuBEYXot5br9kZh2BkynEThipNMSKiIiI1FFcWvRw90XAWbU8NlNLHu6+Cmidr7oiU1YGb7wRdRUiIiJSwGLRoidplJbC55/D6tVRVyIiIiIFSkEvrpI9b+fPj7YOERERKVgKenGlIVZERESkjhT04qq0NKwV9ERERCRHCnpx1b49dOqkoCciIiI5U9CLMw2xIiIiInWgoBdnCnoiIiJSBwp6cVZWBgsXwvbtUVciIiIiBUhBL85KS2HrVliyJOpKREREpAAp6MWZhlgRERGROlDQizMFPREREakDBb0469ULmjZV0BMREZGcKOjFWbNmsMceCnoiIiKSEwW9uNMQKyIiIpIjBb24KyuD+fOjrkJEREQKkIJe3JWWwvLlsH591JWIiIhIgVHQi7tkz1u16omIiEiWFPTiTkOsiIiISI4U9OJOQU9ERERypKAXdx07Qvv2CnoiIiKSNQW9uDPTECsiIiKSEwW9QqCgJyIiIjlQ0CsEZWWwYAFUVERdiYiIiBQQBb1CUFoKmzfDsmVRVyIiIiIFREGvEKjnrYiIiORAQa8QKOiJiIhIDhT0CsEee4Tetwp6IiIikgUFvULQogX06qWgJyIiIllR0CsUGmJFREREsqSgVyjKymD+/KirEBERkQISm6BnZj3N7G4zW2pmW8xsgZmNM7OOOZxriJk9aGaLE+f61MxeNLMz6qP2BlFaCkuXwqZNUVciIiIiBSIWQc/M+gLTgLOAN4AbgI+Ai4ApZtYpi3OdD7wJjAGeB/4IPAY0BY7Lb+UNKNnzdsGCSMsQERGRwtEs6gISbgW6Ahe6+83JjWZ2PXAJcA1wTk0nMbMxwE3As8DJ7r6uyv7m+Sy6QaUOsTJwYLS1iIiISEGIvEUv0Zo3BlgA3FJl95XABmCsmbWtxen+AGwCvlM15AG4+9a6VRshjaUnIiIiWYpDi96oxHqCu+8wmau7rzOzVwhBcDjhUmxaZrYvsD/wOPC5mY0CDgQcmA5MrHr+gtKlC7Rtq6AnIiIitRaHoNc/sZ6TYf9cQtDbi2qCHjA0sV4OTAIOq7L/PTP7hrt/mGOd0TLTECsiIiKSlcgv3QLtE+s1GfYnt3eo4TxdE+sfAH2A4xPn3gt4ANgP+K+ZtUj3ZDM728ymmtnUFStW1K7yhqagJyIiIlmIQ9DLl+R7aQqc5u5Puvtad58LnAFMJYS+b6Z7srvf6e7l7l7epUuXhqk4W8mx9NyjrkREREQKQByCXrLFrn2G/cntq2s4T3L/MnefkrrD3R0Yn3g4LMv64qO0FDZsgLi2OIqIiEisxCHozU6s98qwv19inekevqrnWZ1h/6rEunXtyooh9bwVERGRLMQh6E1MrMeY2Q71mFkJMALYCLxWw3leIwzF0ifDUCz7JtaFO4+Ygp6IiIhkIfKg5+7zgAmEDhTnVdl9NdAWuN/dNyQ3mtkAMxtQ5Twbgb8ArYDfmpmlHL8f8D1gG/Bo/t9FA+nTJ6wV9ERERKQW4jC8CsC5wKvATWZ2BDATOIgwxt4c4Ioqx89MrK3K9l8ShlW5GPhKYgy+bsA3CAHw4kSwLEytW8NuuynoiYiISK1E3qIHX7bqlQP3EgLepUBf4EZguLuvrOV51gKHAv8H7AqcD3wVeBk42t1vzHvxDU1DrIiIiEgtxaVFD3dfBJxVy2OrtuSl7ltPaAGs2gpYHMrKYNKkqKsQERGRAhCLFj3JQmkpLFoEX3wRdSUiIiIScwp6haasLAyYvHBh1JWIiIhIzCnoFRoNsSIiIiK1pKBXaBT0REREpJYU9ApN9+7QqpWCnoiIiNRIQa/QNGkSOmQo6ImIiEgNFPQKkcbSExERkVpQ0CtEyaDnHnUlIiIiEmMKeoWotBTWroVVq6KuRERERGJMQa8QqeetiIiI1IKCXiFS0BMREZFaUNArRKWlYa2gJyIiItVQ0CtE7dpB164KeiIiIlItBb1CpSFWREREpAYKeoWqrAzmz4+6ChEREYkxBb1CVVoKCxfCtm1RVyIiIiIxpaBXqMrKYPt2WLQo6kpEREQkphT0CpWGWBEREZEaKOgVKgU9ERERqUHWQc/MOprZ3mbWssr2s8xsvJk9aGbD8leipLX77tC8uYKeiIiIZNQsh+f8H3A60DW5wcwuAMYBlth0kpmVu/sHda5Q0mvaFPr0UdATERGRjHK5dDsCeN7dN6VsuwxYAhwGfCux7X/qWJvURGPpiYiISDVyCXq7A18O4GZmewO9gJvd/WV3fxR4ghD6pD5pLD0RERGpRi5BrzWwOeXxCMCB51K2zSMEQqlPpaWwciWsWRN1JSIiIhJDuQS9JcCAlMdHA2uBd1K2dQRSL+1KfUj2vFWrnoiIiKSRS9CbCBxnZueb2Q+BE4Gn3b0i5Zi+gEbyrW8aYkVERESqkUvQuxZYD9wI3Em4jHtVcqeZ7QIcAryah/qkOgp6IiIiUo2sh1dx9/lmtg9wcmLTv93945RD9gTuAB7MQ31SnfbtYdddFfREREQkrVzG0cPdlwF/yrDvLeCtuhQlWdAQKyIiIpJB3qZAM7NOZvZ1MzvazJrm8PyeZna3mS01sy1mtsDMxplZxyzOMcnMvJqlVbZ1xZ6GWBEREZEMsm7RM7OfAN8DjnX3zxPbDgSeBnZNHDbVzEa7+4ZanrMv4Z6+rsB4YBYwDLgIOMbMRrj7yizKvDrD9m1ZnKMwlJbC44/D9u1htgwRERGRhFwu3Z4KeDLkJfyBMKTKPUA34HjgHOCPtTznrYSQd6G735zcaGbXA5cA1yTOVyvuflVtjy14ZWXwxRewdCn06hV1NSIiIhIjuVy67Qe8m3xgZp2Bw4G/uPsP3f0E4E3gO7U5WaI1bwywALilyu4rgQ3AWDNrm0OtxU89b0VERCSDXIJeJ2B5yuMRifVjKdteAvao5flGJdYTqozFh7uvA14B2gDDa1ugmZ1qZpeb2f+Y2bFm1rK2zy04CnoiIiKSQS6Xbj8HOqc8PhyoYMdx8xyobceH/on1nAz75xJa/PYCnq/lOR+q8ni5mZ2XmIe3uPTqFe7NU9ATERGRKnJp0ZsJnJDoZdsBOA14093XphzTB1hWy/O1T6wzTdia3N6hFucaD5wA9CTMyTuAMMBzB+AfZnZMpiea2dlmNtXMpq5YsaIWLxUTzZtD794wa1bUlYiIiEjM5BL0bgR6AIsJ05x1I3SmSDWcHee+bRDufoO7/8fdl7j7Znef7e4/By4lvNdrq3nune5e7u7lXbp0abCa8+Loo0PPW4U9ERERSZF10HP3fxN6wL4PzAYuc/cHkvvNbCTQDnimlqdMtti1z7A/uX11lqWm+jNhaJVBZlZSh/PE09VXQ9u2cMkl4B51NSIiIhITOQ2YnNr65e43VNk3yd07uvudtTzd7MR6rwz7+yXWme7hq5G7bwbWJR4WX+/drl3hyivh6afhv/+NuhoRERGJibzNjFEHExPrMWa2Qz2J1rcRwEbgtVxfwMz6E8b5Wwd8lut5Yu3882HAgNCqt2VL1NWIiIhIDOQc9MxsuJn92cymmdk8M3vLzO4ys4OzOY+7zwMmEDpwnFdl99WEFrj7U2fZMLMBZjagSj2lZrZrledjZl0IAzkDPOTuxTc7BoROGePGwYcfwk03RV2NiIiIxIB5Dvd0mdlvgZ8Blma3A79LdIKo7fmqToE2EziIMMbeHODg1CnQzMwB3N1Stn0PuB14GfiIMAxMb+A4wn1+U4Gj3H11TfWUl5f71KlTa1t+vJx4IkyaBHPmQPfuUVcjIiIi9czMprl7ebp9WbfomdkpwM+Bj4EfAmWEoUzKEo8/Bn5qZt+q7TkTrXrlwL2EgHcp0JfQw3d4Lee5nUYYP68b8M3EOY4B3gMuBEbUJuQVvD/+ETZvhp/9LOpKREREJGJZt+iZ2WRCB4n93H2n+90SU6LNAGa7++F5qbKBFXSLHsBPfwq//z28/joMGxZ1NSIiIlKP8tqiBxwAPJou5AEktj8CDMrh3JIPv/hFuGx74YVQUVHz8SIiIlKUcgl6zQi9YKuzkdymV5N8KCmB664LLXp/+1vU1YiIiEhEcgl684CvVh0KJSmx/bjEcRKVsWPDZduf/hTWrav5eBERESk6uQS9B4GBwHgz65e6I9F79lFg78RxEpUmTcIwK598Av/3f1FXIyIiIhHIJehdD0wGjgdmmtnHZva6mS0kzHJxEvBK4jiJ0kEHwRlnwPXXwzw1sIqIiDQ2ucx1+wVwFHAFMB/oCQwFeiUeXwEckThOonbdddCiBVx6adSViIiISAPLda7bre5+rbv3A3YhhLxd3L2fu18LNDWzXfJZqOSoR4/QC3f8eHj22airERERkQZU57lu3X29uy9x9/Upm28jzEwhcXDxxdC3L1x0EWzdGnU1IiIi0kDqHPSqkW56NIlCy5bhPr2ZM+G226KuRkRERBpIfQY9iZMTToAxY+DKK2HFiqirERERkQagoNdYmMG4cWFMvV/+MupqREREpAEo6DUmAwfC+efDnXfC9OlRVyMiIiL1TEGvsbnqKujUKXTMcI+6GhEREalHCnqNTYcOcM01MHkyPPJI1NWIiIhIPTKvRauOmW3P5eTu3jSX50WtvLzcp06dGnUZ9Wf7dhg6FD77DGbNgjZtoq5IREREcmRm09y9PN2+2rboWQ6LxFXTpnDjjbBoEfz+91FXIyIiIvWkVkHP3ZvksBRka16jceihcOqp8LvfwccfR12NiIiI1APdo9eY/eEPYdiV//3fqCsRERGReqCg15j16gWXXw4PPwwvvhh1NSIiIpJnCnqN3f/+L/TuHYZb2Z5TnxsRERGJKQW9xq51a/h//w/eeQf+/OeoqxEREZE8UtATOPlkOPxwuOIKWLUq6mpEREQkTxT0JHTIuOmmEPKuuirqakRERCRPFPQk2H9/+PGP4ZZb4P33o65GRERE8kBBTyr9+tdQUgKXXKJ5cEVERIqAgp5U6tw5hL1nn9WMGSIiIkWgWdQFSMycey689loYX6+kJDwWERGRgqSgJztq2hTuvRfWr4fzzgthb+zYqKsSERGRHOjSreyseXP4xz/giCPgrLPgsceirkhERERyoKAn6bVqBY8/DsOGwWmnwYQJUVckIiIiWYpN0DOznmZ2t5ktNbMtZrbAzMaZWcc6nPMwM9tuZm5mv81nvY1Cu3bw5JOw995w0knw8stRVyQiIiJZiEXQM7O+wDTgLOAN4AbgI+AiYIqZdcrhnCXAfcDGPJba+HToAM88E+bDPf54eOutqCsSERGRWopF0ANuBboCF7r7Se5+ubuPJgS+/sA1OZzzRqA9cG3+ymykunYNQ6507AhjxsAHH0RdkYiIiNRC5EEv0Zo3BlgA3FJl95XABmCsmbXN4pxfI7QOXggszU+l+eMOb74ZdRVZ6tULnnsudNQ46ij46KOoKxIREZEaRB70gFGJ9QR3r0jd4e7rgFeANsDw2pzMzLoCdwGPu/sD+Sw0Xx54IPRxeOGFqCvJ0p57hpa9zZvhyCNhyZKoKxIREZFqxCHo9U+s52TYPzex3quW57uL8L7OqUtR9enkk6G0NAxT98UXUVeTpX33haefhs8+Cy17K1ZEXZGIiIhkEIeg1z6xXpNhf3J7h5pOZGbfB04EznX3T7MpwszONrOpZjZ1RT2Hl9at4eabYdYsuOGGen2p+jF0KPznPzB/Phx9NKzJ9NGJiIhIlOIQ9PLCzPoA44BH3P3hbJ/v7ne6e7m7l3fp0iXf5e3k+OPha18LU8t+/HG9v1z+HXZYGEh5xozwZjZsiLoiERERqSIOQS/ZHNQ+w/7k9tU1nOduYBNQMJOz3nhj6JhxySVRV5KjY46BBx+EKVPg61+HLVuirkhERERSxCHozU6sM92D1y+xznQPX9IQwhAtKxIDJLuZOXBPYv8ViW2P16naPNpjD/jFL+Bf/wq3vRWkk0+Gv/wldNI47TTYti3qikRERCTB3D3aAsLwKh8Shlfpm9rzNjHo8SeAAV3dPeP1QTO7idA7t6p+wGHAdMKgzG+7e9VhXHZQXl7uU6dOze6N5GjLFth/f9i+PVwFbdWqQV42/26+GS68EE4/He67D5rE4f8QIiIixc/Mprl7ebp9zRq6mKrcfZ6ZTSCMpXcecHPK7quBtsAdqSHPzAYknjsr5TwXpju/mX2PEPT+6+6/yPsbqKOWLeGWW0IH1t//Hn71q6grytEFF8C6dXDFFVBSEt6UWdRViYiINGqRB72Ec4FXgZvM7AhgJnAQYYy9OcAVVY6fmVgXRZI48kj41rfg2mtDg1hZWdQV5ehnPws9cH//+xD2rrtOYU9ERCRCsbi+5u7zgHLgXkLAuxToS5jGbLi7r4yuuoZx/fXQrFm4+hnx1fTcmYVwd845Iexdq9nnREREohSXFj3cfRFh2rLaHFvrZiJ3v5cQIGNt993hqqvgssvg3/8OQ68UJLNw2Xb9+srLuBdcEHVVIiIijVIsWvQkuPBC2GcfuOgi2Lgx6mrqoEkTuOceOOmk8KbGjQu9TURERKRBKejFSPPmcOutsHAhXHNN1NXUUbNm8NBDYTDlSy4Jk/u+/HLUVYmIiDQqCnoxc9hhMHYs/OEPMHt2zcfHWsuW8MQTYVDl5cvh0EPhO9+BRYuirkxERKRRUNCLoT/8Adq0gfPPL+COGUlm8O1vh4l9f/WrMG3agAHwm9/Apk1RVyciIlLUFPRiqFs3+O1v4bnn4JFHoq4mT9q2hauvhpkzw+XcX/0KBg6ERx8tgjQrIiISTwp6MfWTn8DgweH2tnXroq4mj/r0gYcfhokToX17OOUUGD0a3n036spERESKjoJeTDVtGjpmLF0aGsKKzsiRMG0a3HYbvPdeSLU/+Ql89lnUlYmIiBQNBb0YGz4cfvjDMDrJjBlRV1MPmjULgyvPnRtuSLzrLujXD266CbZujbo6ERGRgqegF3PXXhuucJ53XhHfytaxI9x4I7zzDpSXh4EEBw2CZ5+NujIREZGCpqAXc507w+9+B5MnwwMPRF1NPdtnH5gwAcaPhy1bYMyYMOjyvHlRVyYiIlKQFPQKwPe/Hy7jXnYZrF4ddTX1zAxOPBHefz/Mm/v887D33vCznxVZrxQREZH6p6BXAJo0CR0zPvsMfvnLqKtpIC1bwk9/GkaN/va3Q+jr3x/++leoqIi6OhERkYKgoFcgBg+Gc88Nge+tt6KupgHtthvcey+89hr06gVnnhmmD9FwLCIiIjVS0Csgv/kNdOkSAl+ja9Q66CCYMgXuvju08g0ZAv/zP7B2bdSViYiIxJaCXgHp0CFMj/b66yHvNDpNmsBZZ4Wglxx3ZsAA+Mc/irhLsoiISO4U9ArM6aeHK5eXXw4rV0ZdTUR23RVuvz1czu3RA047LfTQnT076spERERiRUGvwJjBLbeE3rc/+1nU1URs2DB44w3405/gzTdhv/3gF7+AjRujrkxERCQWFPQK0L77wsUXw5//HC7jNmpNm4bRpJO9c6+5JgzH8sQTUVcmIiISOQW9AnXlleGq5bnnwvbtUVcTA926wX33waRJ0LZtGIvvxBNhwYKoKxMREYmMgl6BKimBG24IQ63cfnvU1cTI4YfD9Omh18oLL4TWvWuuCTNtiIiINDIKegXslFPgyCPhiivg00+jriZGmjcP04jMmgXHHx/u29t/f3juuagrExERaVAKegXMLPRD2LgRvvlNdTrdSc+e8Mgj8PTTYeDBo44KPXSXLo26MhERkQahoFfg+veHv/wF3nsvdDq9/HJYvz7qqmLm6KPDL+jqq+Hxx8PYezfcANu2RV2ZiIhIvVLQKwJjx8KcOfDd78LvfgcDB8LDD2sM4R20agW/+hW8/z4cemiYVWPIEJg4Ub8oEREpWgp6RaJbN7jnHnjllTBN2qmnhvv3Pvgg6spipm9f+M9/4F//CoMRjh4NI0aE4Vga3bxyIiJS7BT0iszBB4exg2+5JfTIPeAAuPRSTQm7AzP4+tfDTY233AKffBKGYjngAPjb33RJV0REioaCXhFq2jSMrzdnDnzve+F2tAEDQobRVcoUrVtX/qL++tfQonf66bDXXnDbbbB5c9QVioiI1ImCXhHr0gXuuitMCbv77iHDjBwZ+iVIiubNw42O770H48dD164hAPbpE256VHOoiIgUKAW9RmDYsBD27rwz9EUYPBguuijcoiYpmjQJl3CnTAmDLe+/f+jG3Lt3GKxw+fKoKxQREcmKgl4j0bQp/OhH4ba0s8+Gm28OQ7Pcd5/6IOzEDEaNggkTYOrU0Kvl2mtDC98FF8DChVFXKCIiUiuxCXpm1tPM7jazpWa2xcwWmNk4M+uYxTn+18yeTDx3vZmtNbP3zOx6M+tZn/UXik6d4NZbQ34pKwv38B16KLz9dtSVxdSBB8Kjj4buy6edFuab23NPOPNMdWkWEZHYi0XQM7O+wDTgLOAN4AbgI+AiYIqZdarlqX4M7Aa8CNwK/AVYCVwCvG9mg/NcesEaMiQMxXL33TB3LpSXw3nnweefR11ZTA0YEH5ZH30UflGPPgr77BN6777xRtTViYiIpGUeg26YZvYMMAa40N1vTtl+PSGk3eHu59TiPK3cfaeukmb2I+BO4Cl3P66m85SXl/vUqVOzeQsFbfXqMJbwLbfArrvCb38bxuHr0CHqymLss8/gppvCNfDkeHyXXw5HHBHu9RMREWkgZjbN3cvT7os66CVa8z4EFgB93b0iZV8J8AlgQFd335Dja7QHVgMfunu/mo5vbEEv6Z134Pzz4eWXQ1YZMiTkl1Gj4JBDoF27qCuMoXXr4I474Prrw3h83brB8cfDCSeEuXXbto26QhERKXLVBb04ND2MSqwnpIY8AHdfB7wCtAGG1+E1Tkis363DOYreAQfA5Mlh+eUvwzBzN9wAxx4LHTuGsPfLX4ZZwzTEXEJJCVx2Wbik++CDIRX/85/hkm6nTiH03XEHLFkSdaUiItIIxaFF7w/AZcBl7v7HNPv/BJwHnOvut9XynD8EegLtgP2AI4FFwFHuPrem5zfWFr10NmyAV18No41MnBhm3aiogJYtwywcyRa/oUOhRYuoq42JrVvhpZfCtGr//ncIgRCaSE88MbT2DR4ceveKiIjUUdwv3d4J/Aj4kbv/Oc3+a4CfAz9392trec7XgINSNr0JfMfdP6zmOWcDZwP07t37wIUaQiOtNWtChpk4MYS/d94Js220bRta/EaPDsvgwWFIl0bPHWbOrAx9U6aEbT17wle/GkLf6NHQqlXUlYqISIFqdEEv5bmdgCHANUB/4Fvu/kxNz1OLXu2tXAkvvljZ4pcccaR9ezj88NDaN3o07Luv+igAYdDlJ58Mwe+ZZ0KTadu24X6+E08Ml3q7do26ShERKSDVBb1mDV1MGmsS6/YZ9ie3r872xO6+EnjWzN4EZgH3m9ke7r4p6yolrU6d4BvfCAvAsmUh8CVb/P7977C9S5cQ+I44IixlZdHVHKmuXcPghd/7XrjRcdKkEPqeeAIefzxczj3ooBD6Tj4Z+tXYd0hERCSjOLTo/RC4C7jT3X+cZn9y6JUj3f35OrzOY8BJwFB3r7a5Ti16+fPxxyHwPf98WD75JGzv06cy9I0eHTqrNmru4Tp48hJv8s/f0UeHrtDHHqtr4SIiklbcL93W+/AqiXO9AQwFBrv79OqOVdCrH+4wa1Zl6Js4MdzzB+HSbjL4HX447LJLtLVGbskSuOceuO02WLo0NIGeey6cdVYY7FBERCQh1kEPsh8w2cwGALj7rJRtvYEt7v5pmvP/GLid0PO21N23V1ePgl7D2L4d3norhL7nngszdWzeHBquhg4NU8wecQR85Suhl2+jtHUrjB8fBmaePDmMefPd74ZWvgMOiLo6ERGJgUIIen2BV4GuwHhgJqHX7ChgDnBw4n675PEO4O6Wsu0k4BFgCqGF8FOgE2H8vf2A9cBX3f3FmupR0IvG5s1hKJdki19yKJfWrUOP3iOOCOFvyJBGOjLJu++G6Uvuvx82bQqTFJ9/fhizr3nzqKsTEZGIxD7oAZhZL+DXwDGEgPYJ8BhwtbuvqnJsuqDXG7gQOBToA+wKbCbMmfsscKO7L6pNLQp68bBmTejRmwx+778ftvfvD9//PowdCz16RFtjJFatCpd1b7kljNHXoweccw6cfTZ07x51dSIi0sAKIujFiYJePC1bBv/9b8g4r7wSLvEed1wIfccf3wgbtSoq4Omn4U9/gqeeCr+Ak0+GCy6A4cMbabOniEjjo6CXJQW9+Js9OwS+++4LAbBrVzjjjBD6Bg6MuroIzJ0Lt94afilr1oTr2+efD6edFq59i4hI0Yr7XLciWevfH667DhYtCqORHHwwjBsHe+8dOm/cdResXRt1lQ2oX78wMfHixXD77fDFFyH19uwJP/0pLFgQdYUiIhIBteiloRa9wvTpp/DAA/CXv4RZx9q0gVNOCXnn0EMb2ZVM99BL9+abw0DM27eH0Lf33jsvHTtGXa2IiNSBLt1mSUGvsLnD66/D3XfDQw/BunWw555hCLozz4Tdd4+6wga2aFH4Rbz3XpijbuZM2Lixcn+PHukDYOfO0dUsIiK1pqCXJQW94rFhA/zzn6GVb/LkMN/uMceEVr4TToAWLaKuMAIVFWHKkg8+qFzefz+s16+vPK5Ll8rQt88+lT937drImkdFROJNQS9LCnrFae5cuPfesCxdGhqsjjoKevXaeencuRFmGfdwj19qAEyGwOQUJhBm5hg8OMxdN3o0lJdDszhMmy0i0jgp6GVJQa+4bdsGEyaEDqrTpoVss3Xrjse0ahVuaUsXApNL+/aNJAy6h67NydD3/vvw2mthAGeAkhI47LDK4Lf//qHpVEREGoSCXpYU9BqXigpYvjzcypZuWbw4tABurzJxXrt2IfAlA2HPniHztG1bu6XgLxuvWBFGtH7hhbDMnh2277orjBxZGfwGDGgkiVhEJBoKellS0JOqtm0LjVqZwuCiRaHXbzZfp2bNqg+CrVuHMNiiRZjrN/lzpqW6Y1q1Cn0uunSpx8y1ZAlMnBhC3/PPh/sAIczWkQx9o0dDaWk9FSAi0jgp6GVJQU9ysX176PyRr2Xz5jAcXtWl6mXmbLRsWfMl6Q4d8hAG3WH+/MrWvhdeCEkYoE+fytA3ahTstlsdX0xEpHFT0MuSgp7EmXsIe+lCYOqyZUvlzxs3hsvPVVshlyzZ+ZJ027Y1h8GSkhyKnjmzssVv4sQwZy+E0a9HjQqXew8/XPP1ihS5Vavgww/DfzxTl1atKn9u2jTqKguLgl6WFPSksdi+vfKS9OLF6S9Jf/LJzpeka7pkXOO25hW0WPMZLT9ZQMvF8+iy6C26b1lAd5bRvawt3UYOpOSo4dhIBT+RQuce+m49+WRYpkzZ+T+YVTVrljkEpv6cfNy8eXhOs2aVP6fbVpuf27QJFxp22y2MJlUIgwoo6GVJQU+k0tatO7cGrlqVvuUw18fp/hpqzcYQ/FqvoXt36Na3hO6DutF9z5LwuBtfruM8na97aDmdPTt04Bk2TH1TopL8j02rVuEWhWJsNdqyBebNg1mzYNMmGDQo9Idq6Pe6di0891wIdk89Ff4OgTAN93HHhVGZtm0L9W7ZEm5VSf5c0+N0+7ZtC39XbdtWuaQ+3ro1dLzLVpMm4e+YZPDLtHTuHO1gA9UFvQLIqSISpebNYY89wlIfKipg5crwD/CXy9LtfPreWpZ9sJllHxtzFrbgpfnt+ey59NeM27evDH277w69e1fWvMce4XHWl5uztGEDzJkTAl3qMmfOjuNQ9+0LZ5wRlj596remxmj79tAP6MMPw9iZH35YucybF/5zASFsd+gQOonvuit06lT5c3WP4xAQ3eGzz8Kfr1mzwpL8+aOPdg40bdrAAQeEkDVkCBx4YBj7vHnz/NY0c2Zlq91LL4WAtcsucPTRcOyxYbD6Hj3y95rZqqgIfz6qBsCqP69fH65kLF0aliVLwvrjj8PIUitW7HzuZs3Ce0sXAocPD2E7KmrRS0MteiIxs20bvP02W5+fzPJn32HZawtYtrGET+nGss77saz7IJa168sy786SFS1YtGjnTisdO+4c/lIf16ZHckVFuMSdDHHJf2Bnzw4tnUlm4Zz9+++4LF4M990XblGEcEvimWfCySfXfxAtJtu2wYIFO4a4ZLCbP3/Hz7516zAF4p57Qr9+IVx/8QV8/nlYVq6s/Dn5ePXq6l+/Q4edg2DHjjU/btkyu/e5dWt4P8kwlxroPv+88riWLWGvvUKY6N8/rAcMCLdJvP02vPVWWN5+u/I/HS1ahCEvk8FvyBDYd9/Q2llbGzaEW26T4S7Z0X6//UKr3XHHwVe+kt9AGQdffBH+Q5oMgpmW5G3If/gDXHZZ/dakS7dZUtATiblE8GPSpLC89FKY1BhgyBC2n3Mey0Z9m4XLW/Pxx7Bw4c5LaisbhH/gqoa/7t1DgEv+4zpnTrgcllRSUvkPa2qg69ev+svJCxfCAw+E0Dd3bjj2G98IoW/06IZrMVq7NrRQvPJKeF9f+QqMGBHuS4qD9evhzTfDNM2prXMLFoQ/Aklt24bfeTLQJUPdnnuGVpZsL5Vv3x7CXnVhMPXnVavCz6tXV395sE2b6sNgScmOf94+/HDH99mtW2WISw10vXvX7s9MRUX4PSaDX3JJBttmzcJsh6nh74ADQt1Jc+dWBrtJk0Loads2zDJ03HGh1a5Xr+x+38Vq06bQMlhSEv4jWZ8U9LKkoCdSYLZtg+nTQ/PC/ffDjBnhX84f/AB+8pOdxu5zD/+4VQ1/qaFw+fJwrFl4emqQS/5D27173e63c4fXXw+B76GHQk277w6nnx5C38CBuZ873Wt9/HEIdcnlvffCP/5NmoR/5JOXNfv1C4FvxAg45JDwXuv7vsKKihBwXnutcpkxozI4lZTsGOZSf+7WLR73PVZUhPCcDIHJAJi6pNv2+efhXjMIrV977rlzoOvfP7Qk5ltyJKTU4DdtWrg0DOHPxoAB4VLv9OkhfELYlmy1O+SQ7FsrJb8U9LKkoCdSwNxh8mT405/gscfCv75f/Sqcfz4ceWSt75jetCkM/de9e3aXs3K1eTM88QT89a/h5vXt22Ho0HAv37e/HS4VZmPbNnjnnR2D3ZIlYV+7duG+oWSYO+ig8A/1tGnhuJdfDuuVK8PxnTrBwQdXBr8DD6z77+Tzz0PITYa611+vnFK5Q4dQ0/DhYRk8OLQyxiHM1ZdNm8L779w5+l6eyWmvU8PfjBkh7B13XLjfrqws2hplRwp6WVLQEykSixfDHXfAnXeGJrr+/eG880Jz2S67RF1dRp9+Cg8+GFr63nkntPJ89ash9B13XPrp81Ivw77ySvh5w4awr1evylA3YkS4h6qmMOEeLlWnBr85c8K+Fi1Cr8lk8Dv44BBQMtm2LbQeprbWJc/VpEmoJxnqhg8P95tpumSR2lPQy5KCnkiR2bIFHnkktPK9/npo0jrjjBD69t476uqq9c47oZXvb38LAbBz59DCd/LJIcemuwy7//47BrvevfNTy4oV8OqrlcFv6tTKjg/9+1e+3oEHht6fyVA3dWoYtBvCZdbUUFdeHj4OEcmdgl6WFPREitibb8Itt4Sb4rZsCb0fLrggNJlFfc2sGtu2wTPPhNA3fnzlPV1VL8MOH95wPXg3bQohLhk2X311x96gLVqEG/qHD6+8FLvHHsV9CVYkCgp6WVLQE2kEVqyAv/wFbr01dHXs3Tt03PjhD6u/Dpmr5FQAeehSu2pV6HdSVla7y7ANpaIi9BadPj2MFzhokG7SF2kICnpZUtATaUS2bYP//Cdc1n3++ZBMvv3t0HnjwAN3Pn7z5pq7UabbtmZNCHk9eoRQWXUC4eS22gzoJyKSQkEvSwp6Io3UBx+Ey7r33Rd6MgwZEq6Npga41IH0qmratPoRc7dvr5xH7uOPwzp5DTapZUvo2XPH8Fc1DLZvX7+/BxEpKAp6WVLQE2nk1qwJN8P94x/humh1o9ymbispya41LjmXVWrwqxoEly7deQb4kpIQ+gYNCpPnDhsWfm6IcWBEJHYU9LKkoCcisbFtWxhev2oInD8/DHz3ySfhuGbNQnfboUMrw9/AgdFPzCoi9a66oBeTW3hFRCStZs0qL92ms2QJvPFG6E38xhuhN/Edd4R9bduG+wyHDasMgOr2KtKoqEUvDbXoiUjBSk5omhr+pk+vvBewS5cdW/2GDq2fXsYi0mDUoici0lg0aVI5Ke/YsWHbF1+EEZVTw99TT4V7BCFM5nvwwTBmTFi6d4+ufhHJK7XopaEWPREpeuvWhUlM33gjLJMnh2niIHTsOProsIwYkX7ONRGJjYLojGFmPYFfA8cAnYBPgMeBq919VS2e3xY4CTgeGAL0AiqA2cDfgZvd/Yva1KKgJyKNTkVFuMT7zDNheeWV0BGkbVsYNaoy+O25p+7xE4mZ2Ac9M+sLvAp0BcYDs4BhwChCUBvh7itrOMcxwFPA58BE4EOgI3Ai0D1x/iPcfXNN9SjoiUijt24dTJxYGfzmzQvbS0vhmGNC6Bs9uuHmWxORjAoh6D0DjAEudPebU7ZfD1wC3OHu59RwjkHAPsAjqS13ZlYCTCK08l3m7n+sqR4FPRGRKj78sDL0vfBCGFC6WbNwb1+ytW/w4HCPoIg0qFgHvURr3ofAAqCvu1ek7CshXMI1oKu7b8jxNb4D/A34j7ufUNPxCnoiItX44gt49dXK4Pf222F7ly5w1FGhQ0efPtCpU1h23VWT3orUo7j3uh2VWE9IDXkA7r7OzF4htPYNB57P8TW2Jtbbcny+iIgktWgBI0eG5dpr4dNP4dln4emnYcIEePDBnZ/Ttm0IfKnhL3WdblvHjqHVUERyFodvUP/Eek6G/XMJQW8vcg9630+sn87x+SIikkm3bnD66WGpqICZM2HZMli5Miyff77jeuVKWLy4cltFReZzt28f7gNs1izM8lHXdYsWsNdeYR7jIUM0hqAUvTgEveTs3Gsy7E9u75DLyc3sfEJP3unA3dUcdzZwNkDv3r1zeSkREWnSBPbZJyy1UVEBa9fuGAKrBsN168J8v9u3h57AVX9OXW/eXP0xmzbBPfdUvn7v3pWhL7n06FE/vxuRCMQh6NUbM/sGMA5YBnzT3bdmOtbd7wTuhHCPXoMUKCLS2DVpAh06hKWsrGFec9WqcF/hW29VLuPHVw4g3b17mDouNfz16qVhZaQgxSHoJVvs2mfYn9y+OpuTmtlJwEPAcmCUu3+US3EiIlJkOnYMQ8OMHl25bd26MI5gavh76qnKy8qdO+/c8ldWFq/w5x5aNDdsCEvbtro0LbEIerMT670y7O+XWGe6h28nZnYK8CChJW+0u8/NvTwRESl6JSVw6KFhSdq4Ed59d8fw98c/wtbExaH27cMA0i1bhqVFi8p16s/ptqXb37x5mJM4GdRqu2zcWPlz1ZE09t03BNojjoDDDw81S6NSdMOrmNl3gfuAJeTYkqfhVUREJK0tW2DGjBD6pk2Djz8Ow80kly1bdlxX3bZ9e3av16xZaJmrurRpk3576vLpp2HQ65dfDvcmNmkC5eWVwe/gg8N5pODFehw9yH7AZDMbAODus6qc50xCh4uFhJC3MJd6FPRERKRebN8eWgTTBcIvvgite6lhLR/zDG/ZAq+9Bs8/Hwa7fv310DmlRQv4yldC6Bs9GoYNC62KUnAKIehVnQJtJnAQYYy9OcDBqVOgmZkDuLulbBsFPAc0IYS9RWlearW7j6upHgU9EREpWuvXw0svhdD3/PPh3kT3ECwPO6wy+B1wgGY6KRCxD3oAZtYL+DVhKJROhEu2jwFXu/uqKsemC3rfA+6hegvdvU9NtSjoiYhIo7FyJUyaFILfCy/ArMTFsl13hVGjQugbOTIEwfXrd7w/MPVxpp/T7WvaNAxt06dP5bLHHpXrdu2i+m0UpIIIenGioCciIo3WkiXh3r7nnw/LonQXyDJo2jQEwnbtKi8/p/t561ZYuLBy2bJlx/N07lwZ/NIFwV12yd/7LQIKellS0BMRESFc0p03L3ToqKioOcS1aJH9kDMVFaHjyMKFsGBB5ZL6ePPmHZ/TseOOAbB3b+jZs3Lp0aNRTZ+noJclBT0REZGYcIcVKzKHwAULwhAzqZo0CQNfp4a/Xr12fLzbbvnp7BID1QW9xhN3RUREpPCYQdeuYRk2bOf97rB6dZg/Od0ycyZMmBDuDayqW7cdw1/PnqF1sKwsLF26xGtQ7Bwo6ImIiEjhMguXcjt2hP32y3zc2rWZw+BHH8HkyWF6vFTt2lWGvuTSt29Y77FHGA4n5hT0REREpPjtsgvsvXdYMtmwIVwWnj8/hL9588J67lx45pkw8HSSWWgBTAa/qkGwU6dYtAYq6ImIiIhA6FSSKQy6w7JlIfgll2QQfOop+OSTHY8vKQmB79JLYezYhqk/DQU9ERERkZqYhd68PXrAiBE779+4sbIlMHVp3brha02hoCciIiJSV23awD77hCVGNLeJiIiISJFS0BMREREpUgp6IiIiIkVKQU9ERESkSCnoiYiIiBQpBT0RERGRIqWgJyIiIlKkFPREREREipSCnoiIiEiRUtATERERKVIKeiIiIiJFSkFPREREpEgp6ImIiIgUKXP3qGuIHTNbASys55fpDHxWz68hNdPnEA/6HKKnzyAe9DlErxA/gz3cvUu6HQp6ETGzqe5eHnUdjZ0+h3jQ5xA9fQbxoM8hesX2GejSrYiIiEiRUtATERERKVIKetG5M+oCBNDnEBf6HKKnzyAe9DlEr6g+A92jJyIiIlKk1KInIiIiUqQU9ERERESKlIJeAzKznmZ2t5ktNbMtZrbAzMaZWceoa2ssEr9zz7Asi7q+YmJmJ5vZzWb2kpmtTfyOH6jhOQeb2ZNm9rmZbTKzd83sYjNr2lB1F5tsPgcz61PN98PN7KGGrr/QmVknM/uhmT1mZh8m/lyvMbOXzewHZpb232F9F/Ir28+hmL4LzaIuoLEws77Aq0BXYDwwCxgGXAQcY2Yj3H1lhCU2JmuAcWm2r2/gOordL4ADCL/XxcCA6g42s68B/wQ2A/8APgdOAG4ARgCn1GexRSyrzyHhHeDxNNtn5K+sRuMU4DbgE2Ai8DHQDfgG8GfgWDM7xVNumNd3oV5k/TkkFP53wd21NMACPAM4cEGV7dcntt8edY2NYQEWAAuirqMxLMAooB9gwMjEn/MHMhy7C7Ac2AKUp2xvRfgPkgOnRf2eCnHJ8nPok9h/b9R1F8sCjCaEtCZVtncnhA0HvpmyXd+FeHwORfNd0KXbBpBozRtDCBm3VNl9JbABGGtmbRu4NJF64+4T3X2uJ/7WrMHJQBfgIXefmnKOzYQWKYCf1EOZRS/Lz0HyzN1fcPcn3L2iyvZlwO2JhyNTdum7UA9y+ByKhi7dNoxRifWENH/I1pnZK4QgOBx4vqGLa4RamtnpQG9CyH4XmOzu26Mtq1EbnVg/nWbfZGAjcLCZtXT3LQ1XVqO1m5n9GOgErASmuPu7EddUjLYm1ttStum70PDSfQ5JBf9dUNBrGP0T6zkZ9s8lBL29UNBrCN2B+6tsm29mZ7n7i1EUJJm/I+6+zczmA/sAZcDMhiyskToqsXzJzCYBZ7r7x5FUVGTMrBlwRuJhaqjTd6EBVfM5JBX8d0GXbhtG+8R6TYb9ye0d6r+URu8e4AhC2GsL7AfcQbgf4ykzOyC60ho1fUfiYSPwG+BAoGNiOZxw8/pI4HndYpI31wH7Ak+6+zMp2/VdaFiZPoei+S4o6Emj4u5XJ+7V+NTdN7r7DHc/h9AppjVwVbQVikTH3Ze7+6/c/S13X51YJhOuOLwO7An8MNoqC5+ZXQhcShh9YWzE5TRa1X0OxfRdUNBrGMn/gbXPsD+5fXX9lyIZJG/GPSzSKhovfUdizN23EYagAH1H6sTMzgduBD4ARrn751UO0XehAdTic0irEL8LCnoNY3ZivVeG/f0S60z38En9W5FYF0RTfBHK+B1J3ENTSrhR+qOGLEp2oO9IHZnZxcDNhDHYRiV6fFal70I9q+XnUJ2C+i4o6DWMiYn1mDSjb5cQBsDcCLzW0IXJl4Yn1vrLMxovJNbHpNl3GNAGeFW9DCOl70gdmNlPCQMeTyeEi+UZDtV3oR5l8TlUp6C+Cwp6DcDd5wETCDf8n1dl99WE/xXc7+4bGri0RsXMBqa7edbM+gB/SjysdoouqTePAp8Bp5lZeXKjmbUCfpt4eFsUhTUmZjYk3ZRcZnYEcEniob4jWTKzXxJu+p8GHOHun1VzuL4L9SSbz6GYvgumMTQbRpop0GYCBxHG2JsDHOyaAq1emdlVhBtvJwMLgXVAX+B4wqjzTwJfd/cvoqqxmJjZScBJiYfdgaMJ/wN+KbHtM3e/rMrxjxKmfXqIMO3TiYThJh4FvqVBf7OXzeeQGDaiH+HvqsWJ/ftTObbbL909GTakFszsTOBeYDvhcmG63rQL3P3elOechL4LeZXt51BM3wUFvQZkZr2AXxOa5DsR5tx7DLja3VdFWVtjYGaHA+cAg6kcXmU1oQn/fkKrqr4QeZII1ldWc8hCd+9T5TkjgCuArxDC94fA3cBNGtA6N9l8Dmb2A+DrhOEmOgPNgU+BKcCf3P2lTCeR9Grx+wd40d1HVnmevgt5lO3nUEzfBQU9ERERkSKle/REREREipSCnoiIiEiRUtATERERKVIKeiIiIiJFSkFPREREpEgp6ImIiIgUKQU9ERERkSKloCciUoDM7CozczMbGXUtIhJfCnoi0iglQlJNy8io6xQRqYtmURcgIhKxq6vZt6ChihARqQ8KeiLSqLn7VVHXICJSX3TpVkSkFlLviTOzM83sbTPbZGbLzexuM+ue4Xn9zOyvZrbEzL4ws6WJx/0yHN/UzM4xs1fMbE3iNT40sz9X85yTzewNM9toZp+b2UNmtns+37+IFCa16ImIZOcSYAzwD+Bp4BDgLGCkmR3k7iuSB5rZUOA5oAT4N/ABMAA4HfiamR3p7m+mHN8C+A9wFLAIeBBYC/QBvg68DMytUs+5wImJ878IHAScChxgZoPcfUs+37yIFBYFPRFp1Mzsqgy7Nrv7dWm2Hwsc5O5vp5zjBuBi4DrgB4ltBvwV2AU43d3/lnL8qcBDwP1mtre7VyR2XUUIeU8Ap6SGNDNrmThXVccAQ939vZRjHwS+DXwNeDjTexeR4mfuHnUNIiINzsxq+stvjbt3SDn+KuBK4G53/0GVc7UHFgItgQ7uvsXMRhBa4Ka4+8FpXv8lQmvg4e4+2cyaAiuBFsCe7r60hvqT9Vzj7r+osm8U8ALwR3e/rIb3KSJFTPfoiUij5u6WYemQ4SkvpjnHGmA60AoYmNg8JLF+IcN5ktsHJ9YDgPbAuzWFvCqmptm2KLHumMV5RKQIKeiJiGTn0wzblyXW7ausP8lwfHJ7hyrrJVnWszrNtm2JddMszyUiRUZBT0QkO90ybE/2ul1TZZ22Ny7Qo8pxqxNr9ZYVkbxR0BMRyc7hVTck7tEbBGwGZiY2JztrjMxwnlGJ9VuJ9SxC2NvfzHbLQ50iIgp6IiJZGmtmg6tsu4pwqfbvKT1lXwFmA4eY2cmpByceHwrMIXTYwN23A7cCrYHbE71sU5/Twsy65Pm9iEiR0/AqItKoVTO8CsDj7j69yrangFfM7GHCfXaHJJYFwOXJg9zdzexM4FngH2Y2ntBq1x84CVgHnJEytAqE6dgOAk4A5pjZfxLH9SKM3fe/wL05vE0RaaQU9ESksbuymn0LCL1pU90APEYYN+9UYD0hfP3c3ZenHujurycGTf4FcCQhwH0G/B34jbvPrnL8F2Z2DHAOcAZwJmDA0sRrvpztmxORxk3j6ImI1ELKuHWj3H1StNWIiNSO7tETERERKVIKeiIiIiJFSkFPREREpEjpHj0RERGRIqUWPREREZEipaAnIiIiUqQU9ERERESKlIKeiIiISJFS0BMREREpUgp6IiIiIkXq/wO04vUSrj+N9AAAAABJRU5ErkJggg==",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 720x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "# Plot the loss function\n",
        "fig, ax = pyplot.subplots(1, 1, figsize=(10,6))\n",
        "ax.plot(np.sqrt(history.history['loss']), 'r', label='train')\n",
        "ax.plot(np.sqrt(history.history['val_loss']), 'b' ,label='val')\n",
        "ax.set_xlabel(r'Epoch', fontsize=20)\n",
        "ax.set_ylabel(r'Loss', fontsize=20)\n",
        "ax.legend()\n",
        "ax.tick_params(labelsize=20)\n",
        "\n",
        "# Plot the accuracy\n",
        "fig, ax = pyplot.subplots(1, 1, figsize=(10,6))\n",
        "ax.plot(np.sqrt(history.history['accuracy']), 'r', label='train')\n",
        "ax.plot(np.sqrt(history.history['val_accuracy']), 'b' ,label='val')\n",
        "ax.set_xlabel(r'Epoch', fontsize=20)\n",
        "ax.set_ylabel(r'Accuracy', fontsize=20)\n",
        "ax.legend()\n",
        "ax.tick_params(labelsize=20)"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "MfNhwVLyRccg",
      "metadata": {
        "id": "MfNhwVLyRccg"
      },
      "source": [
        "### Prediction \n",
        "\n",
        "1. Read one image\n",
        "2. preprocessing , scaling and normalisation\n",
        "3. prediction"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "-R2700_VRvDR",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "id": "-R2700_VRvDR",
        "outputId": "a93a90ef-dc21-42a5-8feb-1d81f27fa566"
      },
      "outputs": [
        {
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/NK7nSAAAACXBIWXMAAAsTAAALEwEAmpwYAAANrklEQVR4nO3db4xV9Z3H8c9Hl0YCPACJZKRmgQYfNCbCSkhxjcE0bZSQYJ805cGKWZNpYg012aikDanJponZpfrM6tQqsKk2jUpqyEaqpC5rYhrAuIKDRTRgwZERMSloYhG++2AOzYhzf3e4/85lvu9XMpl7z/eee7658Jlz7vnde36OCAGY+i6ruwEAvUHYgSQIO5AEYQeSIOxAEv/Qy43Z5tQ/0GUR4YmWt7Vnt32r7T/bPmR7QzvPBaC73Oo4u+3LJR2U9B1JRyXtlrQ2IoYL67BnB7qsG3v25ZIORcR7EfE3Sb+VtKaN5wPQRe2Efb6kv4y7f7Ra9iW2B23vsb2njW0BaFPXT9BFxJCkIYnDeKBO7ezZj0m6Ztz9r1fLAPShdsK+W9Ji2wttf03SDyS90Jm2AHRay4fxEfGF7Xsk7ZB0uaQnI+KtjnUGoKNaHnpraWO8Zwe6risfqgFw6SDsQBKEHUiCsANJEHYgCcIOJEHYgSQIO5AEYQeSIOxAEoQdSIKwA0kQdiCJnl5KGpee9evXF+uPPPJI17Z92WXlfdH27duL9RdffLFhbWhoqLjumTNnivVLEXt2IAnCDiRB2IEkCDuQBGEHkiDsQBKEHUiCcXYU3XDDDcV6N69OfO7cuWL9tttua7n+9ttvF9fduXNnsX4pYs8OJEHYgSQIO5AEYQeSIOxAEoQdSIKwA0kwzo6izZs3F+tXXXVVsT4wMNCwdt111xXXPXjwYLF+4sSJYn3FihUNa/fff39x3VdffbVY//zzz4v1ftRW2G0flnRK0llJX0TEsk40BaDzOrFnvyUiyn9iAdSO9+xAEu2GPST9wfZe24MTPcD2oO09tve0uS0AbWj3MP6miDhm+ypJL9l+OyJ2jX9ARAxJGpIk29371gSAorb27BFxrPo9KmmbpOWdaApA57UcdtszbM86f1vSdyXt71RjADrLrX4f2fYije3NpbG3A09HxM+brMNhfDKzZ89uWJs/f35x3ePHjxfrV199dbG+d+/eYr2d5x4dHW35ubstIjzR8pbfs0fEe5Kub7kjAD3F0BuQBGEHkiDsQBKEHUiCsANJ8BVXdNUnn3zSUk2SVq5cWayvXr26lZYkScPDw8X6Z5991vJz9yv27EAShB1IgrADSRB2IAnCDiRB2IEkCDuQBOPs6Krp06c3rD3wwAPFdVetWlWsN5tOuqTZpaJPnz7d8nP3K/bsQBKEHUiCsANJEHYgCcIOJEHYgSQIO5AE4+zJbd26tVhv9VLj5y1cuLBh7cYbbyyua094ReS/e//994v1J554omHt0UcfLa47FbFnB5Ig7EAShB1IgrADSRB2IAnCDiRB2IEkWp6yuaWNMWVzzy1YsKBYf/fdd4v1Xv7/uNDGjRuL9c2bNxfrIyMjHezm0tFoyuame3bbT9oetb1/3LI5tl+y/U71u/Ek3AD6wmQO4zdLuvWCZRsk7YyIxZJ2VvcB9LGmYY+IXZJOXrB4jaQt1e0tkm7vbFsAOq3Vz8bPi4jzb4g+lDSv0QNtD0oabHE7ADqk7S/CRESUTrxFxJCkIYkTdECdWh16O257QJKq36OdawlAN7Qa9hckratur5P0+860A6Bbmh7G235G0kpJc20flfQzSQ9J+p3tuyQdkfT9bjaJ1n366ad1t9DQhg3lQZxNmzb1qJMcmoY9ItY2KH27w70A6CI+LgskQdiBJAg7kARhB5Ig7EASXEp6ivvoo4+K9TqHtx577LHatp0Re3YgCcIOJEHYgSQIO5AEYQeSIOxAEoQdSIJLSaNv3XLLLcX62bNni/Vdu3Z1sp1LRsuXkgYwNRB2IAnCDiRB2IEkCDuQBGEHkiDsQBJ8nx1969prry3Wd+zY0aNOpgb27EAShB1IgrADSRB2IAnCDiRB2IEkCDuQRJrvs8+aNatYX7FiRbH+2muvNaydOnWqpZ7QnkWLFhXr119/fcPatm3bOt1O32j5++y2n7Q9anv/uGUP2j5m+43qZ1UnmwXQeZM5jN8s6dYJlj8SEUuqn//ubFsAOq1p2CNil6STPegFQBe1c4LuHttvVof5sxs9yPag7T2297SxLQBtajXsv5T0DUlLJI1I+kWjB0bEUEQsi4hlLW4LQAe0FPaIOB4RZyPinKRfSVre2bYAdFpLYbc9MO7u9yTtb/RYAP2h6ffZbT8jaaWkubaPSvqZpJW2l0gKSYcl/bB7LU7OzTffXKw//vjjxfrixYuL9SVLljSs7d/P37o6NPvsxFNPPdWwtm/fvuK6hw4daqmnftY07BGxdoLFv+5CLwC6iI/LAkkQdiAJwg4kQdiBJAg7kMSUuZT06tWri/VmQ2tHjhwp1j/++OOL7gn1mjlzZsPaFVdc0cNO+gN7diAJwg4kQdiBJAg7kARhB5Ig7EAShB1IYsqMs9sTXj130vWnn366WB8ZGbnontCeGTNmFOvTp08v1l9++eWGtYxfS2bPDiRB2IEkCDuQBGEHkiDsQBKEHUiCsANJTJlx9mZTTzer33nnncX6K6+80rB24MCB4roffPBBsX4pa3Y559J1BubOnVtc94477ijWm30n/eGHHy7Ws2HPDiRB2IEkCDuQBGEHkiDsQBKEHUiCsANJTJlx9nYNDAwU6zt27GhYa3bN+WbT/zb7DEA7du/eXawvX768WG/W25VXXlmsL126tFgvOXPmTLF+9913F+ulKZszarpnt32N7T/aHrb9lu0fV8vn2H7J9jvV79ndbxdAqyZzGP+FpH+LiG9K+pakH9n+pqQNknZGxGJJO6v7APpU07BHxEhEvF7dPiXpgKT5ktZI2lI9bIuk27vUI4AOuKj37LYXSFoq6U+S5kXE+QuzfShpXoN1BiUNttEjgA6Y9Nl42zMlPSfp3oj46/hajJ3FmfBMTkQMRcSyiFjWVqcA2jKpsNueprGg/yYinq8WH7c9UNUHJI12p0UAneBmQyseuwbzFkknI+Leccv/U9LHEfGQ7Q2S5kTE/U2eq2tjTM2GztavX1+s33fffZ1s50uaXca6m0NvzdTZ2/DwcLG+adOmYn3r1q2dbGfKiIgJ/1En8579nyX9i6R9tt+olv1E0kOSfmf7LklHJH2/A30C6JKmYY+IVyU1+vP/7c62A6Bb+LgskARhB5Ig7EAShB1IgrADSTQdZ+/oxro4zt7MtGnTivXZs8tf2lu3bl3D2saNG4vrNpt6uJ/H2UdHy5+VajbWXbrM9rPPPltc9/Tp08U6JtZonJ09O5AEYQeSIOxAEoQdSIKwA0kQdiAJwg4kkWacHciCcXYgOcIOJEHYgSQIO5AEYQeSIOxAEoQdSIKwA0kQdiAJwg4kQdiBJAg7kARhB5Ig7EAShB1IomnYbV9j+4+2h22/ZfvH1fIHbR+z/Ub1s6r77QJoVdOLV9gekDQQEa/bniVpr6TbNTYf++mI2DTpjXHxCqDrGl28YjLzs49IGqlun7J9QNL8zrYHoNsu6j277QWSlkr6U7XoHttv2n7S9oTzJ9ketL3H9p72WgXQjklfg872TEn/I+nnEfG87XmSTkgKSf+usUP9f23yHBzGA13W6DB+UmG3PU3Sdkk7IuLhCeoLJG2PiOuaPA9hB7qs5QtOemyaz19LOjA+6NWJu/O+J2l/u00C6J7JnI2/SdL/Ston6Vy1+CeS1kpaorHD+MOSflidzCs9F3t2oMvaOozvFMIOdB/XjQeSI+xAEoQdSIKwA0kQdiAJwg4kQdiBJAg7kARhB5Ig7EAShB1IgrADSRB2IAnCDiTR9IKTHXZC0pFx9+dWy/pRv/bWr31J9NaqTvb2j40KPf0++1c2bu+JiGW1NVDQr731a18SvbWqV71xGA8kQdiBJOoO+1DN2y/p1976tS+J3lrVk95qfc8OoHfq3rMD6BHCDiRRS9ht32r7z7YP2d5QRw+N2D5se181DXWt89NVc+iN2t4/btkc2y/Zfqf6PeEcezX11hfTeBemGa/1tat7+vOev2e3fbmkg5K+I+mopN2S1kbEcE8bacD2YUnLIqL2D2DYvlnSaUlbz0+tZfs/JJ2MiIeqP5SzI+KBPuntQV3kNN5d6q3RNON3qsbXrpPTn7eijj37ckmHIuK9iPibpN9KWlNDH30vInZJOnnB4jWStlS3t2jsP0vPNeitL0TESES8Xt0+Jen8NOO1vnaFvnqijrDPl/SXcfePqr/mew9Jf7C91/Zg3c1MYN64abY+lDSvzmYm0HQa7166YJrxvnntWpn+vF2coPuqmyLinyTdJulH1eFqX4qx92D9NHb6S0nf0NgcgCOSflFnM9U0489Jujci/jq+VudrN0FfPXnd6gj7MUnXjLv/9WpZX4iIY9XvUUnbNPa2o58cPz+DbvV7tOZ+/i4ijkfE2Yg4J+lXqvG1q6YZf07SbyLi+Wpx7a/dRH316nWrI+y7JS22vdD21yT9QNILNfTxFbZnVCdOZHuGpO+q/6aifkHSuur2Okm/r7GXL+mXabwbTTOuml+72qc/j4ie/0hapbEz8u9K+mkdPTToa5Gk/6t+3qq7N0nPaOyw7ozGzm3cJelKSTslvSPpZUlz+qi3/9LY1N5vaixYAzX1dpPGDtHflPRG9bOq7teu0FdPXjc+LgskwQk6IAnCDiRB2IEkCDuQBGEHkiDsQBKEHUji/wEVWlspoPgKkAAAAABJRU5ErkJggg==",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          },
          "output_type": "display_data"
        }
      ],
      "source": [
        "import cv2\n",
        "image_file_path = r\"/content/sample.png\"\n",
        "image = cv2.imread(image_file_path, cv2.IMREAD_ANYDEPTH)\n",
        "pyplot.imshow(image, cmap=pyplot.get_cmap('gray'))\n",
        "pyplot.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "EULfr6-oYSl5",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "EULfr6-oYSl5",
        "outputId": "59293c8d-469c-4114-8ff5-41f336c8914a"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "array([[  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0, 110, 253, 253, 253, 255, 180, 107,   5,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0, 161, 252, 252, 252, 253, 252, 252,  67,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,  65, 183, 208, 252, 215, 196, 252, 185,   7,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,  17,  45,  21,  47, 252, 252,  22,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,  64, 140, 252, 210,  12,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   9, 128, 202, 253, 253, 150,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "        162, 161, 161, 194, 252, 253, 223, 123,  17,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  32,\n",
              "        253, 252, 252, 252, 252, 122,  25,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  21,\n",
              "        253, 252, 252, 252, 116,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "        137, 137, 137, 137,  12,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,  66,  62,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0, 103,  33,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,  13,  70,  13,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,  34,  82,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,  62, 215, 252,  45,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,  47, 240,  37,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0, 138, 252, 221,  25,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,  47, 252,  45,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0, 139, 253, 131,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,  11,  57, 184,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0, 138, 252, 227,  67,  26,   0,   0,   0,   0,   0,\n",
              "          0,   0,  13,  89, 109, 203, 211,  98,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0, 138, 252, 252, 252, 221, 185, 184, 184, 184, 131,\n",
              "        185, 184, 203, 252, 252, 247,  98,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,  25, 135, 252, 252, 252, 253, 252, 252, 252, 252,\n",
              "        253, 252, 252, 252, 210,  94,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   5,  22, 117, 242, 253, 252, 252, 252, 252,\n",
              "        253, 231, 137,  96,  12,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0],\n",
              "       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,\n",
              "          0,   0]], dtype=uint8)"
            ]
          },
          "execution_count": 106,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "image"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "soLL33IKYhEU",
      "metadata": {
        "id": "soLL33IKYhEU"
      },
      "outputs": [],
      "source": [
        "image=np.reshape(image, [-1, input_size])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "7DI2JjRMa830",
      "metadata": {
        "id": "7DI2JjRMa830"
      },
      "outputs": [],
      "source": [
        "Im_pred = tf.cast(image, tf.float32) / 255.0"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "IUmw0XqBbE9B",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "IUmw0XqBbE9B",
        "outputId": "7d92ce42-01ca-43da-8e70-086c353b7c90"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "1/1 [==============================] - 0s 23ms/step\n"
          ]
        }
      ],
      "source": [
        "Im_pred_y=best_model.predict(Im_pred)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "zbi6ShgBbMuv",
      "metadata": {
        "id": "zbi6ShgBbMuv"
      },
      "outputs": [],
      "source": [
        "Im_pred_y=np.argmax(Im_pred_y, axis=1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "hFesgygMbQ_A",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "hFesgygMbQ_A",
        "outputId": "b70cbe9e-6853-4880-ec04-6a24c5074ad8"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "3"
            ]
          },
          "execution_count": 156,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "Im_pred_y[0]"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "wqswVWr_02V-",
      "metadata": {
        "id": "wqswVWr_02V-"
      },
      "source": [
        "### Export Env"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "avhKdoRz2gM4",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "avhKdoRz2gM4",
        "outputId": "499f2f8b-b627-4843-9373-8bca6e9ce0ed"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Python 3.9.16\n"
          ]
        }
      ],
      "source": [
        "%%python --version"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "uyKbhJ_I01hw",
      "metadata": {
        "id": "uyKbhJ_I01hw"
      },
      "outputs": [],
      "source": [
        "%pip freeze > requirements.txt\n"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "8z_eEt9EbrS-",
      "metadata": {
        "id": "8z_eEt9EbrS-"
      },
      "source": [
        "### Save model\n",
        "\n",
        "\n",
        "> Save model in binary format for serving in a web app.\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "zXVE-zVvcLrw",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zXVE-zVvcLrw",
        "outputId": "c7aa63ed-a0f9-4cdc-beb7-eda585a95b22"
      },
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "WARNING:absl:Found untraced functions such as _update_step_xla while saving (showing 1 of 1). These functions will not be directly callable after loading.\n"
          ]
        }
      ],
      "source": [
        "best_model.save(\"digits_classifier\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "QjjsfPQJeubq",
      "metadata": {
        "id": "QjjsfPQJeubq"
      },
      "outputs": [],
      "source": [
        "loaded_model=keras.models.load_model(\"digits_classifier\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "RsIaIjblkAom",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "id": "RsIaIjblkAom",
        "outputId": "87c8a2ee-a302-4db8-f427-d832d8645d1a"
      },
      "outputs": [
        {
          "data": {
            "application/vnd.google.colaboratory.intrinsic+json": {
              "type": "string"
            },
            "text/plain": [
              "'/content/updated_digits_classifier_archived.zip'"
            ]
          },
          "execution_count": 159,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "import shutil\n",
        "shutil.make_archive(\"updated_digits_classifier_archived\", 'zip', \"digits_classifier\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "u3WdKm4iB-Y_",
      "metadata": {
        "id": "u3WdKm4iB-Y_"
      },
      "outputs": [],
      "source": []
    }
  ],
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 2
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython2",
      "version": "2.7.3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 5
}