Disease_Classification.ipynb

{
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "LzJNqVmRr-5b",
        "outputId": "c48ee40f-91c3-4313-d985-92184aec02ae"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /content/drive\n"
          ]
        }
      ],
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "id": "PU3yss0IwOPs"
      },
      "outputs": [],
      "source": [
        "import csv\n",
        "import pandas as pd\n",
        "import scipy.io\n",
        "import numpy as np\n",
        "import copy\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "import h5py\n",
        "import scipy\n",
        "from PIL import Image\n",
        "from scipy import ndimage\n",
        "\n",
        "from tensorflow import keras\n",
        "import tensorflow as tf\n",
        "\n",
        "%matplotlib inline\n",
        "%load_ext autoreload\n",
        "%autoreload 2"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!pip install watermark\n",
        "%load_ext watermark\n",
        "\n",
        "%watermark --iversions "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QsKVEMsOJr2R",
        "outputId": "c6d3d614-d37d-4c4c-b194-a72a2cf0e145"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "numpy     : 1.21.6\n",
            "matplotlib: 3.2.2\n",
            "tensorflow: 2.8.0+zzzcolab20220506162203\n",
            "h5py      : 3.1.0\n",
            "keras     : 2.8.0\n",
            "PIL       : 7.1.2\n",
            "sklearn   : 0.0\n",
            "google    : 2.0.3\n",
            "scipy     : 1.4.1\n",
            "IPython   : 5.5.0\n",
            "pandas    : 1.3.5\n",
            "csv       : 1.0\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DNAY6K38-84F"
      },
      "source": [
        "#Disease Classification"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hJf0Us5pwVjV"
      },
      "outputs": [],
      "source": [
        "bal_all = np.loadtxt('/content/drive/MyDrive/ML/Disease_classification/Dataset/both_phases.csv', delimiter=',')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bDjSpN-fwWCI",
        "outputId": "094350be-6912-4490-c182-51ab70f6c1e8"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "(47790, 1024)\n",
            "(47790,)\n",
            "(47790, 1)\n"
          ]
        }
      ],
      "source": [
        "x_bal = bal_all[:,0:1024]\n",
        "y_disease = bal_all[:,-1]\n",
        "print(np.shape(x_bal))\n",
        "print(np.shape(y_disease))\n",
        "y_disease=y_disease.reshape(-1,1 )\n",
        "print(np.shape(y_disease))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PokYvcoMwZir",
        "outputId": "8c4c8d97-569e-4da4-e181-fc02ab0ed49e"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "0.0\n",
            "[1. 0. 0. 0. 0.]\n"
          ]
        }
      ],
      "source": [
        "from sklearn.model_selection import train_test_split \n",
        "from sklearn.preprocessing import OneHotEncoder\n",
        "\n",
        "print(y_disease[8500,0])\n",
        "enc = OneHotEncoder(handle_unknown='ignore')\n",
        "# 2. FIT\n",
        "enc.fit(y_disease)\n",
        "\n",
        "# 3. Transform\n",
        "onehotlabels = enc.transform(y_disease).toarray()\n",
        "\n",
        "print(onehotlabels[8500])\n",
        "\n",
        "x_train1, x_test1, y_train1, y_test1 = train_test_split(x_bal, onehotlabels, test_size = 0.2, shuffle = True)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TXzOVw_3_Fe0"
      },
      "source": [
        "## 1D CNN"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "IU2Ofo-SYEh_"
      },
      "outputs": [],
      "source": [
        "# GRADED FUNCTION: convolutional_model\n",
        "import tensorflow as tf\n",
        "from keras.callbacks import EarlyStopping\n",
        "import tensorflow.keras.layers as tfl\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.layers import Dropout\n",
        "from keras.utils.vis_utils import plot_model\n",
        "from keras.layers.merge import concatenate\n",
        "from time import time\n",
        "\n",
        "def disease_classification(input_shape, x_train,y_train,test_data):\n",
        "    \"\"\"\n",
        "    \n",
        "    Arguments:\n",
        "    input_shape\n",
        "\n",
        "    Returns:\n",
        "    model -- TF Keras model (object containing the information for the entire training process) \n",
        "    history\n",
        "    \"\"\"\n",
        "\n",
        "    start = time()\n",
        "\n",
        "    input_img = tf.keras.Input(shape=input_shape)\n",
        "    ## CONV1D: 8 filters 4x4, stride of 1, padding 'SAME'\n",
        "    Z1 = tfl.Conv1D(filters=25,\n",
        "                    kernel_size=15,\n",
        "                    strides=1,\n",
        "                    padding='same')(input_img) \n",
        "    Z11 = tfl.Conv1D(filters=25,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z1)\n",
        "    #Z11op=tf.nn.dropout(Z11, rate = 0.2)\n",
        "    ## RELU\n",
        "    A1 = tfl.ReLU()(Z11)\n",
        "    ## MAXPOOL: window 8x8, stride 8, padding 'SAME'\n",
        "    P1 = tfl.MaxPooling1D(pool_size= 5,\n",
        "                       padding='same')(A1)\n",
        "   \n",
        "    ## CONV2D: 16 filters 3x3, stride 1, padding 'SAME'\n",
        "    Z2 = tfl.Conv1D(filters=15,\n",
        "                    kernel_size=15,\n",
        "                    strides=1,\n",
        "                    padding='same')(P1)\n",
        "    Z21 = tfl.Conv1D(filters=15,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z2)\n",
        "    #Z21op=tf.nn.dropout(Z21, rate = 0.2)\n",
        "    ## RELU\n",
        "    A2 = tfl.ReLU()(Z21)\n",
        "    ## MAXPOOL: window 4x4, stride 4, padding 'SAME'\n",
        "    P2 = tfl.MaxPooling1D(pool_size=5,\n",
        "                       padding='same')(A2)\n",
        "    \n",
        "\n",
        "    Z3 = tfl.Conv1D(filters=10,\n",
        "                    kernel_size=15,\n",
        "                    strides=1,\n",
        "                    padding='same')(P2)\n",
        "    Z31 = tfl.Conv1D(filters=10,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z3)\n",
        "    Z31=tf.nn.dropout(Z31, rate = 0.2)\n",
        "    ## RELU\n",
        "    A3 = tfl.ReLU()(Z31)\n",
        "    ## MAXPOOL: window 4x4, stride 4, padding 'SAME'\n",
        "    P3 = tfl.MaxPooling1D(pool_size=5,\n",
        "                       padding='same')(A3)\n",
        "\n",
        "\n",
        "    ## FLATTEN\n",
        "    F = tfl.Flatten()(P3)\n",
        "    ## Dense layer\n",
        "    op = tfl.Dense(units=128,\n",
        "                        activation='relu')(F)\n",
        "    op=tf.nn.dropout(op, rate = 0.2)\n",
        "\n",
        "    op1 = tfl.Dense(units=48,\n",
        "                        activation='relu')(op)\n",
        "\n",
        "    op2 = tfl.Dense(units=16,\n",
        "                        activation='relu')(op1)\n",
        "          \n",
        "    outputs = tfl.Dense(units=5,\n",
        "                        activation='softmax')(op2)\n",
        "\n",
        "    model = tf.keras.Model(inputs=input_img, outputs=outputs)\n",
        "\n",
        "    #compile\n",
        "    model.compile(optimizer = 'adam',\n",
        "                  loss='categorical_crossentropy',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "    model.summary()\n",
        "\n",
        "    early_stopping = EarlyStopping(monitor='val_loss', patience=15, verbose=2)\n",
        "    history_m = model.fit(x_train,y_train, epochs=200, validation_split =0.2, callbacks = [early_stopping], batch_size = 75, shuffle = True)\n",
        "\n",
        "    t=time()-start\n",
        "    print('Total training time:', t)\n",
        "    history_m.history['training_time']=t\n",
        "    \n",
        "    return model, history_m"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9Nk_Md-_Rtix"
      },
      "source": [
        "##Inception Model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "1494jVNdXRAs"
      },
      "outputs": [],
      "source": [
        "from tensorflow import keras\n",
        "from keras.callbacks import EarlyStopping\n",
        "from tensorflow import keras\n",
        "import tensorflow.keras.layers as tfl\n",
        "from tensorflow.keras.layers import *\n",
        "from tensorflow.keras.utils import *\n",
        "from time import time\n",
        "import tensorflow.keras.layers as tfl\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.layers import Dropout\n",
        "from keras.utils.vis_utils import plot_model\n",
        "from keras.layers.merge import concatenate\n",
        "from time import time\n",
        "\n",
        "def disease_classification2(input_shape, x_train,y_train,test_data):\n",
        "    \"\"\"\n",
        "    \n",
        "    Arguments:\n",
        "    input_shape\n",
        "\n",
        "    Returns:\n",
        "    model -- TF Keras model (object containing the information for the entire training process) \n",
        "    history\n",
        "    \"\"\"\n",
        "\n",
        "    start = time()\n",
        "\n",
        "    input_img = tf.keras.Input(shape=input_shape)\n",
        "    ## CONV1D: 8 filters 4x4, stride of 1, padding 'SAME'\n",
        "    Z1 = tfl.Conv1D(filters=100,\n",
        "                    kernel_size=25,\n",
        "                    strides=1,\n",
        "                    padding='same')(input_img) \n",
        "    M1 = MaxPool1D()(Z1)\n",
        "    Z2 = tfl.Conv1D(filters=80,\n",
        "                    kernel_size=40,\n",
        "                    strides=1,\n",
        "                    padding='same')(M1) \n",
        "\n",
        "    M2 = MaxPool1D()(Z2)\n",
        "    Z3 = tfl.Conv1D(filters=60,\n",
        "                    kernel_size=30,\n",
        "                    strides=1,\n",
        "                    padding='same')(M2) \n",
        "\n",
        "    M3 = MaxPool1D()(Z3)\n",
        "\n",
        "    Z4 = tfl.Conv1D(filters=40,\n",
        "                    kernel_size=20,\n",
        "                    strides=1,\n",
        "                    padding='same')(M3) \n",
        "\n",
        "    M4 = MaxPool1D()(Z4)\n",
        "\n",
        "    gru = GRU(40, return_sequences=True)(M4)\n",
        "    gru1 = GRU(40, return_sequences=True)(gru)\n",
        "\n",
        "    gru1=tf.nn.dropout(gru1, rate = 0.25)\n",
        "    \n",
        "    F = tfl.Flatten()(gru1)\n",
        "    ## Dense layer\n",
        "  \n",
        "\n",
        "    \n",
        "    outputs = tfl.Dense(units=5,\n",
        "                        activation='softmax')(F)\n",
        "\n",
        "    model = tf.keras.Model(inputs=input_img, outputs=outputs)\n",
        "\n",
        "    #compile\n",
        "    model.compile(optimizer = 'adam',\n",
        "                  loss='categorical_crossentropy',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "    model.summary()\n",
        "\n",
        "    early_stopping = EarlyStopping(monitor='val_loss', patience=15, verbose=2)\n",
        "    history_m = model.fit(x_train,y_train, epochs=1, validation_split =0.2, callbacks = [early_stopping], batch_size = 75, shuffle = True)\n",
        "\n",
        "    t=time()-start\n",
        "    print('Total training time:', t)\n",
        "    history_m.history['training_time']=t\n",
        "    \n",
        "    return model, history_m\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nlXrWtfxYYFp",
        "outputId": "9730a02b-4245-48a1-e680-5ba7a35d4bd9"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model_1\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input_2 (InputLayer)        [(None, 1024, 1)]         0         \n",
            "                                                                 \n",
            " conv1d_4 (Conv1D)           (None, 1024, 100)         2600      \n",
            "                                                                 \n",
            " max_pooling1d_4 (MaxPooling  (None, 512, 100)         0         \n",
            " 1D)                                                             \n",
            "                                                                 \n",
            " conv1d_5 (Conv1D)           (None, 512, 80)           320080    \n",
            "                                                                 \n",
            " max_pooling1d_5 (MaxPooling  (None, 256, 80)          0         \n",
            " 1D)                                                             \n",
            "                                                                 \n",
            " conv1d_6 (Conv1D)           (None, 256, 60)           144060    \n",
            "                                                                 \n",
            " max_pooling1d_6 (MaxPooling  (None, 128, 60)          0         \n",
            " 1D)                                                             \n",
            "                                                                 \n",
            " conv1d_7 (Conv1D)           (None, 128, 40)           48040     \n",
            "                                                                 \n",
            " max_pooling1d_7 (MaxPooling  (None, 64, 40)           0         \n",
            " 1D)                                                             \n",
            "                                                                 \n",
            " gru_2 (GRU)                 (None, 64, 40)            9840      \n",
            "                                                                 \n",
            " gru_3 (GRU)                 (None, 64, 40)            9840      \n",
            "                                                                 \n",
            " tf.nn.dropout_2 (TFOpLambda  (None, 64, 40)           0         \n",
            " )                                                               \n",
            "                                                                 \n",
            " flatten_1 (Flatten)         (None, 2560)              0         \n",
            "                                                                 \n",
            " dense_1 (Dense)             (None, 5)                 12805     \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 547,265\n",
            "Trainable params: 547,265\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "Epoch 1/200\n",
            "408/408 [==============================] - 47s 92ms/step - loss: 1.6923 - accuracy: 0.2094 - val_loss: 1.6453 - val_accuracy: 0.2108\n",
            "Epoch 2/200\n",
            "408/408 [==============================] - 35s 87ms/step - loss: 1.6303 - accuracy: 0.2265 - val_loss: 1.6262 - val_accuracy: 0.2266\n",
            "Epoch 3/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.6289 - accuracy: 0.2258 - val_loss: 1.6302 - val_accuracy: 0.2188\n",
            "Epoch 4/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.6159 - accuracy: 0.2290 - val_loss: 1.6223 - val_accuracy: 0.2257\n",
            "Epoch 5/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.6073 - accuracy: 0.2385 - val_loss: 1.6073 - val_accuracy: 0.2328\n",
            "Epoch 6/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.6027 - accuracy: 0.2451 - val_loss: 1.6027 - val_accuracy: 0.2371\n",
            "Epoch 7/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.6009 - accuracy: 0.2456 - val_loss: 1.6037 - val_accuracy: 0.2431\n",
            "Epoch 8/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5966 - accuracy: 0.2511 - val_loss: 1.6020 - val_accuracy: 0.2388\n",
            "Epoch 9/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5921 - accuracy: 0.2551 - val_loss: 1.6013 - val_accuracy: 0.2435\n",
            "Epoch 10/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5959 - accuracy: 0.2525 - val_loss: 1.5997 - val_accuracy: 0.2504\n",
            "Epoch 11/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5953 - accuracy: 0.2532 - val_loss: 1.5991 - val_accuracy: 0.2462\n",
            "Epoch 12/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5954 - accuracy: 0.2539 - val_loss: 1.5991 - val_accuracy: 0.2387\n",
            "Epoch 13/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5963 - accuracy: 0.2524 - val_loss: 1.5997 - val_accuracy: 0.2496\n",
            "Epoch 14/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5960 - accuracy: 0.2501 - val_loss: 1.6024 - val_accuracy: 0.2343\n",
            "Epoch 15/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.6003 - accuracy: 0.2474 - val_loss: 1.5987 - val_accuracy: 0.2418\n",
            "Epoch 16/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5988 - accuracy: 0.2447 - val_loss: 1.6008 - val_accuracy: 0.2401\n",
            "Epoch 17/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5949 - accuracy: 0.2564 - val_loss: 1.5957 - val_accuracy: 0.2506\n",
            "Epoch 18/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5926 - accuracy: 0.2574 - val_loss: 1.5946 - val_accuracy: 0.2448\n",
            "Epoch 19/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5876 - accuracy: 0.2590 - val_loss: 1.6018 - val_accuracy: 0.2363\n",
            "Epoch 20/200\n",
            "408/408 [==============================] - 35s 85ms/step - loss: 1.5839 - accuracy: 0.2681 - val_loss: 1.5932 - val_accuracy: 0.2592\n",
            "Epoch 21/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5780 - accuracy: 0.2742 - val_loss: 1.5938 - val_accuracy: 0.2581\n",
            "Epoch 22/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5780 - accuracy: 0.2738 - val_loss: 1.5987 - val_accuracy: 0.2443\n",
            "Epoch 23/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5881 - accuracy: 0.2630 - val_loss: 1.5980 - val_accuracy: 0.2449\n",
            "Epoch 24/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5897 - accuracy: 0.2603 - val_loss: 1.5938 - val_accuracy: 0.2472\n",
            "Epoch 25/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5845 - accuracy: 0.2646 - val_loss: 1.5949 - val_accuracy: 0.2441\n",
            "Epoch 26/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5863 - accuracy: 0.2623 - val_loss: 1.5948 - val_accuracy: 0.2542\n",
            "Epoch 27/200\n",
            "408/408 [==============================] - 34s 83ms/step - loss: 1.5876 - accuracy: 0.2602 - val_loss: 1.5976 - val_accuracy: 0.2479\n",
            "Epoch 28/200\n",
            "408/408 [==============================] - 35s 85ms/step - loss: 1.5883 - accuracy: 0.2564 - val_loss: 1.5945 - val_accuracy: 0.2666\n",
            "Epoch 29/200\n",
            "408/408 [==============================] - 35s 85ms/step - loss: 1.5827 - accuracy: 0.2674 - val_loss: 1.5891 - val_accuracy: 0.2572\n",
            "Epoch 30/200\n",
            "408/408 [==============================] - 35s 85ms/step - loss: 1.5874 - accuracy: 0.2624 - val_loss: 1.5885 - val_accuracy: 0.2589\n",
            "Epoch 31/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5927 - accuracy: 0.2560 - val_loss: 1.5926 - val_accuracy: 0.2474\n",
            "Epoch 32/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5923 - accuracy: 0.2541 - val_loss: 1.5922 - val_accuracy: 0.2515\n",
            "Epoch 33/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5909 - accuracy: 0.2599 - val_loss: 1.5935 - val_accuracy: 0.2500\n",
            "Epoch 34/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5953 - accuracy: 0.2520 - val_loss: 1.5931 - val_accuracy: 0.2558\n",
            "Epoch 35/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5931 - accuracy: 0.2534 - val_loss: 1.5879 - val_accuracy: 0.2584\n",
            "Epoch 36/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5894 - accuracy: 0.2558 - val_loss: 1.5918 - val_accuracy: 0.2577\n",
            "Epoch 37/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5906 - accuracy: 0.2603 - val_loss: 1.5931 - val_accuracy: 0.2562\n",
            "Epoch 38/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5919 - accuracy: 0.2568 - val_loss: 1.5987 - val_accuracy: 0.2470\n",
            "Epoch 39/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5889 - accuracy: 0.2594 - val_loss: 1.5903 - val_accuracy: 0.2689\n",
            "Epoch 40/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5893 - accuracy: 0.2579 - val_loss: 1.5888 - val_accuracy: 0.2572\n",
            "Epoch 41/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5822 - accuracy: 0.2693 - val_loss: 1.5889 - val_accuracy: 0.2572\n",
            "Epoch 42/200\n",
            "408/408 [==============================] - 34s 83ms/step - loss: 1.5795 - accuracy: 0.2723 - val_loss: 1.5874 - val_accuracy: 0.2554\n",
            "Epoch 43/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5747 - accuracy: 0.2769 - val_loss: 1.5918 - val_accuracy: 0.2526\n",
            "Epoch 44/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5739 - accuracy: 0.2779 - val_loss: 1.5862 - val_accuracy: 0.2651\n",
            "Epoch 45/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5732 - accuracy: 0.2793 - val_loss: 1.5944 - val_accuracy: 0.2536\n",
            "Epoch 46/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5758 - accuracy: 0.2786 - val_loss: 1.5879 - val_accuracy: 0.2601\n",
            "Epoch 47/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5701 - accuracy: 0.2814 - val_loss: 1.5854 - val_accuracy: 0.2640\n",
            "Epoch 48/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5728 - accuracy: 0.2810 - val_loss: 1.5833 - val_accuracy: 0.2659\n",
            "Epoch 49/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5802 - accuracy: 0.2724 - val_loss: 1.5953 - val_accuracy: 0.2528\n",
            "Epoch 50/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5904 - accuracy: 0.2576 - val_loss: 1.5995 - val_accuracy: 0.2441\n",
            "Epoch 51/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5894 - accuracy: 0.2589 - val_loss: 1.5881 - val_accuracy: 0.2614\n",
            "Epoch 52/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5852 - accuracy: 0.2656 - val_loss: 1.5837 - val_accuracy: 0.2643\n",
            "Epoch 53/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5796 - accuracy: 0.2686 - val_loss: 1.5870 - val_accuracy: 0.2601\n",
            "Epoch 54/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5746 - accuracy: 0.2757 - val_loss: 1.5927 - val_accuracy: 0.2546\n",
            "Epoch 55/200\n",
            "408/408 [==============================] - 34s 84ms/step - loss: 1.5766 - accuracy: 0.2718 - val_loss: 1.5909 - val_accuracy: 0.2623\n",
            "Epoch 56/200\n",
            "252/408 [=================>............] - ETA: 12s - loss: 1.5783 - accuracy: 0.2755"
          ]
        }
      ],
      "source": [
        "conv_model,history = disease_classification2((1024,1), x_train1, y_train1, (x_test1,y_test1))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "kWaptx2gHpcW"
      },
      "outputs": [],
      "source": [
        "# GRADED FUNCTION: convolutional_model\n",
        "from tensorflow import keras\n",
        "import tensorflow.keras.layers as tfl\n",
        "from tensorflow.keras.layers import *\n",
        "from tensorflow.keras.utils import *\n",
        "from time import time\n",
        "from keras.callbacks import ModelCheckpoint\n",
        "\n",
        "# Architecture of the model\n",
        "def inception_module(layer, f1, f2, f3, f4):\n",
        "    conv1 = Conv1D(f1, 1, activation='relu', padding='same')(layer)\n",
        "    conv2 = Conv1D(f1, 1, activation='relu', padding='same')(layer)\n",
        "    conv3 = Conv1D(f1, 1, activation='relu', padding='same')(layer)\n",
        "    maxp = MaxPool1D(pool_size=3, strides=1, padding='same')(layer)\n",
        "\n",
        "    conv2 = Conv1D(f2, 3, activation='relu', padding='same')(conv2)\n",
        "    conv3 = Conv1D(f3, 5, activation='relu', padding='same')(conv3)\n",
        "    conv4 = Conv1D(f4, 1, activation='relu', padding='same')(maxp)\n",
        "\n",
        "    return Concatenate()([conv1, conv2, conv3, conv4])\n",
        "\n",
        "def build_incep(input_shape,  x_train,y_train,test_data):\n",
        "    start=time()\n",
        "    input_signal = Input(shape=input_shape)\n",
        "\n",
        "\n",
        "    incept = inception_module(input_signal, 16, 32, 64, 16)\n",
        "    conv = Conv1D(96, 5, activation='relu', padding='same')(incept)\n",
        "    conv = Conv1D(64, 3, activation='relu', padding='same')(conv)\n",
        "    gru = GRU(64, return_sequences=True)(conv)\n",
        "\n",
        "    maxp = MaxPool1D()(gru)\n",
        "\n",
        "    incept = inception_module(maxp, 8, 16, 32, 8)\n",
        "    conv = Conv1D(48, 5, activation='relu', padding='same')(incept)\n",
        "    conv = Conv1D(32, 3, activation='relu', padding='same')(conv)\n",
        "    gru = GRU(32, return_sequences=True)(conv)\n",
        "\n",
        "    maxp = MaxPool1D()(gru)\n",
        "\n",
        "\n",
        "    incept = inception_module(maxp, 2, 4, 8, 2)\n",
        "    conv = Conv1D(12, 5, activation='relu', padding='same')(incept)\n",
        "    conv = Conv1D(8, 3, activation='relu', padding='same')(conv)\n",
        "    gru = GRU(8, return_sequences=True)(conv)\n",
        "\n",
        "    flat = Flatten()(gru)\n",
        "    output = Dense(5, activation='softmax')(flat)\n",
        "\n",
        "\n",
        "    #if 'model' in globals() or 'model' in locals():\n",
        "    #    del model\n",
        "\n",
        "    model = keras.Model(inputs=input_signal, outputs=output)\n",
        "    model.summary()\n",
        "    init_weights = model.get_weights()\n",
        "    \n",
        "    #plot_model(model, show_shapes=True)\n",
        "\n",
        "    #compile\n",
        "    opt = tf.keras.optimizers.Adam( learning_rate=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-07)\n",
        "    model.load_weights(\"/content/drive/MyDrive/ML/Disease_classification/weights/day2/weights-improvement-24-0.80.hdf5\")\n",
        "    #compile\n",
        "    model.compile(optimizer = 'adam',\n",
        "                  loss='categorical_crossentropy',\n",
        "              metrics=['accuracy'])\n",
        "    \n",
        "    model.summary()\n",
        "\n",
        "    #filepath=\"/content/drive/MyDrive/ML/Disease_classification/weights/day2/weights-improvement-{epoch:02d}-{val_accuracy:.2f}.hdf5\"\n",
        "    #checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')\n",
        "    #callbacks_list = [checkpoint]\n",
        "\n",
        "    early_stopping = keras.callbacks.EarlyStopping(monitor='val_loss', patience=15, verbose=2)\n",
        "    #history_m = model.fit(x_train,y_train, epochs=100, validation_split = 0.2, callbacks=[early_stopping,checkpoint] ,batch_size = 100, shuffle = True)\n",
        "\n",
        "    t=time()-start\n",
        "    print('Total training time:', t)\n",
        "    #history_m.history['time']=t\n",
        "    \n",
        "    return model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4OWKoOB-I8_a",
        "outputId": "178dfdad-9d81-4960-919d-554540b91803"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model_1\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_2 (InputLayer)           [(None, 1024, 1)]    0           []                               \n",
            "                                                                                                  \n",
            " conv1d_25 (Conv1D)             (None, 1024, 16)     32          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_26 (Conv1D)             (None, 1024, 16)     32          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " max_pooling1d_5 (MaxPooling1D)  (None, 1024, 1)     0           ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_24 (Conv1D)             (None, 1024, 16)     32          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_27 (Conv1D)             (None, 1024, 32)     1568        ['conv1d_25[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_28 (Conv1D)             (None, 1024, 64)     5184        ['conv1d_26[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_29 (Conv1D)             (None, 1024, 16)     32          ['max_pooling1d_5[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_3 (Concatenate)    (None, 1024, 128)    0           ['conv1d_24[0][0]',              \n",
            "                                                                  'conv1d_27[0][0]',              \n",
            "                                                                  'conv1d_28[0][0]',              \n",
            "                                                                  'conv1d_29[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_30 (Conv1D)             (None, 1024, 96)     61536       ['concatenate_3[0][0]']          \n",
            "                                                                                                  \n",
            " conv1d_31 (Conv1D)             (None, 1024, 64)     18496       ['conv1d_30[0][0]']              \n",
            "                                                                                                  \n",
            " gru_3 (GRU)                    (None, 1024, 64)     24960       ['conv1d_31[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d_6 (MaxPooling1D)  (None, 512, 64)     0           ['gru_3[0][0]']                  \n",
            "                                                                                                  \n",
            " conv1d_33 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_34 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " max_pooling1d_7 (MaxPooling1D)  (None, 512, 64)     0           ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_32 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_35 (Conv1D)             (None, 512, 16)      400         ['conv1d_33[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_36 (Conv1D)             (None, 512, 32)      1312        ['conv1d_34[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_37 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_7[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_4 (Concatenate)    (None, 512, 64)      0           ['conv1d_32[0][0]',              \n",
            "                                                                  'conv1d_35[0][0]',              \n",
            "                                                                  'conv1d_36[0][0]',              \n",
            "                                                                  'conv1d_37[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_38 (Conv1D)             (None, 512, 48)      15408       ['concatenate_4[0][0]']          \n",
            "                                                                                                  \n",
            " conv1d_39 (Conv1D)             (None, 512, 32)      4640        ['conv1d_38[0][0]']              \n",
            "                                                                                                  \n",
            " gru_4 (GRU)                    (None, 512, 32)      6336        ['conv1d_39[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d_8 (MaxPooling1D)  (None, 256, 32)     0           ['gru_4[0][0]']                  \n",
            "                                                                                                  \n",
            " conv1d_41 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_42 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " max_pooling1d_9 (MaxPooling1D)  (None, 256, 32)     0           ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_40 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_43 (Conv1D)             (None, 256, 4)       28          ['conv1d_41[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_44 (Conv1D)             (None, 256, 8)       88          ['conv1d_42[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_45 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_9[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_5 (Concatenate)    (None, 256, 16)      0           ['conv1d_40[0][0]',              \n",
            "                                                                  'conv1d_43[0][0]',              \n",
            "                                                                  'conv1d_44[0][0]',              \n",
            "                                                                  'conv1d_45[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_46 (Conv1D)             (None, 256, 12)      972         ['concatenate_5[0][0]']          \n",
            "                                                                                                  \n",
            " conv1d_47 (Conv1D)             (None, 256, 8)       296         ['conv1d_46[0][0]']              \n",
            "                                                                                                  \n",
            " gru_5 (GRU)                    (None, 256, 8)       432         ['conv1d_47[0][0]']              \n",
            "                                                                                                  \n",
            " flatten_1 (Flatten)            (None, 2048)         0           ['gru_5[0][0]']                  \n",
            "                                                                                                  \n",
            " dense_1 (Dense)                (None, 5)            10245       ['flatten_1[0][0]']              \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 154,373\n",
            "Trainable params: 154,373\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n",
            "Model: \"model_1\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_2 (InputLayer)           [(None, 1024, 1)]    0           []                               \n",
            "                                                                                                  \n",
            " conv1d_25 (Conv1D)             (None, 1024, 16)     32          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_26 (Conv1D)             (None, 1024, 16)     32          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " max_pooling1d_5 (MaxPooling1D)  (None, 1024, 1)     0           ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_24 (Conv1D)             (None, 1024, 16)     32          ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_27 (Conv1D)             (None, 1024, 32)     1568        ['conv1d_25[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_28 (Conv1D)             (None, 1024, 64)     5184        ['conv1d_26[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_29 (Conv1D)             (None, 1024, 16)     32          ['max_pooling1d_5[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_3 (Concatenate)    (None, 1024, 128)    0           ['conv1d_24[0][0]',              \n",
            "                                                                  'conv1d_27[0][0]',              \n",
            "                                                                  'conv1d_28[0][0]',              \n",
            "                                                                  'conv1d_29[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_30 (Conv1D)             (None, 1024, 96)     61536       ['concatenate_3[0][0]']          \n",
            "                                                                                                  \n",
            " conv1d_31 (Conv1D)             (None, 1024, 64)     18496       ['conv1d_30[0][0]']              \n",
            "                                                                                                  \n",
            " gru_3 (GRU)                    (None, 1024, 64)     24960       ['conv1d_31[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d_6 (MaxPooling1D)  (None, 512, 64)     0           ['gru_3[0][0]']                  \n",
            "                                                                                                  \n",
            " conv1d_33 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_34 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " max_pooling1d_7 (MaxPooling1D)  (None, 512, 64)     0           ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_32 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_6[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_35 (Conv1D)             (None, 512, 16)      400         ['conv1d_33[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_36 (Conv1D)             (None, 512, 32)      1312        ['conv1d_34[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_37 (Conv1D)             (None, 512, 8)       520         ['max_pooling1d_7[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_4 (Concatenate)    (None, 512, 64)      0           ['conv1d_32[0][0]',              \n",
            "                                                                  'conv1d_35[0][0]',              \n",
            "                                                                  'conv1d_36[0][0]',              \n",
            "                                                                  'conv1d_37[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_38 (Conv1D)             (None, 512, 48)      15408       ['concatenate_4[0][0]']          \n",
            "                                                                                                  \n",
            " conv1d_39 (Conv1D)             (None, 512, 32)      4640        ['conv1d_38[0][0]']              \n",
            "                                                                                                  \n",
            " gru_4 (GRU)                    (None, 512, 32)      6336        ['conv1d_39[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d_8 (MaxPooling1D)  (None, 256, 32)     0           ['gru_4[0][0]']                  \n",
            "                                                                                                  \n",
            " conv1d_41 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_42 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " max_pooling1d_9 (MaxPooling1D)  (None, 256, 32)     0           ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_40 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_8[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_43 (Conv1D)             (None, 256, 4)       28          ['conv1d_41[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_44 (Conv1D)             (None, 256, 8)       88          ['conv1d_42[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_45 (Conv1D)             (None, 256, 2)       66          ['max_pooling1d_9[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_5 (Concatenate)    (None, 256, 16)      0           ['conv1d_40[0][0]',              \n",
            "                                                                  'conv1d_43[0][0]',              \n",
            "                                                                  'conv1d_44[0][0]',              \n",
            "                                                                  'conv1d_45[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_46 (Conv1D)             (None, 256, 12)      972         ['concatenate_5[0][0]']          \n",
            "                                                                                                  \n",
            " conv1d_47 (Conv1D)             (None, 256, 8)       296         ['conv1d_46[0][0]']              \n",
            "                                                                                                  \n",
            " gru_5 (GRU)                    (None, 256, 8)       432         ['conv1d_47[0][0]']              \n",
            "                                                                                                  \n",
            " flatten_1 (Flatten)            (None, 2048)         0           ['gru_5[0][0]']                  \n",
            "                                                                                                  \n",
            " dense_1 (Dense)                (None, 5)            10245       ['flatten_1[0][0]']              \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 154,373\n",
            "Trainable params: 154,373\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n",
            "Total training time: 2.058302402496338\n"
          ]
        }
      ],
      "source": [
        "incep_model = build_incep((1024,1), x_train1, y_train1, (x_test1,y_test1))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 421
        },
        "id": "t_cmn8CWIwDR",
        "outputId": "404087fc-e601-4828-c380-674a8c32ac29"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "{'matrix': array([[1620,   76,   85,   51,  116],\n",
            "       [  51, 1591,   92,   39,   88],\n",
            "       [  62,   53, 1538,  181,   79],\n",
            "       [  17,    8,   48, 1786,   44],\n",
            "       [ 109,   53,  123,  173, 1475]]), 'kappa': 0.7975529685638654, 'accuracy': 0.8380414312617702, 'precision': array([0.87143626, 0.89331836, 0.8154825 , 0.80089686, 0.81853496]), 'recall': array([0.83162218, 0.85491671, 0.80397282, 0.93851813, 0.7630626 ])}\n",
            "299/299 [==============================] - 11s 28ms/step - loss: 0.4524 - accuracy: 0.8380\n",
            "Accuracy 83.80414247512817\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "from sklearn import metrics\n",
        " \n",
        "def analysis(actual, predict, classes):   \n",
        "    real = np.zeros(actual.shape[0], dtype=int) \n",
        "    pred = np.zeros(predict.shape[0], dtype=int)            \n",
        "    \n",
        "    for i in range(predict.shape[0]):\n",
        "        real[i] = np.argmax(actual[i])  \n",
        "        pred[i] = np.argmax(predict[i]) \n",
        " \n",
        "    param = {'matrix': metrics.confusion_matrix(real, pred),           \n",
        "            'kappa': metrics.cohen_kappa_score(real, pred),      \n",
        "            'accuracy': metrics.accuracy_score(real, pred),         \n",
        "            }\n",
        "    precision = []\n",
        "    recall = []\n",
        "    for i in range(classes):\n",
        "        precision.append(param['matrix'][i,i] / sum(param['matrix'][:, i]))               \n",
        "        recall.append(param['matrix'][i,i] / sum(param['matrix'][i, :]))         \n",
        "    param['precision'] = np.array(precision)   \n",
        "    param['recall'] = np.array(recall)  \n",
        "\n",
        "    return param, real, pred  \n",
        "\n",
        "pred = incep_model.predict(x_test1)  \n",
        "#(loss, acc) = resnet_model.evaluate(x_test1, y_test1)   \n",
        "\n",
        "results, real, prediction = analysis(y_test1, pred, 5)        \n",
        "print(results)\n",
        "\n",
        "(loss, acc) = incep_model.evaluate(x_test1, y_test1)\n",
        "print(\"Accuracy\", acc*100)\n",
        "\n",
        "labels = [\"ins\", \"nfle\", \"narco\", \"rbd\", \"plm\"]\n",
        "\n",
        "disp = metrics.ConfusionMatrixDisplay(confusion_matrix=results['matrix'], display_labels=labels)\n",
        "disp.plot()\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Szpx4I7B_IaB"
      },
      "source": [
        "\n",
        "\n",
        "## Skip Connection"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "3Vpf62wswa-8"
      },
      "outputs": [],
      "source": [
        "# GRADED FUNCTION: convolutional_model\n",
        "import tensorflow as tf\n",
        "from keras.callbacks import EarlyStopping\n",
        "import tensorflow.keras.layers as tfl\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.layers import Dropout\n",
        "from keras.utils.vis_utils import plot_model\n",
        "from keras.layers.merge import concatenate\n",
        "from tensorflow import keras\n",
        "from time import time\n",
        "\n",
        "def skip_model(input_shape, x_train,y_train,test_data):\n",
        "    i = 0\n",
        "    skips = []\n",
        "    start = time()\n",
        "    input_img = tf.keras.Input(shape=input_shape)\n",
        "    ## CONV1D: 8 filters 4x4, stride of 1, padding 'SAME'\n",
        "    Z11 = tfl.Conv1D(filters=32,\n",
        "                    kernel_size=7,\n",
        "                    strides=1,\n",
        "                    padding='same')(input_img)\n",
        "    Z12  = tfl.BatchNormalization()(Z11)\n",
        "    Z13 = tfl.ReLU()(Z12)\n",
        "    skips.append(Z13)\n",
        "\n",
        "\n",
        "    Z21 = tfl.Conv1D(filters=32,\n",
        "                    kernel_size=7,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z13)\n",
        "    Z22  = tfl.BatchNormalization()(Z21)\n",
        "    Z23 = tfl.ReLU()(Z22)\n",
        "    Z24 = tfl.Conv1D(filters = 32, kernel_size = 9, strides = 1 , padding = 'same')(Z23)\n",
        "    skip = skips[i]\n",
        "    i = i+1\n",
        "    x = tfl.Concatenate()([Z24, skip])\n",
        "    skips.append(x)\n",
        "\n",
        "    \n",
        "    Z31  = tfl.BatchNormalization()(x)\n",
        "    Z32  = tfl.ReLU()(Z31)\n",
        "\n",
        "    #Z1=tf.nn.dropout(Z1, rate = 0.3)\n",
        "    Z33 = tfl.Conv1D(filters=16,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z32)\n",
        "    Z34 = tfl.MaxPooling1D(pool_size = 2 , padding = 'same')(Z33)\n",
        "    Z35 = tfl.BatchNormalization()(Z34)\n",
        "    Z36 = tfl.ReLU()(Z35)\n",
        "\n",
        "    A1 = tfl.Conv1D(filters=16,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z36)\n",
        "\n",
        "    Z41 = tfl.Conv1D(filters=32,\n",
        "                    kernel_size=7,\n",
        "                    strides=1,\n",
        "                    padding='same')(A1)\n",
        "    Z42  = tfl.BatchNormalization()(Z41)\n",
        "    Z43 = tfl.ReLU()(Z42)\n",
        "\n",
        "    skips.append(Z43)\n",
        "\n",
        "\n",
        "    Z51 = tfl.Conv1D(filters=32,\n",
        "                    kernel_size=7,\n",
        "                    strides=1,\n",
        "                    padding='same')(A1)\n",
        "    Z52  = tfl.BatchNormalization()(Z51)\n",
        "    Z53 = tfl.ReLU()(Z52)\n",
        "    Z54 = tfl.Conv1D(filters = 32, kernel_size = 9, strides = 1 , padding = 'same')(Z53)\n",
        "    skip = skips[2]\n",
        "    i = i+1\n",
        "    x = tfl.Concatenate()([Z54, skip])\n",
        "    skips.append(x)\n",
        "    \n",
        "    # x = tfl.Concatenate()([Z37, Z42])\n",
        "    ## RELU\n",
        "    ## MAXPOOL: window 8x8, stride 8, padding 'SAME'\n",
        "    P1 = tfl.AveragePooling1D(pool_size= 16,\n",
        "                       padding='same')(x)\n",
        "    #P1=tf.nn.dropout(P1, rate = 0.3)\n",
        "\n",
        "\n",
        "    Z1 = tfl.Conv1D(filters=32,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same')(P1)\n",
        "    #Z1=tf.nn.dropout(Z1, rate = 0.3)\n",
        "    Z11 = tfl.Conv1D(filters=12,\n",
        "                    kernel_size=7,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z1)\n",
        "    ## RELU\n",
        "    A1 = tfl.ReLU()(Z11)\n",
        "    ## MAXPOOL: window 8x8, stride 8, padding 'SAME'\n",
        "    P1 = tfl.MaxPooling1D(pool_size= 8,\n",
        "                       padding='same')(A1)\n",
        "   \n",
        "    ## CONV2D: 16 filters 3x3, stride 1, padding 'SAME'\n",
        "    Z2 = tfl.Conv1D(filters=16,\n",
        "                    kernel_size=5,\n",
        "                    strides=1,\n",
        "                    padding='same')(P1)\n",
        "    #Z2=tf.nn.dropout(Z2, rate = 0.4)\n",
        "    Z21 = tfl.Conv1D(filters=8,\n",
        "                    kernel_size=3,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z2)\n",
        "    ## RELU\n",
        "    A2 = tfl.ReLU()(Z21)\n",
        "    ## MAXPOOL: window 4x4, stride 4, padding 'SAME'\n",
        "    P2 = tfl.MaxPooling1D(pool_size=4,\n",
        "                       padding='same')(A2)\n",
        "\n",
        "    Z3 = tfl.Conv1D(filters=8,\n",
        "                    kernel_size=5,\n",
        "                    strides=1,\n",
        "                    padding='same')(P2)\n",
        "    Z3=tf.nn.dropout(Z3, rate = 0.2)\n",
        "    Z31 = tfl.Conv1D(filters=4,\n",
        "                    kernel_size=3,\n",
        "                    strides=1,\n",
        "                    padding='same')(Z3)\n",
        "    ## RELU\n",
        "    A3 = tfl.ReLU()(Z31)\n",
        "    ## MAXPOOL: window 8x8, stride 8, padding 'SAME'\n",
        "    P3 = tfl.MaxPooling1D(pool_size= 4,\n",
        "                       padding='same')(A3)\n",
        "\n",
        "    ## FLATTEN\n",
        "    F = tfl.Flatten()(P3)\n",
        "    ## Dense layer\n",
        "    op = tfl.Dense(units=55,\n",
        "                        activation='relu')(F)\n",
        "\n",
        "    op1 = tfl.Dense(units=16,\n",
        "                        activation='relu')(F)\n",
        "\n",
        "          \n",
        "    outputs = tfl.Dense(units=5,\n",
        "                        activation='softmax')(op1)\n",
        "\n",
        "    model = tf.keras.Model(inputs=input_img, outputs=outputs)\n",
        "\n",
        "    #compile\n",
        "    model.compile(optimizer = 'adam',\n",
        "                  loss='categorical_crossentropy',\n",
        "              metrics=['CategoricalAccuracy'])\n",
        "\n",
        "    model.summary()\n",
        "\n",
        "    early_stopping = EarlyStopping(monitor='val_loss', patience=30, verbose=2)\n",
        "    history_m = model.fit(x_train,y_train, epochs=1, validation_split =0.2, callbacks = [early_stopping], batch_size = 100, shuffle = True)\n",
        "\n",
        "    t=time()-start\n",
        "    print('Total training time:', t)\n",
        "    history_m.history['training_time']=t\n",
        "\n",
        "    # filepath=\"/content/drive/MyDrive/ML/Disease_classification/weights/weights-improvement-{epoch:02d}-{val_accuracy:.2f}.hdf5\"\n",
        "    # checkpoint = ModelCheckpoint(filepath, monitor='val_accuracy', verbose=1, save_best_only=True, mode='max')\n",
        "    \n",
        "    return model, history_m"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qacUf8OHifYx",
        "outputId": "d0003f18-4de2-4894-aaee-2ed7f9a727ac"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_1 (InputLayer)           [(None, 1024, 1)]    0           []                               \n",
            "                                                                                                  \n",
            " conv1d (Conv1D)                (None, 1024, 32)     256         ['input_1[0][0]']                \n",
            "                                                                                                  \n",
            " batch_normalization (BatchNorm  (None, 1024, 32)    128         ['conv1d[0][0]']                 \n",
            " alization)                                                                                       \n",
            "                                                                                                  \n",
            " re_lu (ReLU)                   (None, 1024, 32)     0           ['batch_normalization[0][0]']    \n",
            "                                                                                                  \n",
            " conv1d_1 (Conv1D)              (None, 1024, 32)     7200        ['re_lu[0][0]']                  \n",
            "                                                                                                  \n",
            " batch_normalization_1 (BatchNo  (None, 1024, 32)    128         ['conv1d_1[0][0]']               \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " re_lu_1 (ReLU)                 (None, 1024, 32)     0           ['batch_normalization_1[0][0]']  \n",
            "                                                                                                  \n",
            " conv1d_2 (Conv1D)              (None, 1024, 32)     9248        ['re_lu_1[0][0]']                \n",
            "                                                                                                  \n",
            " concatenate (Concatenate)      (None, 1024, 64)     0           ['conv1d_2[0][0]',               \n",
            "                                                                  're_lu[0][0]']                  \n",
            "                                                                                                  \n",
            " batch_normalization_2 (BatchNo  (None, 1024, 64)    256         ['concatenate[0][0]']            \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " re_lu_2 (ReLU)                 (None, 1024, 64)     0           ['batch_normalization_2[0][0]']  \n",
            "                                                                                                  \n",
            " conv1d_3 (Conv1D)              (None, 1024, 16)     9232        ['re_lu_2[0][0]']                \n",
            "                                                                                                  \n",
            " max_pooling1d (MaxPooling1D)   (None, 512, 16)      0           ['conv1d_3[0][0]']               \n",
            "                                                                                                  \n",
            " batch_normalization_3 (BatchNo  (None, 512, 16)     64          ['max_pooling1d[0][0]']          \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " re_lu_3 (ReLU)                 (None, 512, 16)      0           ['batch_normalization_3[0][0]']  \n",
            "                                                                                                  \n",
            " conv1d_4 (Conv1D)              (None, 512, 16)      2320        ['re_lu_3[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_6 (Conv1D)              (None, 512, 32)      3616        ['conv1d_4[0][0]']               \n",
            "                                                                                                  \n",
            " batch_normalization_5 (BatchNo  (None, 512, 32)     128         ['conv1d_6[0][0]']               \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " conv1d_5 (Conv1D)              (None, 512, 32)      3616        ['conv1d_4[0][0]']               \n",
            "                                                                                                  \n",
            " re_lu_5 (ReLU)                 (None, 512, 32)      0           ['batch_normalization_5[0][0]']  \n",
            "                                                                                                  \n",
            " batch_normalization_4 (BatchNo  (None, 512, 32)     128         ['conv1d_5[0][0]']               \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " conv1d_7 (Conv1D)              (None, 512, 32)      9248        ['re_lu_5[0][0]']                \n",
            "                                                                                                  \n",
            " re_lu_4 (ReLU)                 (None, 512, 32)      0           ['batch_normalization_4[0][0]']  \n",
            "                                                                                                  \n",
            " concatenate_1 (Concatenate)    (None, 512, 64)      0           ['conv1d_7[0][0]',               \n",
            "                                                                  're_lu_4[0][0]']                \n",
            "                                                                                                  \n",
            " average_pooling1d (AveragePool  (None, 32, 64)      0           ['concatenate_1[0][0]']          \n",
            " ing1D)                                                                                           \n",
            "                                                                                                  \n",
            " conv1d_8 (Conv1D)              (None, 32, 32)       18464       ['average_pooling1d[0][0]']      \n",
            "                                                                                                  \n",
            " conv1d_9 (Conv1D)              (None, 32, 12)       2700        ['conv1d_8[0][0]']               \n",
            "                                                                                                  \n",
            " re_lu_6 (ReLU)                 (None, 32, 12)       0           ['conv1d_9[0][0]']               \n",
            "                                                                                                  \n",
            " max_pooling1d_1 (MaxPooling1D)  (None, 4, 12)       0           ['re_lu_6[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_10 (Conv1D)             (None, 4, 16)        976         ['max_pooling1d_1[0][0]']        \n",
            "                                                                                                  \n",
            " conv1d_11 (Conv1D)             (None, 4, 8)         392         ['conv1d_10[0][0]']              \n",
            "                                                                                                  \n",
            " re_lu_7 (ReLU)                 (None, 4, 8)         0           ['conv1d_11[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d_2 (MaxPooling1D)  (None, 1, 8)        0           ['re_lu_7[0][0]']                \n",
            "                                                                                                  \n",
            " conv1d_12 (Conv1D)             (None, 1, 8)         328         ['max_pooling1d_2[0][0]']        \n",
            "                                                                                                  \n",
            " tf.nn.dropout (TFOpLambda)     (None, 1, 8)         0           ['conv1d_12[0][0]']              \n",
            "                                                                                                  \n",
            " conv1d_13 (Conv1D)             (None, 1, 4)         100         ['tf.nn.dropout[0][0]']          \n",
            "                                                                                                  \n",
            " re_lu_8 (ReLU)                 (None, 1, 4)         0           ['conv1d_13[0][0]']              \n",
            "                                                                                                  \n",
            " max_pooling1d_3 (MaxPooling1D)  (None, 1, 4)        0           ['re_lu_8[0][0]']                \n",
            "                                                                                                  \n",
            " flatten (Flatten)              (None, 4)            0           ['max_pooling1d_3[0][0]']        \n",
            "                                                                                                  \n",
            " dense_1 (Dense)                (None, 16)           80          ['flatten[0][0]']                \n",
            "                                                                                                  \n",
            " dense_2 (Dense)                (None, 5)            85          ['dense_1[0][0]']                \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 68,693\n",
            "Trainable params: 68,277\n",
            "Non-trainable params: 416\n",
            "__________________________________________________________________________________________________\n",
            "306/306 [==============================] - 44s 84ms/step - loss: 1.4958 - categorical_accuracy: 0.3096 - val_loss: 1.4379 - val_categorical_accuracy: 0.3485\n",
            "Total training time: 50.533894300460815\n"
          ]
        }
      ],
      "source": [
        " model, history_m = skip_model((1024,1), x_train1, y_train1, (x_test1, y_test1))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "sJtNGpbfizsq"
      },
      "outputs": [],
      "source": [
        "from tensorflow import keras\n",
        "model1 = keras.models.load_model(\"model_skip_D70.h5\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 440
        },
        "id": "psjmZUBfio8X",
        "outputId": "67938619-3fb3-45ef-9242-abea8dc1250c"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "[[ 976   83   46   91  712]\n",
            " [  48 1448   88   34  261]\n",
            " [  27  105 1316  174  317]\n",
            " [   1   14   30 1841   89]\n",
            " [  22   24   40  190 1581]]\n",
            "Acc 0.7493199414103369\n",
            "Precision [0.90875233 0.86499403 0.86578947 0.79012876 0.53412162]\n",
            "Recall [0.5115304  0.77062267 0.67870036 0.9321519  0.85137318]\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7fc8d0388310>"
            ]
          },
          "execution_count": 10,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "from sklearn import metrics\n",
        " \n",
        "def analysis(actual, predict, classes):   \n",
        "    real = np.zeros(actual.shape[0], dtype=int) \n",
        "    pred = np.zeros(predict.shape[0], dtype=int)            \n",
        "    \n",
        "    for i in range(predict.shape[0]):\n",
        "        real[i] = np.argmax(actual[i])  \n",
        "        pred[i] = np.argmax(predict[i]) \n",
        " \n",
        "    param = {'matrix': metrics.confusion_matrix(real, pred),           \n",
        "            'kappa': metrics.cohen_kappa_score(real, pred),      \n",
        "            'accuracy': metrics.accuracy_score(real, pred),         \n",
        "            }\n",
        "    precision = []\n",
        "    recall = []\n",
        "    for i in range(classes):\n",
        "        precision.append(param['matrix'][i,i] / sum(param['matrix'][:, i]))               \n",
        "        recall.append(param['matrix'][i,i] / sum(param['matrix'][i, :]))         \n",
        "    param['precision'] = np.array(precision)   \n",
        "    param['recall'] = np.array(recall)  \n",
        "\n",
        "    return param, real, pred  \n",
        "\n",
        "pred = model1.predict(x_test1)  \n",
        "#(loss, acc) = resnet_model.evaluate(x_test1, y_test1)   \n",
        "\n",
        "results, real, prediction = analysis(y_test1, pred, 5)        \n",
        "print(results['matrix'])\n",
        "print(\"Acc\",results['accuracy'])\n",
        "print(\"Precision\",results['precision'])\n",
        "print(\"Recall\",results['recall'])\n",
        "\n",
        "\n",
        "labels =  [\"narco\",\"ins\",\"nfle\",\"plm\",\"rbd\"]\n",
        "\n",
        "\n",
        "disp = metrics.ConfusionMatrixDisplay(confusion_matrix=results['matrix'], display_labels=labels)\n",
        "disp.plot()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QEK2hQhG_Oc_"
      },
      "source": [
        "##LSTM Model"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "NlKrxidG7gCx"
      },
      "outputs": [],
      "source": [
        "bal_all = np.loadtxt('/content/drive/MyDrive/FYP/combine_classification/dataset/both_phases.csv', delimiter=',')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Zn4oTM5x8QdW",
        "outputId": "56f632b3-6640-445b-d53d-73eb3d82b291"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "(47790, 1024)\n",
            "(47790,)\n",
            "(47790, 1)\n",
            "0.0\n",
            "[1. 0. 0. 0. 0.]\n"
          ]
        }
      ],
      "source": [
        "x_bal = bal_all[:,0:1024]\n",
        "y_disease = bal_all[:,-1]\n",
        "print(np.shape(x_bal))\n",
        "print(np.shape(y_disease))\n",
        "y_disease=y_disease.reshape(-1,1 )\n",
        "print(np.shape(y_disease))\n",
        "\n",
        "from sklearn.model_selection import train_test_split \n",
        "from sklearn.preprocessing import OneHotEncoder\n",
        "\n",
        "print(y_disease[8500,0])\n",
        "enc = OneHotEncoder(handle_unknown='ignore')\n",
        "# 2. FIT\n",
        "enc.fit(y_disease)\n",
        "\n",
        "# 3. Transform\n",
        "onehotlabels = enc.transform(y_disease).toarray()\n",
        "\n",
        "print(onehotlabels[8500])\n",
        "\n",
        "x_train1, x_test1, y_train1, y_test1 = train_test_split(x_bal, onehotlabels, test_size = 0.2, shuffle = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wyxzlaLuRJYU",
        "outputId": "556a71da-78ca-4dca-a9f1-550708f9f18e"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " conv1d (Conv1D)             (None, 1024, 32)          256       \n",
            "                                                                 \n",
            " conv1d_1 (Conv1D)           (None, 1024, 16)          4624      \n",
            "                                                                 \n",
            " conv1d_2 (Conv1D)           (None, 1024, 32)          2592      \n",
            "                                                                 \n",
            " lstm (LSTM)                 (None, 100)               53200     \n",
            "                                                                 \n",
            " dense (Dense)               (None, 5)                 505       \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 61,177\n",
            "Trainable params: 61,177\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n",
            "306/306 [==============================] - 32s 67ms/step - loss: 1.5609 - categorical_accuracy: 0.2837 - val_loss: 1.5625 - val_categorical_accuracy: 0.2972\n",
            "Total training time: 35.7595739364624\n",
            "Accuracy: 29.76%\n"
          ]
        }
      ],
      "source": [
        "# LSTM and CNN for sequence classification in the IMDB dataset\n",
        "import numpy\n",
        "from keras.datasets import imdb\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Dense\n",
        "from keras.layers import LSTM\n",
        "from keras.layers.convolutional import Conv1D\n",
        "from keras.layers.convolutional import MaxPooling1D\n",
        "from keras.callbacks import EarlyStopping\n",
        "\n",
        "from keras.layers.embeddings import Embedding\n",
        "from keras.preprocessing import sequence\n",
        "from time import time\n",
        "\n",
        "# fix random seed for reproducibility\n",
        "\n",
        "start = time()\n",
        "model = Sequential()\n",
        "model.add(keras.Input(shape=(1024,1)))\n",
        "model.add(Conv1D(filters=32,\n",
        "                    kernel_size=7,\n",
        "                    strides=1,\n",
        "                    padding='same'))\n",
        "model.add(Conv1D(filters=16,\n",
        "                    kernel_size=9,\n",
        "                    strides=1,\n",
        "                    padding='same'))\n",
        "# model.add(MaxPooling1D(pool_size= 16,\n",
        "                      #  padding='same'))\n",
        "model.add(Conv1D(filters=32, kernel_size=5, padding='same', activation='relu'))\n",
        "# model.add(MaxPooling1D(pool_size=8))\n",
        "model.add(LSTM(100))\n",
        "model.add(Dense(5, activation='softmax'))\n",
        "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['CategoricalAccuracy'])\n",
        "model.build((1024,1,1))\n",
        "print(model.summary())\n",
        "# model.fit(x_train1, y_train1, epochs=25, batch_size=64)\n",
        "early_stopping = EarlyStopping(monitor='val_loss', patience=30, verbose=2)\n",
        "history_m = model.fit(x_train1,y_train1, epochs=1, validation_split =0.2, callbacks = [early_stopping], batch_size = 100, shuffle = True)\n",
        "t=time()-start\n",
        "print('Total training time:', t)\n",
        "history_m.history['training_time']=t\n",
        "# Final evaluation of the model\n",
        "scores = model.evaluate(x_test1, y_test1, verbose=0)\n",
        "print(\"Accuracy: %.2f%%\" % (scores[1]*100))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6uhhaOkOVAGX"
      },
      "source": [
        "## Result"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "i7CeuLb9RiRt"
      },
      "outputs": [],
      "source": [
        "from tensorflow import keras\n",
        "model1 = keras.models.load_model(\"lstm_disease80.h5\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 440
        },
        "id": "eKvOitC-RTFM",
        "outputId": "89dbdd19-45cf-4e4f-c0e2-31b777d03bca"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "[[1728   36   29   32   78]\n",
            " [  37 1743   44   33   55]\n",
            " [  48   39 1732   60   74]\n",
            " [  11    2   15 1824   40]\n",
            " [  59   40   56   75 1668]]\n",
            "Acc 0.909709144172421\n",
            "Precision [0.91768455 0.93709677 0.92324094 0.90118577 0.87101828]\n",
            "Recall [0.90803994 0.91161088 0.88684076 0.9640592  0.87881981]\n"
          ]
        },
        {
          "data": {
            "text/plain": [
              "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7faf62650750>"
            ]
          },
          "execution_count": 8,
          "metadata": {},
          "output_type": "execute_result"
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "from sklearn import metrics\n",
        " \n",
        "def analysis(actual, predict, classes):   \n",
        "    real = np.zeros(actual.shape[0], dtype=int) \n",
        "    pred = np.zeros(predict.shape[0], dtype=int)            \n",
        "    \n",
        "    for i in range(predict.shape[0]):\n",
        "        real[i] = np.argmax(actual[i])  \n",
        "        pred[i] = np.argmax(predict[i]) \n",
        " \n",
        "    param = {'matrix': metrics.confusion_matrix(real, pred),           \n",
        "            'kappa': metrics.cohen_kappa_score(real, pred),      \n",
        "            'accuracy': metrics.accuracy_score(real, pred),         \n",
        "            }\n",
        "    precision = []\n",
        "    recall = []\n",
        "    for i in range(classes):\n",
        "        precision.append(param['matrix'][i,i] / sum(param['matrix'][:, i]))               \n",
        "        recall.append(param['matrix'][i,i] / sum(param['matrix'][i, :]))         \n",
        "    param['precision'] = np.array(precision)   \n",
        "    param['recall'] = np.array(recall)  \n",
        "\n",
        "    return param, real, pred  \n",
        "\n",
        "pred = model1.predict(x_test1)  \n",
        "#(loss, acc) = resnet_model.evaluate(x_test1, y_test1)   \n",
        "\n",
        "results, real, prediction = analysis(y_test1, pred, 5)        \n",
        "print(results['matrix'])\n",
        "print(\"Acc\",results['accuracy'])\n",
        "print(\"Precision\",results['precision'])\n",
        "print(\"Recall\",results['recall'])\n",
        "\n",
        "\n",
        "labels =  [\"narco\",\"ins\",\"nfle\",\"plm\",\"rbd\"]\n",
        "\n",
        "\n",
        "disp = metrics.ConfusionMatrixDisplay(confusion_matrix=results['matrix'], display_labels=labels)\n",
        "disp.plot()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tii3rMntAxGv"
      },
      "source": [
        "# Disease Classification [Phase B]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-Vtrfvb5A1oR"
      },
      "outputs": [],
      "source": [
        "from tensorflow import keras\n",
        "model1 = keras.models.load_model(\"my_h5_model_lstm_phaseB.h5\")\n",
        "bal_all = np.loadtxt('/content/drive/MyDrive/FYP/combine_classification/dataset/B_phases.csv', delimiter=',')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fxzUZOfrA9El",
        "outputId": "7feef89e-6b91-4904-e6b1-305e7610186b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "(23895, 1024)\n",
            "(23895,)\n",
            "(23895, 1)\n",
            "1.0\n",
            "[0. 1. 0. 0. 0.]\n"
          ]
        }
      ],
      "source": [
        "x_bal = bal_all[:,0:1024]\n",
        "y_disease = bal_all[:,-1]\n",
        "print(np.shape(x_bal))\n",
        "print(np.shape(y_disease))\n",
        "y_disease=y_disease.reshape(-1,1 )\n",
        "print(np.shape(y_disease))\n",
        "\n",
        "\n",
        "from sklearn.model_selection import train_test_split \n",
        "from sklearn.preprocessing import OneHotEncoder\n",
        "\n",
        "print(y_disease[8500,0])\n",
        "enc = OneHotEncoder(handle_unknown='ignore')\n",
        "# 2. FIT\n",
        "enc.fit(y_disease)\n",
        "\n",
        "# 3. Transform\n",
        "onehotlabels = enc.transform(y_disease).toarray()\n",
        "\n",
        "print(onehotlabels[8500])\n",
        "\n",
        "x_train1, x_test1, y_train1, y_test1 = train_test_split(x_bal, onehotlabels, test_size = 0.2, shuffle = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 496
        },
        "id": "vZfBOFCpBiIa",
        "outputId": "dc8503f3-1f85-43a3-9a91-78c14106310d"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "[[880   0  33   0   9]\n",
            " [ 12 960  11   2   7]\n",
            " [ 20   0 886   0  61]\n",
            " [  6   9  27 886  21]\n",
            " [  7   0  25   3 914]]\n",
            "Acc 0.9470600544046872\n",
            "Precision [0.95135135 0.99071207 0.90224033 0.99438833 0.90316206]\n",
            "Recall [0.95444685 0.96774194 0.91623578 0.93361433 0.96311907]\n",
            "150/150 [==============================] - 3s 15ms/step - loss: 0.1524 - categorical_accuracy: 0.9471\n",
            "Accuracy 94.70600485801697\n",
            "150/150 - 2s - 2s/epoch - 13ms/step\n"
          ]
        },
        {
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "\n",
        "from sklearn import metrics\n",
        " \n",
        "def analysis(actual, predict, classes):   \n",
        "    real = np.zeros(actual.shape[0], dtype=int) \n",
        "    pred = np.zeros(predict.shape[0], dtype=int)            \n",
        "    \n",
        "    for i in range(predict.shape[0]):\n",
        "        real[i] = np.argmax(actual[i])  \n",
        "        pred[i] = np.argmax(predict[i]) \n",
        " \n",
        "    param = {'matrix': metrics.confusion_matrix(real, pred),           \n",
        "            'kappa': metrics.cohen_kappa_score(real, pred),      \n",
        "            'accuracy': metrics.accuracy_score(real, pred),         \n",
        "            }\n",
        "    precision = []\n",
        "    recall = []\n",
        "    for i in range(classes):\n",
        "        precision.append(param['matrix'][i,i] / sum(param['matrix'][:, i]))               \n",
        "        recall.append(param['matrix'][i,i] / sum(param['matrix'][i, :]))         \n",
        "    param['precision'] = np.array(precision)   \n",
        "    param['recall'] = np.array(recall)  \n",
        "\n",
        "    return param, real, pred  \n",
        "\n",
        "pred = model1.predict(x_test1)  \n",
        "#(loss, acc) = resnet_model.evaluate(x_test1, y_test1)   \n",
        "\n",
        "results, real, prediction = analysis(y_test1, pred, 5)        \n",
        "print(results['matrix'])\n",
        "print(\"Acc\",results['accuracy'])\n",
        "print(\"Precision\",results['precision'])\n",
        "print(\"Recall\",results['recall'])\n",
        "(loss, acc) = model1.evaluate(x_test1, y_test1)\n",
        "print(\"Accuracy\", acc*100)\n",
        "\n",
        "labels =  [\"narco\",\"ins\",\"nfle\",\"plm\",\"rbd\"]\n",
        "\n",
        "disp = metrics.ConfusionMatrixDisplay(confusion_matrix=results['matrix'], display_labels=labels)\n",
        "disp.plot()\n",
        "\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "from sklearn import metrics\n",
        "\n",
        "prob = model1.predict(x_test1,verbose = 2)\n",
        "pred = (prob > 0.5).astype(\"int32\")\n",
        "\n",
        "\n",
        "#(loss, acc) = conv_model.evaluate(x_test1, y_test1)\n",
        "#print(\"Accuracy\", acc*100)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ivQ8EKVpA2Ix"
      },
      "source": [
        "# Disease Classification [Phase A]"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from tensorflow import keras\n",
        "model = tf.keras.models.load_model('/content/drive/MyDrive/ML/Disease_classification/Dhruvi/LSTM/A/model_v1.h5')\n",
        "bal_all = np.loadtxt('/content/drive/MyDrive/FYP/combine_classification/dataset/A_phases.csv', delimiter=',')"
      ],
      "metadata": {
        "id": "XvkGMB0pXUm5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "9pXWiLHsIAoa"
      },
      "outputs": [],
      "source": [
        "x_bal = bal_all[:,0:1024]\n",
        "y_disease = bal_all[:,-1]\n",
        "print(np.shape(x_bal))\n",
        "print(np.shape(y_disease))\n",
        "y_disease=y_disease.reshape(-1,1 )\n",
        "print(np.shape(y_disease))\n",
        "\n",
        "\n",
        "from sklearn.model_selection import train_test_split \n",
        "from sklearn.preprocessing import OneHotEncoder\n",
        "\n",
        "print(y_disease[8500,0])\n",
        "enc = OneHotEncoder(handle_unknown='ignore')\n",
        "# 2. FIT\n",
        "enc.fit(y_disease)\n",
        "\n",
        "# 3. Transform\n",
        "onehotlabels = enc.transform(y_disease).toarray()\n",
        "\n",
        "print(onehotlabels[8500])\n",
        "\n",
        "x_train1, x_test1, y_train1, y_test1 = train_test_split(x_bal, onehotlabels, test_size = 0.2, shuffle = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "UHTZIdNI-65n"
      },
      "outputs": [],
      "source": [
        "\n",
        "from sklearn import metrics\n",
        " \n",
        "def analysis(actual, predict, classes):   \n",
        "    real = np.zeros(actual.shape[0], dtype=int) \n",
        "    pred = np.zeros(predict.shape[0], dtype=int)            \n",
        "    \n",
        "    for i in range(predict.shape[0]):\n",
        "        real[i] = np.argmax(actual[i])  \n",
        "        pred[i] = np.argmax(predict[i]) \n",
        " \n",
        "    param = {'matrix': metrics.confusion_matrix(real, pred),           \n",
        "            'kappa': metrics.cohen_kappa_score(real, pred),      \n",
        "            'accuracy': metrics.accuracy_score(real, pred),         \n",
        "            }\n",
        "    precision = []\n",
        "    recall = []\n",
        "    for i in range(classes):\n",
        "        precision.append(param['matrix'][i,i] / sum(param['matrix'][:, i]))               \n",
        "        recall.append(param['matrix'][i,i] / sum(param['matrix'][i, :]))         \n",
        "    param['precision'] = np.array(precision)   \n",
        "    param['recall'] = np.array(recall)  \n",
        "\n",
        "    return param, real, pred  \n",
        "\n",
        "pred = model1.predict(x_test1)  \n",
        "#(loss, acc) = resnet_model.evaluate(x_test1, y_test1)   \n",
        "\n",
        "results, real, prediction = analysis(y_test1, pred, 5)        \n",
        "print(results['matrix'])\n",
        "print(\"Acc\",results['accuracy'])\n",
        "print(\"Precision\",results['precision'])\n",
        "print(\"Recall\",results['recall'])\n",
        "(loss, acc) = model1.evaluate(x_test1, y_test1)\n",
        "print(\"Accuracy\", acc*100)\n",
        "\n",
        "labels =  [\"narco\",\"ins\",\"nfle\",\"plm\",\"rbd\"]\n",
        "\n",
        "disp = metrics.ConfusionMatrixDisplay(confusion_matrix=results['matrix'], display_labels=labels)\n",
        "disp.plot()\n",
        "\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "from sklearn import metrics\n",
        "\n",
        "prob = model1.predict(x_test1,verbose = 2)\n",
        "pred = (prob > 0.5).astype(\"int32\")"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "collapsed_sections": [],
      "name": "Disease Classification",
      "toc_visible": true,
      "provenance": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}