v_cnn_support (1).ipynb

{
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3",
      "language": "python"
    },
    "accelerator": "GPU",
    "language_info": {
      "name": "python",
      "version": "3.10.10",
      "mimetype": "text/x-python",
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "pygments_lexer": "ipython3",
      "nbconvert_exporter": "python",
      "file_extension": ".py"
    }
  },
  "nbformat_minor": 0,
  "nbformat": 4,
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "# V-CNN code support for paper:\n",
        "R. Dogaru and Ioana Dogaru, \"V-CNN: A versatile light CNN structure for image recognition on resources-constrained platforms\"\n",
        "Copyright Radu and Ioana Dogaru, contact: radu.dogaru@upb.ro"
      ],
      "metadata": {
        "id": "p7gl-yzOUZZ3"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "#CELL 0 - run only when emnist or svhn datasest are needed\n",
        "\n"
      ],
      "metadata": {
        "id": "7y38gUETcaTp"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install extra-keras-datasets\n",
        "from extra_keras_datasets import emnist, svhn"
      ],
      "metadata": {
        "id": "t0DmlWqvQ_Wo",
        "trusted": true
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# CELL 1  -Loading datsets, scaling and defining input/output shapes  \n",
        "may be replaced with other procedures to access data, e.g. using data generators"
      ],
      "metadata": {
        "id": "SSkvfvnuPHbK"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import tensorflow as tf\n",
        "import numpy as np\n",
        "import keras\n",
        "\n",
        "dataset='cifar10' # mnist or f-mnist or cifar10\n",
        "reduced=0  #  a positive value represents a reduced number of training samples (when memory allocation problems occur)\n",
        "dformat='channels_last'\n",
        "\n",
        "from keras.datasets import mnist, cifar10, cifar100, fashion_mnist\n",
        "\n",
        "\n",
        "if dataset=='mnist':\n",
        "    (x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
        "elif  dataset=='cifar10':\n",
        "    (x_train, y_train), (x_test, y_test) = cifar10.load_data()\n",
        "elif  dataset=='cifar100':\n",
        "    (x_train, y_train), (x_test, y_test) = cifar100.load_data()\n",
        "elif dataset=='f-mnist':\n",
        "    (x_train, y_train), (x_test, y_test) =  fashion_mnist.load_data()\n",
        "elif dataset=='emnist':\n",
        "    (x_train, y_train), (x_test, y_test) =  emnist.load_data()\n",
        "elif dataset=='svhn':\n",
        "    (x_train, y_train), (x_test, y_test) =  svhn.load_data()\n",
        "\n",
        "if (np.ndim(x_train)==3):   # E.g.  MNIST or F-MNIST\n",
        "    x_train=np.reshape(x_train, [np.shape(x_train)[0],np.shape(x_train)[1],np.shape(x_train)[2], 1])\n",
        "    x_test=np.reshape(x_test, [np.shape(x_test)[0],np.shape(x_test)[1],np.shape(x_test)[2], 1] )\n",
        "# place a  1 in the end to keep it compatible with kernel in conv2d\n",
        "# scaling in ([0,1])\n",
        "x_train = x_train.astype('float32')\n",
        "x_test = x_test.astype('float32')\n",
        "x_train /= 255\n",
        "x_test /=255\n",
        "inp_chan=np.shape(x_train)[3]\n",
        "print('Number of input channels in image:', inp_chan)\n",
        "num_classes=int(np.max(y_train)+1)\n",
        "num_inputs = np.shape(x_test)[1]\n",
        "input_shape=np.shape(x_train)[1:4]\n",
        "\n",
        "\n",
        "# if reduced >0\n",
        "if reduced>0:\n",
        "    Ntr1=reduced\n",
        "    x_train=x_train[0:Ntr1,:,:,:]\n",
        "    y_train=y_train[0:Ntr1]\n",
        "y_train = tf.keras.utils.to_categorical(y_train, num_classes)\n",
        "y_test = tf.keras.utils.to_categorical(y_test, num_classes)\n",
        "\n",
        "print('Training samples: ',np.shape(x_train)[0]); print('Esantioane test: ',np.shape(x_test)[0]);\n",
        "print('Input data shape : ', np.shape(x_train)[1], 'x', np.shape(x_train)[2] )\n",
        "print('Number of classes: ',num_classes)"
      ],
      "metadata": {
        "id": "yFFIr2Je52eN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "1f61490e-9fc4-439c-dded-b324a5d594e3",
        "execution": {
          "iopub.status.busy": "2023-06-23T15:12:48.569571Z",
          "iopub.execute_input": "2023-06-23T15:12:48.569951Z",
          "iopub.status.idle": "2023-06-23T15:12:56.172229Z",
          "shell.execute_reply.started": "2023-06-23T15:12:48.569918Z",
          "shell.execute_reply": "2023-06-23T15:12:56.171161Z"
        },
        "trusted": true
      },
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
            "170498071/170498071 [==============================] - 13s 0us/step\n",
            "Number of input channels in image: 3\n",
            "Training samples:  50000\n",
            "Esantioane test:  10000\n",
            "Input data shape :  32 x 32\n",
            "Number of classes:  10\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# CELL 2-   V-CNN model (defining function)"
      ],
      "metadata": {
        "id": "kY20uBAHhZEc"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# V-CNN Versatile CNN model -\n",
        "# includes L-CNN, NL-CNN and XNL-CNN as particular cases.\n",
        "# The basic unit is the \"macro-layer\" as in the XNL-CNN but here one can independently choose the\n",
        "# filter size (fil) and\n",
        "# nonlinearity nl (0 means \"linear\" convolution)\n",
        "# It allows any number of additional dense layers e.g. hid=[] (no hidden dense) or hid =[100, 100] (two additional).\n",
        "# Copyright Radu and Ioana DOGARU - correspondence: radu.dogaru@upb.ro\n",
        "# Last update June 21, 2023\n",
        "#-------------------------------------------------------------------------------------------------\n",
        "\n",
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Dense, Dropout, Flatten, Activation, BatchNormalization\n",
        "from tensorflow.keras.layers import Conv2D, DepthwiseConv2D, MaxPooling2D, AveragePooling2D, GlobalAveragePooling2D, SeparableConv2D  # straturi convolutionale si max-pooling\n",
        "from tensorflow.keras.optimizers import RMSprop, SGD, Adadelta, Adam, Nadam\n",
        "\n",
        "#--------------------------  ------------------------------\n",
        "def create_v_cnn_model(input_shape, num_classes, flat=1, fil=[100,100,100,100], nl=[1,1,0,0], hid=[]):\n",
        "    # Note the number of elements in fil list (macrolayers) should be the same in nl list\n",
        "    # hid can be [] while if the are elements, additional dense layers are added in the output classifier\n",
        "\n",
        "    csize=3; stri=2; psiz=4; pad='same';\n",
        "    drop1=0.6  # Best value for CIFAR-100 after tuning in range 0.25 - 0.75 !\n",
        "\n",
        "    nfilmax=np.shape(np.array(fil))[0]\n",
        "    model = Sequential()\n",
        "    # First macrolayer - connected to input  ----------------\n",
        "    layer=0\n",
        "    if nl[layer]>0:\n",
        "        model.add(Conv2D(fil[layer], padding=pad, kernel_size=(csize, csize), input_shape=input_shape ) )\n",
        "        model.add(Activation('relu'))\n",
        "        for nonlin in range(1,nl[0]):\n",
        "            model.add(Conv2D(fil[layer], padding=pad, kernel_size=(csize, csize)) )\n",
        "            model.add(Activation('relu'))\n",
        "\n",
        "        model.add(Conv2D(fil[0], padding=pad, kernel_size=(csize, csize) ) )\n",
        "        model.add(BatchNormalization())\n",
        "        model.add(MaxPooling2D(pool_size=(psiz, psiz),strides=(stri,stri),padding=pad))\n",
        "        model.add(Dropout(drop1))\n",
        "\n",
        "    else:\n",
        "        model.add(Conv2D(fil[0], padding=pad, kernel_size=(csize, csize), input_shape=input_shape ) )\n",
        "        model.add(BatchNormalization())\n",
        "        model.add(MaxPooling2D(pool_size=(psiz, psiz),strides=(stri,stri),padding=pad))\n",
        "        model.add(Dropout(drop1))\n",
        "    # The remaining  macro-layers\n",
        "\n",
        "    for layer in range(1,nfilmax):\n",
        "        #------------------ nonlin layers -----------------\n",
        "        for nonlin in range(nl[layer]):\n",
        "            model.add(Conv2D(fil[layer], padding=pad, kernel_size=(csize, csize)) )\n",
        "            model.add(Activation('relu'))\n",
        "\n",
        "        #----------------- default macrolayer output\n",
        "\n",
        "        model.add(Conv2D(fil[layer], padding=pad, kernel_size=(csize, csize)) )\n",
        "        model.add(BatchNormalization())\n",
        "        model.add(MaxPooling2D(pool_size=(psiz, psiz),strides=(stri,stri),padding=pad))\n",
        "        model.add(Dropout(drop1))\n",
        "\n",
        "    # Exit classifier\n",
        "    # INPUT TO DENSE LAYER (FLATTEN - more data can overfit / GLOBAL - less data - may be a good choice )\n",
        "    if flat==1:\n",
        "        model.add(Flatten())  # alternanta cu GlobalAv ..\n",
        "    elif flat==0:\n",
        "        model.add(GlobalAveragePooling2D()) # pare sa fie mai Ok la cifar\n",
        "    nhid=np.shape(np.array(hid))[0]\n",
        "    if nhid>0:\n",
        "        for lay in range(nhid):\n",
        "            model.add(Dense(hid[lay], activation='relu'))\n",
        "            #model.add(Dropout(drop1))\n",
        "    model.add(Dense(num_classes, activation='softmax'))\n",
        "\n",
        "# END OF MODEL DESCRIPTION\n",
        "    model.compile(\n",
        "        optimizer=tf.keras.optimizers.Adam(),\n",
        "        loss = 'categorical_crossentropy',\n",
        "        metrics=['accuracy']\n",
        "    )\n",
        "    return model\n",
        "\n"
      ],
      "metadata": {
        "execution": {
          "iopub.status.busy": "2023-06-23T15:13:21.811435Z",
          "iopub.execute_input": "2023-06-23T15:13:21.811793Z",
          "iopub.status.idle": "2023-06-23T15:13:21.835468Z",
          "shell.execute_reply.started": "2023-06-23T15:13:21.811763Z",
          "shell.execute_reply": "2023-06-23T15:13:21.834571Z"
        },
        "trusted": true,
        "id": "k6-R6WOdhZEc"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# CELL 3  - instantiate V-CNN model with its hyper-parameters  "
      ],
      "metadata": {
        "id": "ga5ncMpLjn5j"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#with strategy.scope():   #  For TPU support (initialization of strategy needed)\n",
        "model=create_v_cnn_model(input_shape, num_classes, flat=0, fil=[100, 150,200, 200], nl=[2,1,1,0], hid=[100])\n",
        "model.summary()\n",
        "# The above is a maximal (fat) model tuned for cifar100\n",
        "# Parameters have to adjusted (e.g. by division with a factor )\n",
        "# to cope with data and produce lighter models.\n",
        "# For large image datasets TPU usage is recommended and also adding enough macro-layers"
      ],
      "metadata": {
        "execution": {
          "iopub.status.busy": "2023-06-23T15:13:38.385950Z",
          "iopub.execute_input": "2023-06-23T15:13:38.386327Z",
          "iopub.status.idle": "2023-06-23T15:13:38.716434Z",
          "shell.execute_reply.started": "2023-06-23T15:13:38.386297Z",
          "shell.execute_reply": "2023-06-23T15:13:38.715557Z"
        },
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pBal1oh_hZEd",
        "outputId": "623c274c-98f3-49a5-b28e-9b5c867a0251"
      },
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"sequential_3\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " conv2d_16 (Conv2D)          (None, 32, 32, 100)       2800      \n",
            "                                                                 \n",
            " activation_8 (Activation)   (None, 32, 32, 100)       0         \n",
            "                                                                 \n",
            " conv2d_17 (Conv2D)          (None, 32, 32, 100)       90100     \n",
            "                                                                 \n",
            " activation_9 (Activation)   (None, 32, 32, 100)       0         \n",
            "                                                                 \n",
            " conv2d_18 (Conv2D)          (None, 32, 32, 100)       90100     \n",
            "                                                                 \n",
            " batch_normalization_8 (Batc  (None, 32, 32, 100)      400       \n",
            " hNormalization)                                                 \n",
            "                                                                 \n",
            " max_pooling2d_8 (MaxPooling  (None, 16, 16, 100)      0         \n",
            " 2D)                                                             \n",
            "                                                                 \n",
            " dropout_8 (Dropout)         (None, 16, 16, 100)       0         \n",
            "                                                                 \n",
            " conv2d_19 (Conv2D)          (None, 16, 16, 150)       135150    \n",
            "                                                                 \n",
            " activation_10 (Activation)  (None, 16, 16, 150)       0         \n",
            "                                                                 \n",
            " conv2d_20 (Conv2D)          (None, 16, 16, 150)       202650    \n",
            "                                                                 \n",
            " batch_normalization_9 (Batc  (None, 16, 16, 150)      600       \n",
            " hNormalization)                                                 \n",
            "                                                                 \n",
            " max_pooling2d_9 (MaxPooling  (None, 8, 8, 150)        0         \n",
            " 2D)                                                             \n",
            "                                                                 \n",
            " dropout_9 (Dropout)         (None, 8, 8, 150)         0         \n",
            "                                                                 \n",
            " conv2d_21 (Conv2D)          (None, 8, 8, 200)         270200    \n",
            "                                                                 \n",
            " activation_11 (Activation)  (None, 8, 8, 200)         0         \n",
            "                                                                 \n",
            " conv2d_22 (Conv2D)          (None, 8, 8, 200)         360200    \n",
            "                                                                 \n",
            " batch_normalization_10 (Bat  (None, 8, 8, 200)        800       \n",
            " chNormalization)                                                \n",
            "                                                                 \n",
            " max_pooling2d_10 (MaxPoolin  (None, 4, 4, 200)        0         \n",
            " g2D)                                                            \n",
            "                                                                 \n",
            " dropout_10 (Dropout)        (None, 4, 4, 200)         0         \n",
            "                                                                 \n",
            " conv2d_23 (Conv2D)          (None, 4, 4, 200)         360200    \n",
            "                                                                 \n",
            " batch_normalization_11 (Bat  (None, 4, 4, 200)        800       \n",
            " chNormalization)                                                \n",
            "                                                                 \n",
            " max_pooling2d_11 (MaxPoolin  (None, 2, 2, 200)        0         \n",
            " g2D)                                                            \n",
            "                                                                 \n",
            " dropout_11 (Dropout)        (None, 2, 2, 200)         0         \n",
            "                                                                 \n",
            " global_average_pooling2d_3   (None, 200)              0         \n",
            " (GlobalAveragePooling2D)                                        \n",
            "                                                                 \n",
            " dense_3 (Dense)             (None, 100)               20100     \n",
            "                                                                 \n",
            " dense_4 (Dense)             (None, 10)                1010      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 1,535,110\n",
            "Trainable params: 1,533,810\n",
            "Non-trainable params: 1,300\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# EfficientNet - can be used alternatively for comparsions (after installing the library)\n"
      ],
      "metadata": {
        "id": "PQCiGmL4hZEe"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install -U --pre efficientnet"
      ],
      "metadata": {
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YuLNdIG3hZEe",
        "outputId": "bbc0d18b-1721-4ade-d410-1894aa3fd470"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/\n",
            "Collecting efficientnet\n",
            "  Downloading efficientnet-1.1.1-py3-none-any.whl (18 kB)\n",
            "Collecting keras-applications<=1.0.8,>=1.0.7 (from efficientnet)\n",
            "  Downloading Keras_Applications-1.0.8-py3-none-any.whl (50 kB)\n",
            "\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m50.7/50.7 kB\u001b[0m \u001b[31m4.5 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25hRequirement already satisfied: scikit-image in /usr/local/lib/python3.10/dist-packages (from efficientnet) (0.19.3)\n",
            "Requirement already satisfied: numpy>=1.9.1 in /usr/local/lib/python3.10/dist-packages (from keras-applications<=1.0.8,>=1.0.7->efficientnet) (1.22.4)\n",
            "Requirement already satisfied: h5py in /usr/local/lib/python3.10/dist-packages (from keras-applications<=1.0.8,>=1.0.7->efficientnet) (3.8.0)\n",
            "Requirement already satisfied: scipy>=1.4.1 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (1.10.1)\n",
            "Requirement already satisfied: networkx>=2.2 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (3.1)\n",
            "Requirement already satisfied: pillow!=7.1.0,!=7.1.1,!=8.3.0,>=6.1.0 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (8.4.0)\n",
            "Requirement already satisfied: imageio>=2.4.1 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (2.25.1)\n",
            "Requirement already satisfied: tifffile>=2019.7.26 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (2023.4.12)\n",
            "Requirement already satisfied: PyWavelets>=1.1.1 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (1.4.1)\n",
            "Requirement already satisfied: packaging>=20.0 in /usr/local/lib/python3.10/dist-packages (from scikit-image->efficientnet) (23.1)\n",
            "Installing collected packages: keras-applications, efficientnet\n",
            "Successfully installed efficientnet-1.1.1 keras-applications-1.0.8\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import efficientnet.tfkeras as efn\n",
        "def get_efficientNet_model():\n",
        "    enet = efn.EfficientNetB0(\n",
        "        input_shape=input_shape,\n",
        "        weights='noisy-student',\n",
        "        include_top=False\n",
        "    )\n",
        "\n",
        "    model = tf.keras.Sequential([\n",
        "        enet,\n",
        "        tf.keras.layers.GlobalAveragePooling2D(),\n",
        "        tf.keras.layers.Dense(num_classes, activation='softmax')\n",
        "    ])\n",
        "\n",
        "    return model"
      ],
      "metadata": {
        "trusted": true,
        "id": "-Sc7X3ZxhZEf"
      },
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#with strategy.scope():  # if TPU\n",
        "    # Get the model\n",
        "model = get_efficientNet_model()\n",
        "\n",
        "    # Compile the Model\n",
        "model.compile(\n",
        "        optimizer='adam',\n",
        "        loss = 'categorical_crossentropy',\n",
        "        metrics=['accuracy'],\n",
        "        #steps_per_execution=16\n",
        "    )\n",
        "model.summary()"
      ],
      "metadata": {
        "trusted": true,
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ycUn_FzihZEf",
        "outputId": "6270a330-e1bd-4314-ca95-1224a8e37628"
      },
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://github.com/qubvel/efficientnet/releases/download/v0.0.1/efficientnet-b0_noisy-student_notop.h5\n",
            "16696600/16696600 [==============================] - 1s 0us/step\n",
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " efficientnet-b0 (Functional  (None, 1, 1, 1280)       4049564   \n",
            " )                                                               \n",
            "                                                                 \n",
            " global_average_pooling2d (G  (None, 1280)             0         \n",
            " lobalAveragePooling2D)                                          \n",
            "                                                                 \n",
            " dense (Dense)               (None, 10)                12810     \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 4,062,374\n",
            "Trainable params: 4,020,358\n",
            "Non-trainable params: 42,016\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Augmenting data\n",
        "\n"
      ],
      "metadata": {
        "id": "BPzFBp76OHdy"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "augment=0   # 0 - non-augmented (fast training) 1,2,3 .. various schemes ..\n",
        "# ----------------------\n",
        "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
        "if augment==1:\n",
        "  data_generator = ImageDataGenerator(\n",
        "                        ### --- useful for FER2013 ---\n",
        "\n",
        "                          featurewise_center=False,\n",
        "                          featurewise_std_normalization=False,\n",
        "                          rotation_range=10,\n",
        "                          width_shift_range=0.1,\n",
        "                          height_shift_range=0.1,\n",
        "                          zoom_range=.1,\n",
        "                          horizontal_flip=True\n",
        "                          )\n",
        "\n",
        "                        #------- useful for CIFAR10 according to https://github.com/moritzhambach/Image-Augmentation-in-Keras-CIFAR-10-\n",
        "elif augment==2:\n",
        "  data_generator = ImageDataGenerator(\n",
        "                          rotation_range=15,\n",
        "                          horizontal_flip=True,\n",
        "                          width_shift_range=0.1,\n",
        "                          height_shift_range=0.1\n",
        "                        #------------------------------\n",
        "                         )\n",
        "elif augment==3:      # Proposed in https://github.com/khanhnamle1994/fashion-mnist/blob/master/CNN-4Conv.ipynb\n",
        "  data_generator = ImageDataGenerator(rotation_range=8, width_shift_range=0.08, shear_range=0.3,\n",
        "                               height_shift_range=0.08, zoom_range=0.08)\n",
        "\n",
        "\n"
      ],
      "metadata": {
        "id": "GAeThYodvLkY",
        "execution": {
          "iopub.status.busy": "2023-06-23T15:14:11.967877Z",
          "iopub.execute_input": "2023-06-23T15:14:11.968252Z",
          "iopub.status.idle": "2023-06-23T15:14:11.977019Z",
          "shell.execute_reply.started": "2023-06-23T15:14:11.968223Z",
          "shell.execute_reply": "2023-06-23T15:14:11.976040Z"
        },
        "trusted": true
      },
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Checkpoint for best validation accuracy"
      ],
      "metadata": {
        "id": "zqEskmYzok3K"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from keras.callbacks import ModelCheckpoint\n",
        "from keras.models import load_model\n",
        "import time as ti\n",
        "checkpoint = ModelCheckpoint('best_model.h5', monitor= 'val_accuracy', mode= 'max', save_best_only = True, verbose=1)"
      ],
      "metadata": {
        "execution": {
          "iopub.status.busy": "2023-06-23T15:49:46.180350Z",
          "iopub.execute_input": "2023-06-23T15:49:46.180729Z",
          "iopub.status.idle": "2023-06-23T15:49:46.186287Z",
          "shell.execute_reply.started": "2023-06-23T15:49:46.180697Z",
          "shell.execute_reply": "2023-06-23T15:49:46.185185Z"
        },
        "trusted": true,
        "id": "D0TaappDhZEf"
      },
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Training and validation"
      ],
      "metadata": {
        "id": "ImbNMClXriUG"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "epoci=50\n",
        "batch_size=100\n",
        "\n",
        "t1=ti.time()\n",
        "if augment==0:\n",
        "  history = model.fit(x_train, y_train, epochs=epoci, validation_data=(x_test, y_test), batch_size=batch_size, verbose=2,\n",
        "                    callbacks = checkpoint)\n",
        "elif augment>0:\n",
        "  history = model.fit(data_generator.flow(x_train, y_train,\n",
        "              batch_size=batch_size), epochs=epoci, validation_data=(x_test, y_test), verbose=2,\n",
        "                    callbacks = checkpoint)\n",
        "t2=ti.time()\n",
        "print('====================================================')\n",
        "print('Training with  ',epoci,' epochs, lasted  ',int(t2-t1)/60,' minutes')\n",
        "t1=ti.time()\n",
        "model=load_model('best_model.h5')\n",
        "bp=model.get_weights()  # best weights set\n",
        "score = model.evaluate(x_test, y_test, verbose=0)\n",
        "t2=ti.time()\n",
        "print ('Total number of parameters: ',model.count_params())\n",
        "print('Best validation accuracy :', 100*score[1],'%')\n",
        "#print ('Timp predictie pe tot setul de test: ',t2-t1)\n",
        "print('Latency - GPU (per sample):', 1000*(t2-t1)/np.shape(x_test)[0], 'ms')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "O7Yy4vyvhd2o",
        "outputId": "1b9f7d4e-eab0-4e4f-b85e-c10dec8b434c",
        "trusted": true
      },
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "\n",
            "Epoch 1: val_accuracy improved from -inf to 0.76830, saving model to best_model.h5\n",
            "500/500 - 79s - loss: 1.0188 - accuracy: 0.6512 - val_loss: 0.6837 - val_accuracy: 0.7683 - 79s/epoch - 159ms/step\n",
            "Epoch 2/50\n",
            "\n",
            "Epoch 2: val_accuracy improved from 0.76830 to 0.78740, saving model to best_model.h5\n",
            "500/500 - 20s - loss: 0.5807 - accuracy: 0.8004 - val_loss: 0.6274 - val_accuracy: 0.7874 - 20s/epoch - 40ms/step\n",
            "Epoch 3/50\n",
            "\n",
            "Epoch 3: val_accuracy improved from 0.78740 to 0.80650, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.4455 - accuracy: 0.8455 - val_loss: 0.5769 - val_accuracy: 0.8065 - 19s/epoch - 39ms/step\n",
            "Epoch 4/50\n",
            "\n",
            "Epoch 4: val_accuracy improved from 0.80650 to 0.81240, saving model to best_model.h5\n",
            "500/500 - 20s - loss: 0.3640 - accuracy: 0.8735 - val_loss: 0.6023 - val_accuracy: 0.8124 - 20s/epoch - 39ms/step\n",
            "Epoch 5/50\n",
            "\n",
            "Epoch 5: val_accuracy improved from 0.81240 to 0.82590, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.3017 - accuracy: 0.8932 - val_loss: 0.5493 - val_accuracy: 0.8259 - 19s/epoch - 39ms/step\n",
            "Epoch 6/50\n",
            "\n",
            "Epoch 6: val_accuracy did not improve from 0.82590\n",
            "500/500 - 19s - loss: 0.2534 - accuracy: 0.9109 - val_loss: 0.5844 - val_accuracy: 0.8247 - 19s/epoch - 39ms/step\n",
            "Epoch 7/50\n",
            "\n",
            "Epoch 7: val_accuracy did not improve from 0.82590\n",
            "500/500 - 19s - loss: 0.2238 - accuracy: 0.9222 - val_loss: 0.5952 - val_accuracy: 0.8215 - 19s/epoch - 38ms/step\n",
            "Epoch 8/50\n",
            "\n",
            "Epoch 8: val_accuracy improved from 0.82590 to 0.82690, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.1987 - accuracy: 0.9306 - val_loss: 0.6134 - val_accuracy: 0.8269 - 19s/epoch - 39ms/step\n",
            "Epoch 9/50\n",
            "\n",
            "Epoch 9: val_accuracy improved from 0.82690 to 0.83080, saving model to best_model.h5\n",
            "500/500 - 20s - loss: 0.1697 - accuracy: 0.9402 - val_loss: 0.6400 - val_accuracy: 0.8308 - 20s/epoch - 40ms/step\n",
            "Epoch 10/50\n",
            "\n",
            "Epoch 10: val_accuracy did not improve from 0.83080\n",
            "500/500 - 19s - loss: 0.1459 - accuracy: 0.9500 - val_loss: 0.7091 - val_accuracy: 0.8144 - 19s/epoch - 38ms/step\n",
            "Epoch 11/50\n",
            "\n",
            "Epoch 11: val_accuracy did not improve from 0.83080\n",
            "500/500 - 19s - loss: 0.1372 - accuracy: 0.9537 - val_loss: 0.6741 - val_accuracy: 0.8177 - 19s/epoch - 38ms/step\n",
            "Epoch 12/50\n",
            "\n",
            "Epoch 12: val_accuracy improved from 0.83080 to 0.83550, saving model to best_model.h5\n",
            "500/500 - 20s - loss: 0.1209 - accuracy: 0.9587 - val_loss: 0.6665 - val_accuracy: 0.8355 - 20s/epoch - 40ms/step\n",
            "Epoch 13/50\n",
            "\n",
            "Epoch 13: val_accuracy did not improve from 0.83550\n",
            "500/500 - 19s - loss: 0.1116 - accuracy: 0.9611 - val_loss: 0.7138 - val_accuracy: 0.8274 - 19s/epoch - 37ms/step\n",
            "Epoch 14/50\n",
            "\n",
            "Epoch 14: val_accuracy did not improve from 0.83550\n",
            "500/500 - 19s - loss: 0.1083 - accuracy: 0.9631 - val_loss: 0.6774 - val_accuracy: 0.8303 - 19s/epoch - 38ms/step\n",
            "Epoch 15/50\n",
            "\n",
            "Epoch 15: val_accuracy improved from 0.83550 to 0.83620, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.0969 - accuracy: 0.9668 - val_loss: 0.7115 - val_accuracy: 0.8362 - 19s/epoch - 38ms/step\n",
            "Epoch 16/50\n",
            "\n",
            "Epoch 16: val_accuracy did not improve from 0.83620\n",
            "500/500 - 19s - loss: 0.1019 - accuracy: 0.9646 - val_loss: 0.6805 - val_accuracy: 0.8300 - 19s/epoch - 38ms/step\n",
            "Epoch 17/50\n",
            "\n",
            "Epoch 17: val_accuracy did not improve from 0.83620\n",
            "500/500 - 18s - loss: 0.0856 - accuracy: 0.9703 - val_loss: 0.7075 - val_accuracy: 0.8335 - 18s/epoch - 36ms/step\n",
            "Epoch 18/50\n",
            "\n",
            "Epoch 18: val_accuracy did not improve from 0.83620\n",
            "500/500 - 19s - loss: 0.0888 - accuracy: 0.9702 - val_loss: 0.7457 - val_accuracy: 0.8307 - 19s/epoch - 38ms/step\n",
            "Epoch 19/50\n",
            "\n",
            "Epoch 19: val_accuracy did not improve from 0.83620\n",
            "500/500 - 18s - loss: 0.0884 - accuracy: 0.9698 - val_loss: 0.7786 - val_accuracy: 0.8225 - 18s/epoch - 37ms/step\n",
            "Epoch 20/50\n",
            "\n",
            "Epoch 20: val_accuracy improved from 0.83620 to 0.83730, saving model to best_model.h5\n",
            "500/500 - 20s - loss: 0.0776 - accuracy: 0.9743 - val_loss: 0.7598 - val_accuracy: 0.8373 - 20s/epoch - 39ms/step\n",
            "Epoch 21/50\n",
            "\n",
            "Epoch 21: val_accuracy did not improve from 0.83730\n",
            "500/500 - 18s - loss: 0.0766 - accuracy: 0.9739 - val_loss: 0.6910 - val_accuracy: 0.8372 - 18s/epoch - 37ms/step\n",
            "Epoch 22/50\n",
            "\n",
            "Epoch 22: val_accuracy improved from 0.83730 to 0.84270, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.0712 - accuracy: 0.9759 - val_loss: 0.7265 - val_accuracy: 0.8427 - 19s/epoch - 38ms/step\n",
            "Epoch 23/50\n",
            "\n",
            "Epoch 23: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0880 - accuracy: 0.9710 - val_loss: 0.7157 - val_accuracy: 0.8393 - 19s/epoch - 38ms/step\n",
            "Epoch 24/50\n",
            "\n",
            "Epoch 24: val_accuracy did not improve from 0.84270\n",
            "500/500 - 18s - loss: 0.0688 - accuracy: 0.9772 - val_loss: 0.7292 - val_accuracy: 0.8386 - 18s/epoch - 37ms/step\n",
            "Epoch 25/50\n",
            "\n",
            "Epoch 25: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0674 - accuracy: 0.9758 - val_loss: 0.7071 - val_accuracy: 0.8390 - 19s/epoch - 38ms/step\n",
            "Epoch 26/50\n",
            "\n",
            "Epoch 26: val_accuracy did not improve from 0.84270\n",
            "500/500 - 18s - loss: 0.0595 - accuracy: 0.9799 - val_loss: 0.7609 - val_accuracy: 0.8414 - 18s/epoch - 37ms/step\n",
            "Epoch 27/50\n",
            "\n",
            "Epoch 27: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0572 - accuracy: 0.9808 - val_loss: 0.8013 - val_accuracy: 0.8326 - 19s/epoch - 37ms/step\n",
            "Epoch 28/50\n",
            "\n",
            "Epoch 28: val_accuracy did not improve from 0.84270\n",
            "500/500 - 18s - loss: 0.0654 - accuracy: 0.9774 - val_loss: 0.7209 - val_accuracy: 0.8414 - 18s/epoch - 37ms/step\n",
            "Epoch 29/50\n",
            "\n",
            "Epoch 29: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0569 - accuracy: 0.9807 - val_loss: 0.7736 - val_accuracy: 0.8352 - 19s/epoch - 38ms/step\n",
            "Epoch 30/50\n",
            "\n",
            "Epoch 30: val_accuracy did not improve from 0.84270\n",
            "500/500 - 18s - loss: 0.0588 - accuracy: 0.9805 - val_loss: 0.7507 - val_accuracy: 0.8401 - 18s/epoch - 37ms/step\n",
            "Epoch 31/50\n",
            "\n",
            "Epoch 31: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0545 - accuracy: 0.9817 - val_loss: 0.7617 - val_accuracy: 0.8401 - 19s/epoch - 38ms/step\n",
            "Epoch 32/50\n",
            "\n",
            "Epoch 32: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0729 - accuracy: 0.9763 - val_loss: 0.8036 - val_accuracy: 0.8279 - 19s/epoch - 37ms/step\n",
            "Epoch 33/50\n",
            "\n",
            "Epoch 33: val_accuracy did not improve from 0.84270\n",
            "500/500 - 18s - loss: 0.0572 - accuracy: 0.9812 - val_loss: 0.7613 - val_accuracy: 0.8405 - 18s/epoch - 37ms/step\n",
            "Epoch 34/50\n",
            "\n",
            "Epoch 34: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0463 - accuracy: 0.9849 - val_loss: 0.7942 - val_accuracy: 0.8344 - 19s/epoch - 38ms/step\n",
            "Epoch 35/50\n",
            "\n",
            "Epoch 35: val_accuracy did not improve from 0.84270\n",
            "500/500 - 18s - loss: 0.0482 - accuracy: 0.9842 - val_loss: 0.8279 - val_accuracy: 0.8335 - 18s/epoch - 36ms/step\n",
            "Epoch 36/50\n",
            "\n",
            "Epoch 36: val_accuracy did not improve from 0.84270\n",
            "500/500 - 19s - loss: 0.0520 - accuracy: 0.9828 - val_loss: 0.8439 - val_accuracy: 0.8305 - 19s/epoch - 38ms/step\n",
            "Epoch 37/50\n",
            "\n",
            "Epoch 37: val_accuracy improved from 0.84270 to 0.84380, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.0453 - accuracy: 0.9850 - val_loss: 0.7195 - val_accuracy: 0.8438 - 19s/epoch - 38ms/step\n",
            "Epoch 38/50\n",
            "\n",
            "Epoch 38: val_accuracy did not improve from 0.84380\n",
            "500/500 - 19s - loss: 0.0482 - accuracy: 0.9841 - val_loss: 0.7923 - val_accuracy: 0.8426 - 19s/epoch - 39ms/step\n",
            "Epoch 39/50\n",
            "\n",
            "Epoch 39: val_accuracy did not improve from 0.84380\n",
            "500/500 - 18s - loss: 0.0498 - accuracy: 0.9839 - val_loss: 0.7954 - val_accuracy: 0.8398 - 18s/epoch - 36ms/step\n",
            "Epoch 40/50\n",
            "\n",
            "Epoch 40: val_accuracy did not improve from 0.84380\n",
            "500/500 - 18s - loss: 0.0805 - accuracy: 0.9746 - val_loss: 0.7078 - val_accuracy: 0.8416 - 18s/epoch - 36ms/step\n",
            "Epoch 41/50\n",
            "\n",
            "Epoch 41: val_accuracy did not improve from 0.84380\n",
            "500/500 - 18s - loss: 0.0451 - accuracy: 0.9852 - val_loss: 0.8910 - val_accuracy: 0.8278 - 18s/epoch - 35ms/step\n",
            "Epoch 42/50\n",
            "\n",
            "Epoch 42: val_accuracy improved from 0.84380 to 0.84470, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.0410 - accuracy: 0.9868 - val_loss: 0.7802 - val_accuracy: 0.8447 - 19s/epoch - 38ms/step\n",
            "Epoch 43/50\n",
            "\n",
            "Epoch 43: val_accuracy did not improve from 0.84470\n",
            "500/500 - 17s - loss: 0.0369 - accuracy: 0.9880 - val_loss: 0.8363 - val_accuracy: 0.8397 - 17s/epoch - 35ms/step\n",
            "Epoch 44/50\n",
            "\n",
            "Epoch 44: val_accuracy improved from 0.84470 to 0.84540, saving model to best_model.h5\n",
            "500/500 - 19s - loss: 0.0463 - accuracy: 0.9849 - val_loss: 0.7826 - val_accuracy: 0.8454 - 19s/epoch - 38ms/step\n",
            "Epoch 45/50\n",
            "\n",
            "Epoch 45: val_accuracy did not improve from 0.84540\n",
            "500/500 - 17s - loss: 0.0370 - accuracy: 0.9878 - val_loss: 0.8154 - val_accuracy: 0.8391 - 17s/epoch - 35ms/step\n",
            "Epoch 46/50\n",
            "\n",
            "Epoch 46: val_accuracy did not improve from 0.84540\n",
            "500/500 - 18s - loss: 0.0422 - accuracy: 0.9862 - val_loss: 0.8273 - val_accuracy: 0.8337 - 18s/epoch - 35ms/step\n",
            "Epoch 47/50\n",
            "\n",
            "Epoch 47: val_accuracy improved from 0.84540 to 0.84750, saving model to best_model.h5\n",
            "500/500 - 18s - loss: 0.0471 - accuracy: 0.9844 - val_loss: 0.7725 - val_accuracy: 0.8475 - 18s/epoch - 37ms/step\n",
            "Epoch 48/50\n",
            "\n",
            "Epoch 48: val_accuracy did not improve from 0.84750\n",
            "500/500 - 18s - loss: 0.0392 - accuracy: 0.9876 - val_loss: 0.7690 - val_accuracy: 0.8467 - 18s/epoch - 36ms/step\n",
            "Epoch 49/50\n",
            "\n",
            "Epoch 49: val_accuracy did not improve from 0.84750\n",
            "500/500 - 17s - loss: 0.0346 - accuracy: 0.9884 - val_loss: 0.7921 - val_accuracy: 0.8436 - 17s/epoch - 35ms/step\n",
            "Epoch 50/50\n",
            "\n",
            "Epoch 50: val_accuracy did not improve from 0.84750\n",
            "500/500 - 17s - loss: 0.0350 - accuracy: 0.9882 - val_loss: 0.8485 - val_accuracy: 0.8312 - 17s/epoch - 35ms/step\n",
            "====================================================\n",
            "Training with   50  epochs, lasted   16.666666666666668  minutes\n",
            "Total number of parameters:  4062374\n",
            "Best validation accuracy : 84.75000262260437 %\n",
            "Latency - GPU (per sample): 0.9881463766098022 ms\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Plots of training/validation process"
      ],
      "metadata": {
        "id": "Rc1AbXXwmQPj"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from matplotlib import pyplot as plt\n",
        "def display_training_curves(training, validation, title, subplot):\n",
        "    if subplot%10==1: # set up the subplots on the first call\n",
        "        plt.subplots(figsize=(7,3), facecolor='#F0F0F0')\n",
        "        plt.tight_layout()\n",
        "    ax = plt.subplot(subplot)\n",
        "    ax.set_facecolor('#F8F8F8')\n",
        "    ax.plot(training)\n",
        "    ax.plot(validation)\n",
        "    ax.set_title('model '+ title)\n",
        "    ax.set_ylabel(title)\n",
        "    #ax.set_ylim(0.28,1.05)\n",
        "    ax.set_xlabel('epoch')\n",
        "    ax.legend(['train', 'valid.'])\n",
        "display_training_curves(history.history['loss'], history.history['val_loss'], 'loss', 121)\n",
        "display_training_curves(history.history['accuracy'], history.history['val_accuracy'], 'accuracy', 122)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 376
        },
        "id": "45_UlH9tHH1E",
        "outputId": "a807a6c6-9be8-4b02-bfb1-383a1e9000f8",
        "trusted": true
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "<ipython-input-9-951775051a26>:6: MatplotlibDeprecationWarning: Auto-removal of overlapping axes is deprecated since 3.6 and will be removed two minor releases later; explicitly call ax.remove() as needed.\n",
            "  ax = plt.subplot(subplot)\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 700x300 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Saving model\n",
        "#-----------------------------------------------------------\n",
        "model.save('cif10-90_78.h5')  # use your favourite name"
      ],
      "metadata": {
        "id": "7n4z2_eh_9QV",
        "trusted": true
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Evaluate a savde model\n",
        "from keras.models import load_model\n",
        "import time as ti\n",
        "\n",
        "mod=load_model('specify-it')  # fisierul trebuie \"upload\" in sectiunea fisiere\n",
        "mod.summary()\n",
        "t1=ti.time()\n",
        "score = mod.evaluate(x_test, y_test, verbose=0)\n",
        "t2=ti.time()\n",
        "print('Latency (per input sample):', 1000*(t2-t1)/np.shape(x_test)[0], 'ms')\n",
        "print('Validation accuracy: ',100*score[1],'%')"
      ],
      "metadata": {
        "id": "Dpahjo4aCLTo",
        "trusted": true
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# Examples\n",
        "\n",
        "selectia=int(np.shape(x_test)[0]*np.random.rand(1))  #\n",
        "\n",
        "img1=x_test[selectia,:,:,:]\n",
        "import matplotlib.pyplot as plt\n",
        "import matplotlib.image as mpimg\n",
        "if np.shape(x_train)[3]==1:\n",
        "  plt.imshow(img1[:,:,0],cmap='gray')\n",
        "elif np.shape(x_train)[3]==3:\n",
        "  plt.imshow(img1)\n",
        "\n",
        "label=np.dot(y_test[selectia,:],1+np.array(range(num_classes)).T)\n",
        "print('Labeled class :', label-1)\n",
        "z=model.predict(x_test[selectia:selectia+1,:,:,:])\n",
        "print('Recognized class  : ',np.argmax(z))\n",
        "\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "dd_6qiBEeaQj",
        "outputId": "659d4c90-2710-4d94-9066-bda3642840d3",
        "trusted": true
      },
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Labeled class : 9.0\n",
            "1/1 [==============================] - 0s 24ms/step\n",
            "Recognized class  :  9\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Confusion matrix and more specific performance indicators\n",
        "#----------------------------------------------------------------\n",
        "\n",
        "labels=(np.dot(y_test,np.array(range(num_classes)).T)).astype('int16')\n",
        "pred= model.predict(x_test)\n",
        "predicted_class_indices=np.argmax(pred,axis=1)\n",
        "from sklearn.metrics import classification_report, confusion_matrix\n",
        "C=confusion_matrix(predicted_class_indices,labels)\n",
        "print (C)\n",
        "print('Classification Report')\n",
        "print(classification_report(labels,predicted_class_indices ))\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xAjHg2IHetNK",
        "outputId": "c8d1e507-bc27-4e71-eca8-81f4d26a5923",
        "trusted": true
      },
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 [==============================] - 4s 8ms/step\n",
            "[[884  15  42  13   8   7   4   7  35  16]\n",
            " [  9 894   1   6   2   1   2   1  13  42]\n",
            " [ 12   1 768  37  37  29  13   6   7   1]\n",
            " [ 10   6  37 712  27 136  24  35  11   7]\n",
            " [ 10   2  52  53 857  29  17  35   4   4]\n",
            " [  3   1  13  90  13 740  10  17   3   2]\n",
            " [  3   4  58  52  25  23 925   4   5   1]\n",
            " [  9   3  19  21  24  33   3 887   2   7]\n",
            " [ 33  19   6   3   2   1   0   4 898  10]\n",
            " [ 27  55   4  13   5   1   2   4  22 910]]\n",
            "Classification Report\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.86      0.88      0.87      1000\n",
            "           1       0.92      0.89      0.91      1000\n",
            "           2       0.84      0.77      0.80      1000\n",
            "           3       0.71      0.71      0.71      1000\n",
            "           4       0.81      0.86      0.83      1000\n",
            "           5       0.83      0.74      0.78      1000\n",
            "           6       0.84      0.93      0.88      1000\n",
            "           7       0.88      0.89      0.88      1000\n",
            "           8       0.92      0.90      0.91      1000\n",
            "           9       0.87      0.91      0.89      1000\n",
            "\n",
            "    accuracy                           0.85     10000\n",
            "   macro avg       0.85      0.85      0.85     10000\n",
            "weighted avg       0.85      0.85      0.85     10000\n",
            "\n"
          ]
        }
      ]
    }
  ]
}