03 LSTM_model.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eeUZkO7xOr_D"
      },
      "source": [
        "<img src=\"https://github.com/AmsterdamUMC/AmsterdamUMCdb/blob/master/img/logo_c4i_square.png?raw=1\" alt=\"Logo C4I\" width=128px/>\n",
        "\n",
        "<img src=\"https://github.com/AmsterdamUMC/AmsterdamUMCdb/blob/master/img/logo_amds.png?raw=1\" alt=\"Logo AMDS\" width=128px/>\n",
        "\n",
        "# AmsterdamUMCdb with Google BigQuery and Colaboratory\n",
        "## AmsterdamUMCdb - Freely Accessible ICU Database\n",
        "## Task 2: DL model for septic patients"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wkXGVM33tJaP"
      },
      "source": [
        "# Libraries"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "WVZYCgT5QRAG"
      },
      "outputs": [],
      "source": [
        "!pip install keras --quiet"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 2,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KasbePRjQJl1",
        "outputId": "8b8e80b6-11aa-4a64-b2ee-845e0f9628fd"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "\u001b[?25l     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m0.0/275.7 kB\u001b[0m \u001b[31m?\u001b[0m eta \u001b[36m-:--:--\u001b[0m\r\u001b[2K     \u001b[91m━━━━\u001b[0m\u001b[90m╺\u001b[0m\u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m30.7/275.7 kB\u001b[0m \u001b[31m793.7 kB/s\u001b[0m eta \u001b[36m0:00:01\u001b[0m\r\u001b[2K     \u001b[91m━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[90m╺\u001b[0m\u001b[90m━━━━━━━━━━━━━━\u001b[0m \u001b[32m174.1/275.7 kB\u001b[0m \u001b[31m2.3 MB/s\u001b[0m eta \u001b[36m0:00:01\u001b[0m\r\u001b[2K     \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m275.7/275.7 kB\u001b[0m \u001b[31m2.8 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n",
            "\u001b[?25h  Preparing metadata (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Building wheel for lime (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ]
        }
      ],
      "source": [
        "!pip install lime --quiet"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 3,
      "metadata": {
        "id": "1PTVlPCN9g5N"
      },
      "outputs": [],
      "source": [
        "!pip install tensorflow --quiet"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 4,
      "metadata": {
        "id": "ZO4fJTLyQVf0"
      },
      "outputs": [],
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import random\n",
        "import tensorflow as tf"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Y0vyEJxKGBBL"
      },
      "source": [
        "# Train-test split"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 384
        },
        "id": "Q3RvlcXiWX7l",
        "outputId": "97d43c93-7794-4daa-aa5f-54e8a7e93706"
      },
      "outputs": [
        {
          "data": {
            "text/html": [
              "\n",
              "  <div id=\"df-b775061c-9dc1-43a3-87bc-064ca48f7313\" class=\"colab-df-container\">\n",
              "    <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>admissionid</th>\n",
              "      <th>time</th>\n",
              "      <th>steroids</th>\n",
              "      <th>antibiotics</th>\n",
              "      <th>admissioncount</th>\n",
              "      <th>gender</th>\n",
              "      <th>age</th>\n",
              "      <th>weight</th>\n",
              "      <th>sofa_resp</th>\n",
              "      <th>sofa_cns</th>\n",
              "      <th>...</th>\n",
              "      <th>Leucocyten</th>\n",
              "      <th>Thrombocyten</th>\n",
              "      <th>Bilirubine</th>\n",
              "      <th>Kreatinine</th>\n",
              "      <th>Lactaat</th>\n",
              "      <th>Breathing Rate Spontaneous</th>\n",
              "      <th>Sodium</th>\n",
              "      <th>Calcium</th>\n",
              "      <th>HR</th>\n",
              "      <th>Saturation</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>...</td>\n",
              "      <td>5.552900e+04</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "      <td>55529.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mean</th>\n",
              "      <td>12144.419384</td>\n",
              "      <td>164.907130</td>\n",
              "      <td>0.351096</td>\n",
              "      <td>0.872589</td>\n",
              "      <td>1.150768</td>\n",
              "      <td>0.401466</td>\n",
              "      <td>56.416575</td>\n",
              "      <td>71.126258</td>\n",
              "      <td>0.420321</td>\n",
              "      <td>17.960435</td>\n",
              "      <td>...</td>\n",
              "      <td>4.172303e+02</td>\n",
              "      <td>236.206469</td>\n",
              "      <td>22.012570</td>\n",
              "      <td>123.589044</td>\n",
              "      <td>1.914299</td>\n",
              "      <td>15.341875</td>\n",
              "      <td>142.191648</td>\n",
              "      <td>4.157478</td>\n",
              "      <td>89.411803</td>\n",
              "      <td>97.230276</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>std</th>\n",
              "      <td>6684.116227</td>\n",
              "      <td>150.765077</td>\n",
              "      <td>0.477317</td>\n",
              "      <td>0.333436</td>\n",
              "      <td>0.467650</td>\n",
              "      <td>0.490199</td>\n",
              "      <td>17.472973</td>\n",
              "      <td>20.440044</td>\n",
              "      <td>0.618877</td>\n",
              "      <td>10.927056</td>\n",
              "      <td>...</td>\n",
              "      <td>5.477353e+04</td>\n",
              "      <td>171.941634</td>\n",
              "      <td>52.363768</td>\n",
              "      <td>99.479112</td>\n",
              "      <td>2.018998</td>\n",
              "      <td>10.274632</td>\n",
              "      <td>6.341496</td>\n",
              "      <td>0.491417</td>\n",
              "      <td>19.424117</td>\n",
              "      <td>3.252930</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>min</th>\n",
              "      <td>11.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>18.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>...</td>\n",
              "      <td>0.000000e+00</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>-0.800000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.700000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>25%</th>\n",
              "      <td>6409.000000</td>\n",
              "      <td>40.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>50.000000</td>\n",
              "      <td>60.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>18.000000</td>\n",
              "      <td>...</td>\n",
              "      <td>8.800000e+00</td>\n",
              "      <td>113.000000</td>\n",
              "      <td>5.000000</td>\n",
              "      <td>64.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>11.000000</td>\n",
              "      <td>138.000000</td>\n",
              "      <td>3.800000</td>\n",
              "      <td>76.000000</td>\n",
              "      <td>96.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50%</th>\n",
              "      <td>12645.000000</td>\n",
              "      <td>120.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>60.000000</td>\n",
              "      <td>70.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>18.000000</td>\n",
              "      <td>...</td>\n",
              "      <td>1.250000e+01</td>\n",
              "      <td>201.000000</td>\n",
              "      <td>9.000000</td>\n",
              "      <td>91.000000</td>\n",
              "      <td>1.400000</td>\n",
              "      <td>16.000000</td>\n",
              "      <td>142.000000</td>\n",
              "      <td>4.100000</td>\n",
              "      <td>88.000000</td>\n",
              "      <td>98.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>75%</th>\n",
              "      <td>17853.000000</td>\n",
              "      <td>248.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>70.000000</td>\n",
              "      <td>80.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>18.000000</td>\n",
              "      <td>...</td>\n",
              "      <td>1.780000e+01</td>\n",
              "      <td>315.000000</td>\n",
              "      <td>17.000000</td>\n",
              "      <td>147.000000</td>\n",
              "      <td>2.100000</td>\n",
              "      <td>24.000000</td>\n",
              "      <td>146.000000</td>\n",
              "      <td>4.400000</td>\n",
              "      <td>102.000000</td>\n",
              "      <td>99.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>max</th>\n",
              "      <td>23535.000000</td>\n",
              "      <td>712.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>5.000000</td>\n",
              "      <td>1.000000</td>\n",
              "      <td>80.000000</td>\n",
              "      <td>110.000000</td>\n",
              "      <td>2.000000</td>\n",
              "      <td>72.000000</td>\n",
              "      <td>...</td>\n",
              "      <td>7.452101e+06</td>\n",
              "      <td>1351.000000</td>\n",
              "      <td>926.000000</td>\n",
              "      <td>1724.000000</td>\n",
              "      <td>38.099998</td>\n",
              "      <td>44.000000</td>\n",
              "      <td>177.000000</td>\n",
              "      <td>7.500000</td>\n",
              "      <td>206.000000</td>\n",
              "      <td>398.000000</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>8 rows × 26 columns</p>\n",
              "</div>\n",
              "    <div class=\"colab-df-buttons\">\n",
              "\n",
              "  <div class=\"colab-df-container\">\n",
              "    <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-b775061c-9dc1-43a3-87bc-064ca48f7313')\"\n",
              "            title=\"Convert this dataframe to an interactive table.\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\" viewBox=\"0 -960 960 960\">\n",
              "    <path d=\"M120-120v-720h720v720H120Zm60-500h600v-160H180v160Zm220 220h160v-160H400v160Zm0 220h160v-160H400v160ZM180-400h160v-160H180v160Zm440 0h160v-160H620v160ZM180-180h160v-160H180v160Zm440 0h160v-160H620v160Z\"/>\n",
              "  </svg>\n",
              "    </button>\n",
              "\n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    .colab-df-buttons div {\n",
              "      margin-bottom: 4px;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "    <script>\n",
              "      const buttonEl =\n",
              "        document.querySelector('#df-b775061c-9dc1-43a3-87bc-064ca48f7313 button.colab-df-convert');\n",
              "      buttonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "      async function convertToInteractive(key) {\n",
              "        const element = document.querySelector('#df-b775061c-9dc1-43a3-87bc-064ca48f7313');\n",
              "        const dataTable =\n",
              "          await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                    [key], {});\n",
              "        if (!dataTable) return;\n",
              "\n",
              "        const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "          '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "          + ' to learn more about interactive tables.';\n",
              "        element.innerHTML = '';\n",
              "        dataTable['output_type'] = 'display_data';\n",
              "        await google.colab.output.renderOutput(dataTable, element);\n",
              "        const docLink = document.createElement('div');\n",
              "        docLink.innerHTML = docLinkHtml;\n",
              "        element.appendChild(docLink);\n",
              "      }\n",
              "    </script>\n",
              "  </div>\n",
              "\n",
              "\n",
              "<div id=\"df-2a3683df-4edf-4ff1-8020-262d2f2d31f4\">\n",
              "  <button class=\"colab-df-quickchart\" onclick=\"quickchart('df-2a3683df-4edf-4ff1-8020-262d2f2d31f4')\"\n",
              "            title=\"Suggest charts\"\n",
              "            style=\"display:none;\">\n",
              "\n",
              "<svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "     width=\"24px\">\n",
              "    <g>\n",
              "        <path d=\"M19 3H5c-1.1 0-2 .9-2 2v14c0 1.1.9 2 2 2h14c1.1 0 2-.9 2-2V5c0-1.1-.9-2-2-2zM9 17H7v-7h2v7zm4 0h-2V7h2v10zm4 0h-2v-4h2v4z\"/>\n",
              "    </g>\n",
              "</svg>\n",
              "  </button>\n",
              "\n",
              "<style>\n",
              "  .colab-df-quickchart {\n",
              "      --bg-color: #E8F0FE;\n",
              "      --fill-color: #1967D2;\n",
              "      --hover-bg-color: #E2EBFA;\n",
              "      --hover-fill-color: #174EA6;\n",
              "      --disabled-fill-color: #AAA;\n",
              "      --disabled-bg-color: #DDD;\n",
              "  }\n",
              "\n",
              "  [theme=dark] .colab-df-quickchart {\n",
              "      --bg-color: #3B4455;\n",
              "      --fill-color: #D2E3FC;\n",
              "      --hover-bg-color: #434B5C;\n",
              "      --hover-fill-color: #FFFFFF;\n",
              "      --disabled-bg-color: #3B4455;\n",
              "      --disabled-fill-color: #666;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart {\n",
              "    background-color: var(--bg-color);\n",
              "    border: none;\n",
              "    border-radius: 50%;\n",
              "    cursor: pointer;\n",
              "    display: none;\n",
              "    fill: var(--fill-color);\n",
              "    height: 32px;\n",
              "    padding: 0;\n",
              "    width: 32px;\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart:hover {\n",
              "    background-color: var(--hover-bg-color);\n",
              "    box-shadow: 0 1px 2px rgba(60, 64, 67, 0.3), 0 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "    fill: var(--button-hover-fill-color);\n",
              "  }\n",
              "\n",
              "  .colab-df-quickchart-complete:disabled,\n",
              "  .colab-df-quickchart-complete:disabled:hover {\n",
              "    background-color: var(--disabled-bg-color);\n",
              "    fill: var(--disabled-fill-color);\n",
              "    box-shadow: none;\n",
              "  }\n",
              "\n",
              "  .colab-df-spinner {\n",
              "    border: 2px solid var(--fill-color);\n",
              "    border-color: transparent;\n",
              "    border-bottom-color: var(--fill-color);\n",
              "    animation:\n",
              "      spin 1s steps(1) infinite;\n",
              "  }\n",
              "\n",
              "  @keyframes spin {\n",
              "    0% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "      border-left-color: var(--fill-color);\n",
              "    }\n",
              "    20% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    30% {\n",
              "      border-color: transparent;\n",
              "      border-left-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    40% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-top-color: var(--fill-color);\n",
              "    }\n",
              "    60% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "    }\n",
              "    80% {\n",
              "      border-color: transparent;\n",
              "      border-right-color: var(--fill-color);\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "    90% {\n",
              "      border-color: transparent;\n",
              "      border-bottom-color: var(--fill-color);\n",
              "    }\n",
              "  }\n",
              "</style>\n",
              "\n",
              "  <script>\n",
              "    async function quickchart(key) {\n",
              "      const quickchartButtonEl =\n",
              "        document.querySelector('#' + key + ' button');\n",
              "      quickchartButtonEl.disabled = true;  // To prevent multiple clicks.\n",
              "      quickchartButtonEl.classList.add('colab-df-spinner');\n",
              "      try {\n",
              "        const charts = await google.colab.kernel.invokeFunction(\n",
              "            'suggestCharts', [key], {});\n",
              "      } catch (error) {\n",
              "        console.error('Error during call to suggestCharts:', error);\n",
              "      }\n",
              "      quickchartButtonEl.classList.remove('colab-df-spinner');\n",
              "      quickchartButtonEl.classList.add('colab-df-quickchart-complete');\n",
              "    }\n",
              "    (() => {\n",
              "      let quickchartButtonEl =\n",
              "        document.querySelector('#df-2a3683df-4edf-4ff1-8020-262d2f2d31f4 button');\n",
              "      quickchartButtonEl.style.display =\n",
              "        google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "    })();\n",
              "  </script>\n",
              "</div>\n",
              "    </div>\n",
              "  </div>\n"
            ],
            "text/plain": [
              "        admissionid          time      steroids   antibiotics  admissioncount  \\\n",
              "count  55529.000000  55529.000000  55529.000000  55529.000000    55529.000000   \n",
              "mean   12144.419384    164.907130      0.351096      0.872589        1.150768   \n",
              "std     6684.116227    150.765077      0.477317      0.333436        0.467650   \n",
              "min       11.000000      0.000000      0.000000      0.000000        1.000000   \n",
              "25%     6409.000000     40.000000      0.000000      1.000000        1.000000   \n",
              "50%    12645.000000    120.000000      0.000000      1.000000        1.000000   \n",
              "75%    17853.000000    248.000000      1.000000      1.000000        1.000000   \n",
              "max    23535.000000    712.000000      1.000000      1.000000        5.000000   \n",
              "\n",
              "             gender           age        weight     sofa_resp      sofa_cns  \\\n",
              "count  55529.000000  55529.000000  55529.000000  55529.000000  55529.000000   \n",
              "mean       0.401466     56.416575     71.126258      0.420321     17.960435   \n",
              "std        0.490199     17.472973     20.440044      0.618877     10.927056   \n",
              "min        0.000000     18.000000      0.000000      0.000000      0.000000   \n",
              "25%        0.000000     50.000000     60.000000      0.000000     18.000000   \n",
              "50%        0.000000     60.000000     70.000000      0.000000     18.000000   \n",
              "75%        1.000000     70.000000     80.000000      1.000000     18.000000   \n",
              "max        1.000000     80.000000    110.000000      2.000000     72.000000   \n",
              "\n",
              "       ...    Leucocyten  Thrombocyten    Bilirubine    Kreatinine  \\\n",
              "count  ...  5.552900e+04  55529.000000  55529.000000  55529.000000   \n",
              "mean   ...  4.172303e+02    236.206469     22.012570    123.589044   \n",
              "std    ...  5.477353e+04    171.941634     52.363768     99.479112   \n",
              "min    ...  0.000000e+00      0.000000      0.000000      0.000000   \n",
              "25%    ...  8.800000e+00    113.000000      5.000000     64.000000   \n",
              "50%    ...  1.250000e+01    201.000000      9.000000     91.000000   \n",
              "75%    ...  1.780000e+01    315.000000     17.000000    147.000000   \n",
              "max    ...  7.452101e+06   1351.000000    926.000000   1724.000000   \n",
              "\n",
              "            Lactaat  Breathing Rate Spontaneous        Sodium       Calcium  \\\n",
              "count  55529.000000                55529.000000  55529.000000  55529.000000   \n",
              "mean       1.914299                   15.341875    142.191648      4.157478   \n",
              "std        2.018998                   10.274632      6.341496      0.491417   \n",
              "min       -0.800000                    0.000000      0.000000      1.700000   \n",
              "25%        1.000000                   11.000000    138.000000      3.800000   \n",
              "50%        1.400000                   16.000000    142.000000      4.100000   \n",
              "75%        2.100000                   24.000000    146.000000      4.400000   \n",
              "max       38.099998                   44.000000    177.000000      7.500000   \n",
              "\n",
              "                 HR    Saturation  \n",
              "count  55529.000000  55529.000000  \n",
              "mean      89.411803     97.230276  \n",
              "std       19.424117      3.252930  \n",
              "min        0.000000      1.000000  \n",
              "25%       76.000000     96.000000  \n",
              "50%       88.000000     98.000000  \n",
              "75%      102.000000     99.000000  \n",
              "max      206.000000    398.000000  \n",
              "\n",
              "[8 rows x 26 columns]"
            ]
          },
          "execution_count": 7,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "df.describe()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wqw87Hm7GdGY",
        "outputId": "c1c7abc5-2e21-47ff-c074-b7b10da1c61d"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "2809"
            ]
          },
          "execution_count": 8,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "id_list = list(df['admissionid'].unique())\n",
        "SPLIT_SIZE = 0.2\n",
        "len(id_list)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gj5TDS0kGAa0",
        "outputId": "fec01f08-725b-47b5-824d-d3f06f948451"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "train is 2247\n",
            "val and test are 281\n"
          ]
        }
      ],
      "source": [
        "id_train, id_test_val = train_test_split(id_list, test_size = SPLIT_SIZE, random_state = 42) # train set is 80%)\n",
        "print(\"train is %d\" % len(id_train))\n",
        "# remaining 20% split in halves as test and validation 10% and 10%\n",
        "id_valid, id_test = train_test_split(id_test_val, test_size = 0.5, random_state = 42) # test 10% valid 10%\n",
        "print(\"val and test are %d\" %len(id_test))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "metadata": {
        "id": "CRlG4sNfNBHR"
      },
      "outputs": [],
      "source": [
        "train = df[df.admissionid.isin(id_train)].sort_values(by=['admissionid'])\n",
        "test = df[df.admissionid.isin(id_test)].sort_values(by=['admissionid'])\n",
        "validation = df[df.admissionid.isin(id_valid)].sort_values(by=['admissionid'])\n",
        "\n",
        "test = test.sort_values(by=['admissionid', 'time'], ignore_index = True)\n",
        "train = train.sort_values(by=['admissionid', 'time'], ignore_index = True)\n",
        "validation = validation.sort_values(by=['admissionid', 'time'], ignore_index = True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "b9NAgX-G56eZ",
        "outputId": "dcd306a5-3aa4-4a84-b2d0-a93b699089d2"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "Index(['admissionid', 'time', 'steroids', 'antibiotics', 'admissioncount',\n",
              "       'gender', 'age', 'weight', 'sofa_resp', 'sofa_cns', 'sofa_cardio',\n",
              "       'deceased', 'LOS', 'dose_drug', 'Hematocrit', 'Hemoglobine',\n",
              "       'Leucocyten', 'Thrombocyten', 'Bilirubine', 'Kreatinine', 'Lactaat',\n",
              "       'Breathing Rate Spontaneous', 'Sodium', 'Calcium', 'HR', 'Saturation'],\n",
              "      dtype='object')"
            ]
          },
          "execution_count": 11,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "train.columns"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pnUPhbWc2rKq"
      },
      "source": [
        "# LSTM\n",
        "\n",
        "https://www.kaggle.com/code/prashant111/explain-your-model-predictions-with-lime\n",
        "\n",
        "https://github.com/Amario1306619051/Patient-Mortality-Prediction-Using-LSTM?tab=readme-ov-file"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qyaSqLcHDsnA"
      },
      "source": [
        "## Pad variable length sequences"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "metadata": {
        "id": "VQvFGyl7Euut"
      },
      "outputs": [],
      "source": [
        "col_x = ['admissionid', 'time','steroids', 'antibiotics', 'admissioncount',\n",
        "       'gender', 'age', 'weight', 'sofa_resp', 'sofa_cns', 'sofa_cardio',\n",
        "       'dose_drug', 'Hematocrit', 'Hemoglobine',\n",
        "       'Leucocyten', 'Thrombocyten', 'Bilirubine', 'Kreatinine', 'Lactaat',\n",
        "       'Breathing Rate Spontaneous', 'Sodium', 'Calcium', 'HR', 'Saturation']\n",
        "col_y = ['admissionid', 'time', \"deceased\"]\n",
        "\n",
        "x_train = train[col_x]\n",
        "y_train = train[col_y]\n",
        "x_validation = validation[col_x]\n",
        "y_validation = validation[col_y]\n",
        "x_test = test[col_x]\n",
        "y_test = test[col_y]\n",
        "\n",
        "x_train.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_train.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "x_validation.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_validation.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "metadata": {
        "id": "SdxAA9OZI1Lc"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.preprocessing import sequence\n",
        "\n",
        "# max number of sequence length\n",
        "maxlen = 90 # 720 hours max LOS divided by 8 hours timestep\n",
        "\n",
        "# get a list of unique patient encounter IDs\n",
        "teId = x_train.index.levels[1]\n",
        "veId = x_validation.index.levels[1]\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "# pad every patient sequence with 0s to be the same length,\n",
        "# then transforms the list of sequences to one numpy array\n",
        "# this is for efficient minibatching and GPU computations\n",
        "X_train = [x_train.loc[(slice(None), patient), :].values for patient in teId]\n",
        "y_train = [y_train.loc[(slice(None), patient), :].values for patient in teId]\n",
        "\n",
        "X_train = sequence.pad_sequences(X_train, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_train = sequence.pad_sequences(y_train, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "\n",
        "# repeat for the validation data\n",
        "\n",
        "X_valid = [x_validation.loc[(slice(None), patient), :].values for patient in veId]\n",
        "y_valid = [y_validation.loc[(slice(None), patient), :].values for patient in veId]\n",
        "\n",
        "X_valid = sequence.pad_sequences(X_valid, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_valid = sequence.pad_sequences(y_valid, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 14,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "D9UxVKjnGOrS",
        "outputId": "a3ae48ef-61e1-4d32-edb2-ed1e334b79dd"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_train shape: (2247, 90, 22) | y_train shape: (2247, 90, 1)\n",
            "X_valid shape: (281, 90, 22) | y_valid shape: (281, 90, 1)\n",
            "X_test shape: (281, 90, 22) | y_test shape: (281, 90, 1)\n"
          ]
        }
      ],
      "source": [
        "# print the shape of the array which will be used by the network\n",
        "# the shape is of the form (# of encounters, length of sequence, # of features)\n",
        "print(\"X_train shape: %s | y_train shape: %s\" % (str(X_train.shape), str(y_train.shape)))\n",
        "print(\"X_valid shape: %s | y_valid shape: %s\" % (str(X_valid.shape), str(y_valid.shape)))\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 0
        },
        "id": "XYWn6qbYLKHg",
        "outputId": "813d2ead-25b1-4e3c-c456-cfdebd767609"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "ix = 1027\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 2 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "import matplotlib.pyplot as plt\n",
        "# figure out how many encounters we have\n",
        "numencnt = X_train.shape[0]\n",
        "\n",
        "# choose a random patient encounter to plot\n",
        "ix = random.randint(0,numencnt) #Try a few different index values between 0 and 4999\n",
        "print('ix = {}'.format(ix))\n",
        "\n",
        "# plot a matrix of observation values\n",
        "plt.title(\"Patient Encounter Matrix\")\n",
        "plt.pcolor(np.transpose(X_train[ix,:,:]))\n",
        "plt.ylabel(\"variable\")\n",
        "plt.xlabel(\"time/observation\")\n",
        "plt.ylim(0,22)\n",
        "plt.colorbar()\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s815NaaFJKIB"
      },
      "source": [
        "## LSTM"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9duNmOx7I4Nx",
        "outputId": "ce72e2b6-7292-46d6-fcb8-987103f3f7cb"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model_6\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input (InputLayer)          [(None, 90, 22)]          0         \n",
            "                                                                 \n",
            " input_masked (Masking)      (None, 90, 22)            0         \n",
            "                                                                 \n",
            " lstm1 (LSTM)                (None, 90, 90)            40680     \n",
            "                                                                 \n",
            " output (TimeDistributed)    (None, 90, 1)             91        \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 40771 (159.26 KB)\n",
            "Trainable params: 40771 (159.26 KB)\n",
            "Non-trainable params: 0 (0.00 Byte)\n",
            "_________________________________________________________________\n"
          ]
        }
      ],
      "source": [
        "from tensorflow.keras import Input, Model\n",
        "from tensorflow.keras.layers import LSTM, Dense, TimeDistributed, Masking\n",
        "from tensorflow.keras.optimizers import RMSprop\n",
        "\n",
        "# construct inputs\n",
        "x = Input((90, X_train.shape[-1]) , name='input')\n",
        "mask = Masking(0, name='input_masked')(x)\n",
        "\n",
        "# stack LSTMs\n",
        "lstm_kwargs = {'dropout': 0.25, 'recurrent_dropout': 0.3, 'return_sequences': True}\n",
        "lstm1 = LSTM(90, name='lstm1', **lstm_kwargs)(mask)\n",
        "\n",
        "# output: sigmoid layer\n",
        "output = TimeDistributed(Dense(1, activation='sigmoid'), name='output')(lstm1)\n",
        "model = Model(inputs=x, outputs=output)\n",
        "\n",
        "# compile model\n",
        "optimizer = RMSprop(learning_rate=0.005)\n",
        "model.compile(optimizer=optimizer, loss='binary_crossentropy')\n",
        "\n",
        "# print layer shapes and model parameters\n",
        "model.summary()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DMlUch0OG86v",
        "outputId": "a9a6999a-7aa4-47e4-e57e-d864e630892b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "18/18 [==============================] - 16s 563ms/step - loss: 0.5111 - val_loss: 0.5758\n",
            "Epoch 2/50\n",
            "18/18 [==============================] - 9s 480ms/step - loss: 0.5120 - val_loss: 0.5422\n",
            "Epoch 3/50\n",
            "18/18 [==============================] - 11s 628ms/step - loss: 0.5040 - val_loss: 0.5507\n",
            "Epoch 4/50\n",
            "18/18 [==============================] - 8s 426ms/step - loss: 0.5066 - val_loss: 0.5563\n",
            "Epoch 5/50\n",
            "18/18 [==============================] - 11s 626ms/step - loss: 0.5114 - val_loss: 0.5921\n",
            "Epoch 6/50\n",
            "18/18 [==============================] - 9s 495ms/step - loss: 0.5056 - val_loss: 0.5580\n",
            "Epoch 7/50\n",
            "18/18 [==============================] - 10s 533ms/step - loss: 0.5061 - val_loss: 0.5713\n"
          ]
        }
      ],
      "source": [
        "callback = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=5)\n",
        "history = model.fit(X_train, y_train, batch_size=128, validation_data=(X_valid, y_valid), epochs=50, verbose=1, callbacks=[callback])"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "VSy6n5wdd3EM"
      },
      "outputs": [],
      "source": [
        "from datetime import datetime\n",
        "today_date = datetime.now().strftime('%d-%m-%Y')\n",
        "# Save the model\n",
        "model.save(f'/drive/My Drive/MASTER/SEGUNDO/MAM12A ML AND REASONING FOR HEALTH/MAM12A project/models_LSTM_ana/LSTM_checkpoint_{today_date}')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DHLmeU0SJNFE"
      },
      "source": [
        "## Test + ROC"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "y4fOALXpPjDG"
      },
      "outputs": [],
      "source": [
        "model = tf.keras.models.load_model('/drive/My Drive/MASTER/SEGUNDO/MAM12A ML AND REASONING FOR HEALTH/MAM12A project/models_LSTM_ana/LSTM_checkpoint_12-12-2023')"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NxOVyeU85gH4",
        "outputId": "6cc2d112-d33d-4a76-d008-236b4a889cd5"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "9/9 [==============================] - 1s 32ms/step\n"
          ]
        }
      ],
      "source": [
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 18,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "f_9F4NtjDRZ_",
        "outputId": "e0b7bfc2-74d4-415a-cb6d-2608f4fd42fb"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "281"
            ]
          },
          "execution_count": 18,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "len(preds)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 19,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nl6OXShjdsp8",
        "outputId": "39485486-4c20-4a32-bfca-a0141db1592e"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "(281, 90, 1)"
            ]
          },
          "execution_count": 19,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "preds.shape"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 20,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "e-jZrP5Uck-G",
        "outputId": "1dcc3b11-7657-428b-e055-2252aa93d03b"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "69.0"
            ]
          },
          "execution_count": 20,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "sum(y_test[:, 0, :].squeeze())"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 21,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bCrKs3b7MgP3",
        "outputId": "c064c48a-4262-4a27-f3d5-50458ec85b6b"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "AUC: 0.8349056603773585\n",
            "Accuracy: 0.8185\n",
            "F1 Score: 0.6531\n",
            "Negative Log-Likelihood: 6.5417\n",
            "Precision: 0.6154\n",
            "Recall: 0.6957\n"
          ]
        }
      ],
      "source": [
        "from sklearn.metrics import roc_curve, auc, accuracy_score, f1_score, log_loss, precision_score, recall_score\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 22,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 0
        },
        "id": "vFIpecdvRfX8",
        "outputId": "faf92784-ec41-4b73-c8de-2dc873aa46fe"
      },
      "outputs": [
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 700x500 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "# plot rocs & display AUCs\n",
        "plt.figure(figsize=(7, 5))\n",
        "line_kwargs = {'linewidth': 4, 'alpha': 0.8}\n",
        "plt.plot(rnn_roc[0], rnn_roc[1], label='AUC: %0.3f' % rnn_auc, color='#6AA84F', **line_kwargs)\n",
        "plt.legend(loc='lower right', fontsize=20)\n",
        "plt.xlim((-0.05, 1.05))\n",
        "plt.ylim((-0.05, 1.05))\n",
        "plt.xticks([0, 0.25, 0.5, 0.75, 1.0], fontsize=14)\n",
        "plt.yticks([0, 0.25, 0.5, 0.75, 1.0], fontsize=14)\n",
        "plt.xlabel(\"False Positive Rate\", fontsize=18)\n",
        "plt.ylabel(\"True Positive Rate\", fontsize=18)\n",
        "plt.title(\"In-hospital Mortality Prediction ROC Curve\", fontsize=24)\n",
        "plt.grid(alpha=0.25)\n",
        "plt.tight_layout()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XkfiFOvzIbRb"
      },
      "source": [
        "## Performance by subgroup"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yMrN67cSIj1k"
      },
      "source": [
        "### By gender"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "scmca0u6KgHa"
      },
      "source": [
        " 'steroids',\n",
        " 'antibiotics',\n",
        " 'admissioncount',\n",
        " 'gender',\n",
        " 'age',\n",
        " 'weight',\n",
        " 'sofa_resp',\n",
        " 'sofa_cns',\n",
        " 'sofa_cardio',\n",
        " 'dose_drug',\n",
        " 'Hematocrit',\n",
        " 'Hemoglobine',\n",
        " 'Leucocyten',\n",
        " 'Thrombocyten',\n",
        " 'Bilirubine',\n",
        " 'Kreatinine',\n",
        " 'Lactaat',\n",
        " 'Breathing Rate Spontaneous',\n",
        " 'Sodium',\n",
        " 'Calcium',\n",
        " 'HR',\n",
        " 'Saturation'"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "llqWufzrLNOg"
      },
      "outputs": [],
      "source": [
        "woman_subgroup = test[test['gender']==1]\n",
        "x_test = woman_subgroup[col_x]\n",
        "y_test = woman_subgroup[col_y]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WYjkWuGxO8dG",
        "outputId": "7fadf1c2-0f36-4f7e-9760-42f33acc3fb9"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (109, 90, 22) | y_test shape: (109, 90, 1)\n"
          ]
        }
      ],
      "source": [
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JouLeO-APOi4",
        "outputId": "f0a6e1d1-5fa6-4c7a-efd2-e67aa4517ffb"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "4/4 [==============================] - 0s 27ms/step\n",
            "0.8098223615464996\n",
            "Accuracy: 0.8624\n",
            "F1 Score: 0.6809\n",
            "Negative Log-Likelihood: 4.9601\n",
            "Precision: 0.6400\n",
            "Recall: 0.7273\n"
          ]
        }
      ],
      "source": [
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(rnn_auc)\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hDIk7gnDPfB9"
      },
      "outputs": [],
      "source": [
        "man_subgroup = test[test['gender']==0]\n",
        "x_test = man_subgroup[col_x]\n",
        "y_test = man_subgroup[col_y]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SzfJ1glNPiyB",
        "outputId": "31b635bd-ef2b-4ada-dcbd-a49037d21dfd"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (172, 90, 22) | y_test shape: (172, 90, 1)\n"
          ]
        }
      ],
      "source": [
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9MxwYdcUPkb1",
        "outputId": "a22eb2cf-b6e6-4099-a69b-abeb6824e8ce"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "6/6 [==============================] - 0s 27ms/step\n",
            "0.8471489361702128\n",
            "Accuracy: 0.8198\n",
            "F1 Score: 0.6737\n",
            "Negative Log-Likelihood: 6.4962\n",
            "Precision: 0.6667\n",
            "Recall: 0.6809\n"
          ]
        }
      ],
      "source": [
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(rnn_auc)\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ehU4EPLmIliZ"
      },
      "source": [
        "### By age group"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 23,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5cZajSfIQwAJ",
        "outputId": "2aec0a4b-c361-4814-f464-8e47f59db139"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (30, 90, 22) | y_test shape: (30, 90, 1)\n",
            "1/1 [==============================] - 0s 45ms/step\n",
            "AUC: 0.9821428571428572\n",
            "Accuracy: 0.9667\n",
            "F1 Score: 0.8000\n",
            "Negative Log-Likelihood: 1.2015\n",
            "Precision: 0.6667\n",
            "Recall: 1.0000\n"
          ]
        }
      ],
      "source": [
        "young_subgroup = test[test['age']<25]\n",
        "x_test = young_subgroup[col_x]\n",
        "y_test = young_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 24,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "8jm_BIawQ7-t",
        "outputId": "83614439-8f4f-45dc-a797-2a82ab45b181"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (143, 90, 22) | y_test shape: (143, 90, 1)\n",
            "5/5 [==============================] - 0s 28ms/step\n",
            "AUC: 0.8037267080745343\n",
            "Accuracy: 0.8811\n",
            "F1 Score: 0.6792\n",
            "Negative Log-Likelihood: 4.2849\n",
            "Precision: 0.7200\n",
            "Recall: 0.6429\n"
          ]
        }
      ],
      "source": [
        "middleage_subgroup = test[(test['age']>=25) & (test['age']<=60)]\n",
        "x_test = middleage_subgroup[col_x]\n",
        "y_test = middleage_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 25,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "KP4FyDZQRGbb",
        "outputId": "4021e5e3-9649-4a50-bb85-59bd3b81697a"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (108, 90, 22) | y_test shape: (108, 90, 1)\n",
            "4/4 [==============================] - 0s 27ms/step\n",
            "AUC: 0.8063916759568934\n",
            "Accuracy: 0.8056\n",
            "F1 Score: 0.6557\n",
            "Negative Log-Likelihood: 7.0085\n",
            "Precision: 0.9091\n",
            "Recall: 0.5128\n"
          ]
        }
      ],
      "source": [
        "old_subgroup = test[test['age']>60]\n",
        "x_test = old_subgroup[col_x]\n",
        "y_test = old_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nXu9JZjYKnD8"
      },
      "source": [
        "### By admissioncount"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "HA5V2YoBUkMT",
        "outputId": "d8289827-a273-44bb-b63f-6cff14dd2dde"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (247, 90, 22) | y_test shape: (247, 90, 1)\n",
            "8/8 [==============================] - 0s 30ms/step\n",
            "AUC: 0.8387096774193548\n",
            "Accuracy: 0.8097\n",
            "F1 Score: 0.6412\n",
            "Negative Log-Likelihood: 6.8585\n",
            "Precision: 0.6000\n",
            "Recall: 0.6885\n"
          ]
        }
      ],
      "source": [
        "singleadmission_subgroup = test[test['admissioncount']==1]\n",
        "x_test = singleadmission_subgroup[col_x]\n",
        "y_test = singleadmission_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tn1W1aOXUqY2",
        "outputId": "9559717f-1e72-4373-9340-f0716e84d3c0"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (34, 90, 22) | y_test shape: (34, 90, 1)\n",
            "2/2 [==============================] - 0s 20ms/step\n",
            "AUC: 0.7980769230769231\n",
            "Accuracy: 0.8824\n",
            "F1 Score: 0.7500\n",
            "Negative Log-Likelihood: 4.2404\n",
            "Precision: 0.7500\n",
            "Recall: 0.7500\n"
          ]
        }
      ],
      "source": [
        "multiple_admission_subgroup = test[test['admissioncount']>1]\n",
        "x_test = multiple_admission_subgroup[col_x]\n",
        "y_test = multiple_admission_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XNeQ4X5xInnP"
      },
      "source": [
        "### By steroid medication"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 31,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "prtoDarCU7FH",
        "outputId": "fda12dc7-73b2-4fd2-e971-592665a7c1fe"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (111, 90, 22) | y_test shape: (111, 90, 1)\n",
            "4/4 [==============================] - 1s 120ms/step\n",
            "AUC: 0.8738030095759234\n",
            "Accuracy: 0.8559\n",
            "F1 Score: 0.7949\n",
            "Negative Log-Likelihood: 5.1955\n",
            "Precision: 0.8857\n",
            "Recall: 0.7209\n"
          ]
        }
      ],
      "source": [
        "admission_with_steroid = df[df['steroids'] == 1]['admissionid'].unique()\n",
        "steroid_subgroup = test[test['admissionid'].isin(admission_with_steroid)]\n",
        "\n",
        "x_test = steroid_subgroup[col_x]\n",
        "y_test = steroid_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 32,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RxbfKYA2VDac",
        "outputId": "659deecf-ec6d-42af-9c1f-8a843cacccbc"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (170, 90, 22) | y_test shape: (170, 90, 1)\n",
            "6/6 [==============================] - 1s 96ms/step\n",
            "AUC: 0.766826923076923\n",
            "Accuracy: 0.8353\n",
            "F1 Score: 0.5000\n",
            "Negative Log-Likelihood: 5.9366\n",
            "Precision: 0.4667\n",
            "Recall: 0.5385\n"
          ]
        }
      ],
      "source": [
        "admission_with_steroid = df[df['steroids'] == 1]['admissionid'].unique()\n",
        "no_steroid_subgroup = test[~test['admissionid'].isin(admission_with_steroid)]\n",
        "\n",
        "x_test = no_steroid_subgroup[col_x]\n",
        "y_test = no_steroid_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1WTD0gt-IuDb"
      },
      "source": [
        "### By quick SOFA at admission"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 35,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "AvuYuqz8Ii__",
        "outputId": "64d0d9e3-623a-44b3-8819-8c9d01f5ec8f"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (260, 90, 22) | y_test shape: (260, 90, 1)\n",
            "9/9 [==============================] - 1s 103ms/step\n",
            "AUC: 0.6538584183673469\n",
            "Accuracy: 0.5962\n",
            "F1 Score: 0.4560\n",
            "Negative Log-Likelihood: 14.5561\n",
            "Precision: 0.3411\n",
            "Recall: 0.6875\n"
          ]
        }
      ],
      "source": [
        "high_sofa_subgroup = test[((test['sofa_resp']>=1)|(test['sofa_cns']>=1)|(test['sofa_cardio']>=1))&(test['time']==0)]\n",
        "x_test = high_sofa_subgroup[col_x]\n",
        "y_test = high_sofa_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 36,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ToiPE-F2YGRC",
        "outputId": "9dfc395a-35b0-427a-a620-85606c8f74d0"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "X_test shape: (21, 90, 22) | y_test shape: (21, 90, 1)\n",
            "1/1 [==============================] - 0s 45ms/step\n",
            "AUC: 0.6125\n",
            "Accuracy: 0.4762\n",
            "F1 Score: 0.4762\n",
            "Negative Log-Likelihood: 18.8800\n",
            "Precision: 0.3125\n",
            "Recall: 1.0000\n"
          ]
        }
      ],
      "source": [
        "low_sofa_subgroup = test[((test['sofa_resp']==0)&(test['sofa_cns']==0)&(test['sofa_cardio']==0))&(test['time']==0)]\n",
        "x_test = low_sofa_subgroup[col_x]\n",
        "y_test = low_sofa_subgroup[col_y]\n",
        "x_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "y_test.set_index(['admissionid','time'], append=True, inplace=True)\n",
        "\n",
        "teeId = x_test.index.levels[1]\n",
        "\n",
        "X_test = [x_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "y_test = [y_test.loc[(slice(None), patient), :].values for patient in teeId]\n",
        "\n",
        "X_test = sequence.pad_sequences(X_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "y_test = sequence.pad_sequences(y_test, dtype='float32', maxlen=maxlen, padding='post', truncating='post')\n",
        "print(\"X_test shape: %s | y_test shape: %s\" % (str(X_test.shape), str(y_test.shape)))\n",
        "# Generate model results on the holdout validation set\n",
        "preds = model.predict(X_test)\n",
        "# get 0/1 binary label for each patient encounter\n",
        "label = y_test[:, 0, :].squeeze()\n",
        "\n",
        "# get the last prediction in [0,1] for the patient\n",
        "prediction = preds[:, -1, :].squeeze()\n",
        "\n",
        "# compute ROC curve for predictions\n",
        "rnn_roc = roc_curve(label,prediction)\n",
        "\n",
        "# compute the area under the curve of prediction ROC\n",
        "rnn_auc = auc(rnn_roc[0], rnn_roc[1])\n",
        "\n",
        "print(f\"AUC: {rnn_auc}\")\n",
        "thresholds = rnn_roc[2]\n",
        "# Finding best threshold using Youden's J statistic\n",
        "optimal_idx = np.argmax(rnn_roc[1]-rnn_roc[0])\n",
        "optimal_threshold = thresholds[optimal_idx]\n",
        "\n",
        "accuracy = accuracy_score(label, prediction>=optimal_threshold)\n",
        "f1 = f1_score(label, prediction>=optimal_threshold)\n",
        "nll = log_loss(label, prediction>=optimal_threshold)\n",
        "precision = precision_score(label, prediction>=optimal_threshold)\n",
        "recall = recall_score(label, prediction>=optimal_threshold)\n",
        "\n",
        "print(f\"Accuracy: {accuracy:.4f}\")\n",
        "print(f\"F1 Score: {f1:.4f}\")\n",
        "print(f\"Negative Log-Likelihood: {nll:.4f}\")\n",
        "print(f\"Precision: {precision:.4f}\")\n",
        "print(f\"Recall: {recall:.4f}\")\n"
      ]
    }
  ],
  "metadata": {
    "colab": {
      "collapsed_sections": [
        "wkXGVM33tJaP",
        "qyaSqLcHDsnA",
        "DHLmeU0SJNFE"
      ],
      "provenance": [],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.9.12"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}