VAE.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "VAE.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "zl0ZjU1BVw1y",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "80f5d4ba-7ff7-47cc-c8c3-cfbb76d1c0c8"
      },
      "source": [
        "!pip install -U -q PyDrive\n",
        "from pydrive.auth import GoogleAuth\n",
        "from pydrive.drive import GoogleDrive\n",
        "from google.colab import auth\n",
        "from oauth2client.client import GoogleCredentials# Authenticate and create the PyDrive client.\n",
        "import pandas as pd\n",
        "from matplotlib import pyplot\n",
        "from keras.layers import Input, Dense, BatchNormalization, Lambda, GaussianNoise, GaussianDropout\n",
        "from keras.models import Model\n",
        "import keras.backend as backend\n",
        "import numpy as np\n",
        "import keras\n",
        "from sklearn.preprocessing import LabelEncoder, normalize\n",
        "from keras.utils import np_utils\n",
        "import pandas as pd\n",
        "import tensorflow.compat.v1 as tf\n",
        "tf.disable_v2_behavior()\n",
        "import time\n",
        "import math\n",
        "from sklearn.metrics import mean_squared_error,median_absolute_error,mean_absolute_error,explained_variance_score,r2_score\n",
        "from math import sqrt\n",
        "import keras.utils as np_utils\n",
        "from sklearn.preprocessing import LabelEncoder\n",
        "import numpy as np"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.7/dist-packages/tensorflow/python/compat/v2_compat.py:96: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "non-resource variables are not supported in the long term\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "81suZoTimVbL"
      },
      "source": [
        "auth.authenticate_user()\n",
        "gauth = GoogleAuth()\n",
        "gauth.credentials = GoogleCredentials.get_application_default()\n",
        "drive = GoogleDrive(gauth)"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DAKzqYAOWMqT"
      },
      "source": [
        "id = \"1uiIMbheUVJK82LMshquFhDAQW8vAE6Dt\"\n",
        "downloaded = drive.CreateFile({'id':id}) \n",
        "downloaded.GetContentFile('gene.trainset.csv')  \n",
        "gene_1 = pd.read_csv('gene.trainset.csv')"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Uwt9rd5_-vt_"
      },
      "source": [
        "gene_train = gene_1.drop(columns = [\"Cancer_Type\", \"Sample_ID\"])"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AoTu67epXiGt"
      },
      "source": [
        "id = \"1MCvPmLzJnMq-DM9ch0MS7B3btv7RN3QO\"\n",
        "downloaded = drive.CreateFile({'id':id}) \n",
        "downloaded.GetContentFile('gene.val.csv')  \n",
        "gene_2 = pd.read_csv('gene.val.csv')"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "inBO2lUh-204"
      },
      "source": [
        "gene_test = gene_2.drop(columns = [\"Cancer_Type\", \"Sample_ID\"])"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x4xiH2Hn-5vB"
      },
      "source": [
        "id = '1ZhpxOI75zZBBJMRu4supVZzWzGmoFUaR'\n",
        "downloaded = drive.CreateFile({'id':id}) \n",
        "downloaded.GetContentFile('pro.trainset.csv')  \n",
        "pro1 = pd.read_csv('pro.trainset.csv')"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P0ncJRbV-8_P"
      },
      "source": [
        "lable_train = list(pro1.Cancer_Type)\n",
        "pro_train = pro1.drop(columns = [\"Cancer_Type\", \"Sample_ID\"])"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4UT3DVK5_BpA"
      },
      "source": [
        "id = '1Mof6HE8iPsGB5-ZzXvE8pulizup1EVjM'\n",
        "downloaded = drive.CreateFile({'id':id}) \n",
        "downloaded.GetContentFile('pro.val.csv')  \n",
        "pro2 = pd.read_csv('pro.val.csv')"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PjWAphSp_CeN"
      },
      "source": [
        "lable_test = list(pro2.Cancer_Type)\n",
        "pro_test = pro2.drop(columns = [\"Cancer_Type\", \"Sample_ID\"])"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L3rsf-XGX-qp"
      },
      "source": [
        "le = LabelEncoder()\n",
        "lable1 = np.ravel(lable_train)\n",
        "le.fit(lable1)\n",
        "lable1 = le.transform(lable1)\n",
        "\n",
        "le = LabelEncoder()\n",
        "lable2 = np.ravel(lable_test)\n",
        "le.fit(lable2)\n",
        "lable2 = le.transform(lable2)\n"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DxQmmO8rYClc"
      },
      "source": [
        "cat1 = np_utils.to_categorical(lable1)\n",
        "cat1 = pd.DataFrame({'tag' + str(i) : cat1[:, i]for i in range(len(cat1[0]))})\n",
        "cat_train = cat1[:gene_train.shape[0]]\n",
        "\n",
        "cat2 = np_utils.to_categorical(lable2)\n",
        "cat2 = pd.DataFrame({'tag' + str(i) : cat2[:, i]for i in range(len(cat2[0]))})\n",
        "cat_test = cat2[:gene_test.shape[0]]"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "p1o41r2VZxm1"
      },
      "source": [
        "def model(m_rna_train, pro_train,label_train, m_rna_test, pro_test, label_test):\n",
        "        \"\"\"\n",
        "        We conducted an architecture of Variational Autoencoder that is capable of simultaneous inference of various properties of biological samples, \n",
        "        through multi-task and transfer learning. It encodes the whole transcription profile into a strikingly low-dimensional latent vector of size 24, and then \n",
        "        recovers mRNA and protein expression profiles, disease type from this vector. This latent space is significantly better than the original gene \n",
        "        expression profiles for discriminating samples based on their disease. \n",
        "        We employed this architecture on mRNA transcription profiles of 7150 clinical samples from 32 classes.\n",
        "        :param : m_rna_train, pro_train,label_train, m_rna_test, pro_test, label_test\n",
        "        :return: pred_train, pred_pro, score, history\n",
        "        \"\"\"\n",
        "        opt = keras.optimizers.Nadam(learning_rate=1e-3)\n",
        "\n",
        "        inputs = Input(shape=(m_rna_train.shape[1],), name=\"inputs\")\n",
        "        n_z = 24\n",
        "\n",
        "        # Q(z|x) -- encoder\n",
        "        inputs_noise = GaussianNoise(stddev=0.0)(inputs)\n",
        "        drop1 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_noise)\n",
        "        inputs_0 = BatchNormalization(name=\"inputs_0\")(drop1)\n",
        "        inputs_1 = Dense(1024, activation='relu', name=\"inputs_1\")(inputs_0)\n",
        "        inputs_2 = BatchNormalization(name=\"inputs_2\")(inputs_1)\n",
        "        drop2 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_2)\n",
        "        inputs_3 = Dense(512, activation='relu', name=\"inputs_3\")(drop2)\n",
        "        inputs_4 = BatchNormalization(name=\"inputs_4\")(inputs_3)\n",
        "        drop3 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_4)\n",
        "        inputs_5 = Dense(128, activation='relu', name=\"inputs_5\")(inputs_4)\n",
        "        inputs_6= BatchNormalization(name=\"inputs_6\")(inputs_5)\n",
        "        drop4 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_6)\n",
        "\n",
        "\n",
        "        h_q = Dense(n_z, activation=\"relu\", name=\"h_q\")(drop4)\n",
        "        mu = Dense(n_z, activation=None, name=\"mu\")(h_q)\n",
        "        log_sigma = Dense(n_z, activation=None, name=\"log_sigma\")(h_q)\n",
        "\n",
        "        def sample_z(args):\n",
        "            mu_samples, log_sigma_sample = args\n",
        "            eps = backend.random_normal(shape=(24, n_z), mean=0., stddev=1.)\n",
        "            return mu_samples + backend.exp(log_sigma_sample / 2) * eps\n",
        "\n",
        "        # Sample z ~ Q(z|x)\n",
        "        z = Lambda(sample_z, name=\"lambda\")([mu, log_sigma])\n",
        "\n",
        "        # P(x|z) -- decoder\n",
        "        decoder_hidden = Dense(24, activation='linear', name=\"decoder_hidden\")\n",
        "\n",
        "        # Decoder of Gene Expression Profile\n",
        "        h_p = decoder_hidden(z)\n",
        "        inputs_7 = BatchNormalization(name=\"inputs_7\")(h_p)\n",
        "        inputs_8 = Dense(256, activation='relu', name=\"inputs_8\")(inputs_7)\n",
        "        inputs_9 = BatchNormalization(name=\"inputs_9\")(inputs_8)\n",
        "        drop5 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_9)\n",
        "        inputs_11 = Dense(512, activation='relu', name=\"inputs_11\")(drop5)\n",
        "        inputs_12 = BatchNormalization(name=\"inputs_12\")(inputs_11)\n",
        "        drop6 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_12)\n",
        "        inputs_14 = Dense(1024, activation='relu', name=\"inputs_14\")(drop6)\n",
        "        inputs_15 = BatchNormalization(name=\"inputs_15\")(inputs_14)\n",
        "        drop7 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_15)\n",
        "        output_tcga = Dense(m_rna_train.shape[1], activation='relu', name=\"m_rna\")(drop7)\n",
        "\n",
        "        # Decoder of Protein Expression Profile\n",
        "        inputs_16 = BatchNormalization(name=\"inputs_16\")(h_p)\n",
        "        inputs_17 = Dense(64, activation='relu', name=\"inputs_17\")(inputs_16)\n",
        "        inputs_18 = BatchNormalization(name=\"inputs_18\")(inputs_17)\n",
        "        drop8 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_18)\n",
        "        inputs_19 = Dense(128, activation='relu', name=\"inputs_19\")(drop8)\n",
        "        inputs_20 = BatchNormalization(name=\"inputs_20\")(inputs_19)\n",
        "        drop9 = GaussianDropout(rate=0.5 ** 2 / (1 + 0.5 ** 2))(inputs_20)\n",
        "        output_pro = Dense(pro_train.shape[1], activation='relu', name=\"pro\")(drop9)\n",
        "\n",
        "        # Classifier\n",
        "        output_cl_0 = Dense(32, activation=\"softmax\", name=\"cl_cancer\")(h_p)\n",
        "\n",
        "\n",
        "        lambda_value = 2.5596e-6\n",
        "\n",
        "        def vae_loss(y_true, y_pred):\n",
        "            \"\"\" Calculate loss = reconstruction loss + KL loss for each data_pred in minibatch \"\"\"\n",
        "            # E[log P(x|z)]\n",
        "            recon = backend.mean(backend.square(y_true - y_pred), axis=1)\n",
        "            # D_KL(Q(z|x) || P(z|x)); calculate in closed form as both dist. are Gaussian\n",
        "            kl = 0.5 * backend.sum(backend.exp(log_sigma) + backend.square(mu) - 1. - log_sigma, axis=1)\n",
        "\n",
        "            return recon + lambda_value * kl\n",
        "\n",
        "        svae = Model(inputs, [output_tcga, output_pro, output_cl_0])\n",
        "\n",
        "        svae.compile(optimizer='nadam',\n",
        "                     loss=[vae_loss, \"mse\", \"categorical_crossentropy\"],\n",
        "                     loss_weights=[5e-2, 5e-2, 5e-1],\n",
        "                     metrics={\"m_rna\":[vae_loss, \"mse\", \"mae\"], \"pro\": [\"mae\", \"mse\"], \"cl_cancer\": \"acc\"})\n",
        "        \n",
        "        \n",
        "        history = svae.fit([m_rna_train], [m_rna_train, pro_train, label_train],\n",
        "                 batch_size=24, epochs=10, verbose=1, validation_data = (m_rna_test, [m_rna_test, pro_test, label_test]))\n",
        "        score = svae.evaluate(m_rna_test, [m_rna_test, pro_test, label_test], verbose=1,\n",
        "                              batch_size=24)\n",
        "        pred_pro = svae.predict(m_rna_test, verbose=1,\n",
        "                              batch_size=24)\n",
        "        pred_train = svae.predict(m_rna_train, verbose=1,\n",
        "                              batch_size=24)\n",
        "        return pred_train, pred_pro, score, history"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "i-zdkPKInLQJ",
        "outputId": "123b0c6f-8480-45fc-b49b-be2c6ece70ec"
      },
      "source": [
        "train_y = pro_train.to_numpy()\n",
        "test_y = pro_test.to_numpy()\n",
        "label_train = np.array(cat_train)\n",
        "label_test = np.array(cat_test)\n",
        "\n",
        "train_x_log = np.log2(gene_train)\n",
        "train_x_log1 = train_x_log.replace(to_replace = np.nan, value =0.00)\n",
        "train_x_log2 = train_x_log1.replace(to_replace =-math.inf, value =0.0000)\n",
        "arr = []\n",
        "for i in np.arange(train_x_log2.shape[1]):\n",
        "  idx = (train_x_log2.iloc[:,i]==0.0)\n",
        "  arr.append(len(train_x_log2.iloc[:,i][idx]))\n",
        "arr_idx = []\n",
        "for i in np.arange(train_x_log2.shape[1]):\n",
        "  if arr[i] <= 3000:\n",
        "    arr_idx.append(train_x_log2.columns[i])\n",
        "train_x_log3 = train_x_log2[arr_idx][:][:-2]\n",
        "val_x_log = np.log2(gene_test)\n",
        "val_x_log1 = val_x_log.replace(to_replace =-math.inf, value =0.0000)\n",
        "val_x_log2 = val_x_log1.replace(to_replace = np.nan, value =0.00)\n",
        "val_x_log3 = val_x_log2[arr_idx]\n",
        "val_x_log3 = val_x_log3[:][:-1]\n",
        "test_y = test_y[:-1][:]\n",
        "train_y = train_y[:-2][:]\n",
        "label_test = label_test[:-1][:]\n",
        "label_train = label_train[:-2][:]\n",
        "pred_train, pred_pro, score, history = model(train_x_log3, train_y, label_train, val_x_log3, test_y, label_test)"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:6: RuntimeWarning: divide by zero encountered in log2\n",
            "  \n",
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:18: RuntimeWarning: divide by zero encountered in log2\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Train on 5736 samples, validate on 672 samples\n",
            "Epoch 1/10\n",
            "5736/5736 [==============================] - ETA: 0s - loss: 0.9303 - m_rna_loss: 6.8152 - pro_loss: 0.5366 - cl_cancer_loss: 1.1254 - m_rna_vae_loss: 6.8152 - m_rna_mean_squared_error: 6.8150 - m_rna_mean_absolute_error: 1.8184 - pro_mean_absolute_error: 0.4987 - pro_mean_squared_error: 0.5366 - cl_cancer_acc: 0.7083"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/training.py:2325: UserWarning: `Model.state_updates` will be removed in a future version. This property should not be used in TensorFlow 2.0, as `updates` are applied automatically.\n",
            "  warnings.warn('`Model.state_updates` will be removed in a future version. '\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\r5736/5736 [==============================] - 160s 28ms/sample - loss: 0.9303 - m_rna_loss: 6.8152 - pro_loss: 0.5366 - cl_cancer_loss: 1.1254 - m_rna_vae_loss: 6.8152 - m_rna_mean_squared_error: 6.8150 - m_rna_mean_absolute_error: 1.8184 - pro_mean_absolute_error: 0.4987 - pro_mean_squared_error: 0.5366 - cl_cancer_acc: 0.7083 - val_loss: 0.5272 - val_m_rna_loss: 6.2790 - val_pro_loss: 0.4073 - val_cl_cancer_loss: 0.3858 - val_m_rna_vae_loss: 6.2790 - val_m_rna_mean_squared_error: 6.2783 - val_m_rna_mean_absolute_error: 1.6896 - val_pro_mean_absolute_error: 0.4220 - val_pro_mean_squared_error: 0.4073 - val_cl_cancer_acc: 0.9018\n",
            "Epoch 2/10\n",
            "5736/5736 [==============================] - 159s 28ms/sample - loss: 0.4708 - m_rna_loss: 5.6248 - pro_loss: 0.4083 - cl_cancer_loss: 0.3383 - m_rna_vae_loss: 5.6248 - m_rna_mean_squared_error: 5.6244 - m_rna_mean_absolute_error: 1.5479 - pro_mean_absolute_error: 0.4273 - pro_mean_squared_error: 0.4083 - cl_cancer_acc: 0.9038 - val_loss: 0.4254 - val_m_rna_loss: 5.4185 - val_pro_loss: 0.3818 - val_cl_cancer_loss: 0.2707 - val_m_rna_vae_loss: 5.4185 - val_m_rna_mean_squared_error: 5.4180 - val_m_rna_mean_absolute_error: 1.4990 - val_pro_mean_absolute_error: 0.4101 - val_pro_mean_squared_error: 0.3818 - val_cl_cancer_acc: 0.9315\n",
            "Epoch 3/10\n",
            "5736/5736 [==============================] - 166s 29ms/sample - loss: 0.4013 - m_rna_loss: 5.2956 - pro_loss: 0.3961 - cl_cancer_loss: 0.2334 - m_rna_vae_loss: 5.2956 - m_rna_mean_squared_error: 5.2952 - m_rna_mean_absolute_error: 1.4719 - pro_mean_absolute_error: 0.4201 - pro_mean_squared_error: 0.3961 - cl_cancer_acc: 0.9268 - val_loss: 0.4272 - val_m_rna_loss: 5.2413 - val_pro_loss: 0.3756 - val_cl_cancer_loss: 0.2926 - val_m_rna_vae_loss: 5.2413 - val_m_rna_mean_squared_error: 5.2407 - val_m_rna_mean_absolute_error: 1.4592 - val_pro_mean_absolute_error: 0.4083 - val_pro_mean_squared_error: 0.3756 - val_cl_cancer_acc: 0.9301\n",
            "Epoch 4/10\n",
            "5736/5736 [==============================] - 157s 27ms/sample - loss: 0.3745 - m_rna_loss: 5.1684 - pro_loss: 0.3885 - cl_cancer_loss: 0.1933 - m_rna_vae_loss: 5.1684 - m_rna_mean_squared_error: 5.1679 - m_rna_mean_absolute_error: 1.4383 - pro_mean_absolute_error: 0.4161 - pro_mean_squared_error: 0.3885 - cl_cancer_acc: 0.9390 - val_loss: 0.4074 - val_m_rna_loss: 5.0620 - val_pro_loss: 0.3658 - val_cl_cancer_loss: 0.2720 - val_m_rna_vae_loss: 5.0620 - val_m_rna_mean_squared_error: 5.0614 - val_m_rna_mean_absolute_error: 1.4087 - val_pro_mean_absolute_error: 0.4027 - val_pro_mean_squared_error: 0.3658 - val_cl_cancer_acc: 0.9167\n",
            "Epoch 5/10\n",
            "5736/5736 [==============================] - 159s 28ms/sample - loss: 0.3528 - m_rna_loss: 5.1303 - pro_loss: 0.3817 - cl_cancer_loss: 0.1544 - m_rna_vae_loss: 5.1303 - m_rna_mean_squared_error: 5.1298 - m_rna_mean_absolute_error: 1.4256 - pro_mean_absolute_error: 0.4125 - pro_mean_squared_error: 0.3817 - cl_cancer_acc: 0.9496 - val_loss: 0.4124 - val_m_rna_loss: 5.0880 - val_pro_loss: 0.3642 - val_cl_cancer_loss: 0.2797 - val_m_rna_vae_loss: 5.0880 - val_m_rna_mean_squared_error: 5.0874 - val_m_rna_mean_absolute_error: 1.4132 - val_pro_mean_absolute_error: 0.4012 - val_pro_mean_squared_error: 0.3642 - val_cl_cancer_acc: 0.9211\n",
            "Epoch 6/10\n",
            "5736/5736 [==============================] - 159s 28ms/sample - loss: 0.3302 - m_rna_loss: 5.1119 - pro_loss: 0.3777 - cl_cancer_loss: 0.1114 - m_rna_vae_loss: 5.1119 - m_rna_mean_squared_error: 5.1114 - m_rna_mean_absolute_error: 1.4186 - pro_mean_absolute_error: 0.4099 - pro_mean_squared_error: 0.3777 - cl_cancer_acc: 0.9622 - val_loss: 0.3994 - val_m_rna_loss: 5.0168 - val_pro_loss: 0.3583 - val_cl_cancer_loss: 0.2613 - val_m_rna_vae_loss: 5.0168 - val_m_rna_mean_squared_error: 5.0162 - val_m_rna_mean_absolute_error: 1.3886 - val_pro_mean_absolute_error: 0.3975 - val_pro_mean_squared_error: 0.3583 - val_cl_cancer_acc: 0.9375\n",
            "Epoch 7/10\n",
            "5736/5736 [==============================] - 162s 28ms/sample - loss: 0.3294 - m_rna_loss: 5.1020 - pro_loss: 0.3751 - cl_cancer_loss: 0.1111 - m_rna_vae_loss: 5.1020 - m_rna_mean_squared_error: 5.1015 - m_rna_mean_absolute_error: 1.4144 - pro_mean_absolute_error: 0.4079 - pro_mean_squared_error: 0.3751 - cl_cancer_acc: 0.9613 - val_loss: 0.4054 - val_m_rna_loss: 5.0101 - val_pro_loss: 0.3586 - val_cl_cancer_loss: 0.2739 - val_m_rna_vae_loss: 5.0101 - val_m_rna_mean_squared_error: 5.0095 - val_m_rna_mean_absolute_error: 1.3870 - val_pro_mean_absolute_error: 0.3976 - val_pro_mean_squared_error: 0.3586 - val_cl_cancer_acc: 0.9301\n",
            "Epoch 8/10\n",
            "5736/5736 [==============================] - 162s 28ms/sample - loss: 0.3221 - m_rna_loss: 5.0895 - pro_loss: 0.3731 - cl_cancer_loss: 0.0980 - m_rna_vae_loss: 5.0895 - m_rna_mean_squared_error: 5.0889 - m_rna_mean_absolute_error: 1.4100 - pro_mean_absolute_error: 0.4065 - pro_mean_squared_error: 0.3731 - cl_cancer_acc: 0.9662 - val_loss: 0.4183 - val_m_rna_loss: 5.0338 - val_pro_loss: 0.3555 - val_cl_cancer_loss: 0.2978 - val_m_rna_vae_loss: 5.0338 - val_m_rna_mean_squared_error: 5.0331 - val_m_rna_mean_absolute_error: 1.3904 - val_pro_mean_absolute_error: 0.3956 - val_pro_mean_squared_error: 0.3555 - val_cl_cancer_acc: 0.9271\n",
            "Epoch 9/10\n",
            "5736/5736 [==============================] - 161s 28ms/sample - loss: 0.3115 - m_rna_loss: 5.0765 - pro_loss: 0.3701 - cl_cancer_loss: 0.0783 - m_rna_vae_loss: 5.0765 - m_rna_mean_squared_error: 5.0759 - m_rna_mean_absolute_error: 1.4051 - pro_mean_absolute_error: 0.4046 - pro_mean_squared_error: 0.3701 - cl_cancer_acc: 0.9768 - val_loss: 0.3919 - val_m_rna_loss: 4.9913 - val_pro_loss: 0.3575 - val_cl_cancer_loss: 0.2489 - val_m_rna_vae_loss: 4.9913 - val_m_rna_mean_squared_error: 4.9907 - val_m_rna_mean_absolute_error: 1.3784 - val_pro_mean_absolute_error: 0.3961 - val_pro_mean_squared_error: 0.3575 - val_cl_cancer_acc: 0.9286\n",
            "Epoch 10/10\n",
            "5736/5736 [==============================] - 159s 28ms/sample - loss: 0.3178 - m_rna_loss: 5.0703 - pro_loss: 0.3694 - cl_cancer_loss: 0.0916 - m_rna_vae_loss: 5.0703 - m_rna_mean_squared_error: 5.0697 - m_rna_mean_absolute_error: 1.4025 - pro_mean_absolute_error: 0.4039 - pro_mean_squared_error: 0.3694 - cl_cancer_acc: 0.9725 - val_loss: 0.3928 - val_m_rna_loss: 4.9942 - val_pro_loss: 0.3529 - val_cl_cancer_loss: 0.2509 - val_m_rna_vae_loss: 4.9942 - val_m_rna_mean_squared_error: 4.9936 - val_m_rna_mean_absolute_error: 1.3808 - val_pro_mean_absolute_error: 0.3932 - val_pro_mean_squared_error: 0.3529 - val_cl_cancer_acc: 0.9390\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Gc8Ah1CtNEBE",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "cea285d1-6ca1-4f7d-af75-090f0ed8bfea"
      },
      "source": [
        "def GetAverageRank(test, predict):\n",
        "    p = predict\n",
        "    rank  = 0\n",
        "    for k in range(len(p)):\n",
        "        temp = np.vstack((test, p[k]))\n",
        "        for i in range(temp.shape[0]):\n",
        "            temp[i] = np.subtract(temp[i], test[k])\n",
        "        temp = np.square(temp)\n",
        "        sum = temp.sum(axis = 1)\n",
        "        arr = []\n",
        "        for i in range(len(sum)):\n",
        "            arr.append((sum[i], i))\n",
        "        arr = sorted(arr, key=lambda x: x[0])\n",
        "        for i in range(len(arr)):\n",
        "            if arr[i][1] == len(p):\n",
        "                rank += i\n",
        "                break\n",
        "    rank /= len(p)\n",
        "    return rank\n",
        "\n",
        "print(GetAverageRank(test_y,pred_pro[1]))"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "13.705357142857142\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SJ2Gw4PSahwj",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "9fde96f5-a484-4c38-9cb4-2c6879be4ce0"
      },
      "source": [
        "print(\"Eval Metrics for Test Set\")\n",
        "print(\"explained_variance_score\", explained_variance_score(test_y, pred_pro[1]))\n",
        "print(\"median_absolute_error\", median_absolute_error(test_y, pred_pro[1]))\n",
        "print(\"r2_score\", r2_score(test_y, pred_pro[1]))\n",
        "print(\"RMSE\", sqrt(mean_squared_error(test_y, pred_pro[1])))\n",
        "print(\"mean_absolute_error\", mean_absolute_error(test_y, pred_pro[1]))\n",
        "print(\"mean_squared_error\", mean_squared_error(test_y, pred_pro[1]))"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Eval Metrics for Test Set\n",
            "explained_variance_score 0.20025790537236318\n",
            "median_absolute_error 0.3164902983194564\n",
            "r2_score 0.13966418640429293\n",
            "RMSE 0.5938639618948646\n",
            "mean_absolute_error 0.3932677686016908\n",
            "mean_squared_error 0.3526744052374652\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rUftL5P6a2WR",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "3904427a-5e07-4744-a61b-ba3f92e4e54a"
      },
      "source": [
        "print(\"Eval Metrics for Train Set\")\n",
        "print(\"explained_variance_score\", explained_variance_score(train_y,pred_train[1]))\n",
        "print(\"median_absolute_error\", median_absolute_error(train_y,pred_train[1]))\n",
        "print(\"r2_score\", r2_score(train_y,pred_train[1]))\n",
        "print(\"RMSE\", sqrt(mean_squared_error(train_y,pred_train[1])))\n",
        "print(\"mean_absolute_error\", mean_absolute_error(train_y,pred_train[1]))\n",
        "print(\"mean_squared_error\", mean_squared_error(train_y,pred_train[1]))"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Eval Metrics for Train Set\n",
            "explained_variance_score 0.20967799912230342\n",
            "median_absolute_error 0.31880421773218576\n",
            "r2_score 0.14456520605371578\n",
            "RMSE 0.5951626283633519\n",
            "mean_absolute_error 0.3944720391430388\n",
            "mean_squared_error 0.3542185542003733\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "id": "EMJq_20Ll0Vt",
        "outputId": "576f227d-f86e-4335-ded0-9b89d25e4882"
      },
      "source": [
        "pyplot.plot(history.history['cl_cancer_acc'], color = 'green', label = 'classifier_acc')\n",
        "pyplot.xlabel=\"epoch\"\n",
        "pyplot.plot(history.history['val_cl_cancer_acc'], color = 'red', label = \"val_classifier_acc\")\n",
        "\n",
        "pyplot.plot(history.history['cl_cancer_loss'], color = 'Blue', label = \"classifier_loss\")\n",
        "pyplot.plot(history.history['val_cl_cancer_loss'], color = 'yellow', label = \"val_classifier_loss\")\n",
        "\n",
        "pyplot.title('Accuracy')\n",
        "pyplot.legend()\n",
        "pyplot.savefig(\"plot_acc.png\")"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 281
        },
        "id": "abQDZxpBmBdw",
        "outputId": "191529ef-c33d-4a2f-ae94-a6b8a6ecd919"
      },
      "source": [
        "pyplot.plot(history.history['pro_loss'], color = 'green', label = 'pro_loss')\n",
        "pyplot.xlabel=\"epoch\"\n",
        "pyplot.plot(history.history['val_pro_loss'], color = 'red', label = \"val_pro_loss\")\n",
        "pyplot.plot(history.history['pro_mean_absolute_error'], color = 'black', label = 'pro_mean_absolute_error')\n",
        "pyplot.plot(history.history['val_pro_mean_absolute_error'], color = 'yellow', label = 'val_pro_mean_absolute_error')\n",
        "pyplot.title('Loss')\n",
        "pyplot.legend()\n",
        "pyplot.savefig(\"plot_loss.png\")"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}