notebook-wids2021.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Women in Data Science(WiDS) Datathon 2021\n",
    "\n",
    "\n",
    "### Overview\n",
    " The WiDS Datathon 2021 focuses on patient health, with an emphasis on the chronic condition of diabetes. Intensive Care Units (ICUs) often lack verified medical histories for incoming patients. A patient in distress or a patient who is brought in confused or unresponsive may not be able to provide information about chronic conditions such as heart disease, injuries, or diabetes. Medical records may take days to transfer, especially for a patient from another medical provider or system.\n",
    "\n",
    "\n",
    "### Objective\n",
    " The WiDS Datathon 2021 will focus on models to determine whether a patient admitted to an ICU has been diagnosed with a particular type of diabetes, Diabetes Mellitus. Using data from the first 24 hours of intensive care, participants will explore labeled training data for model development. \n",
    "\n",
    "### Data description\n",
    "* TrainingWiDS2021.csv - the training data. You should see 130,157 encounters represented here. Please view the Data Dictionary file for more information about the columns.\n",
    "* UnlabeledWiDS2021.csv - the unlabeled data (data without diabetes_mellitus provided). You are being asked to predict the diabetes_mellitus variable for these encounters.\n",
    "* SampleSubmissionWiDS2021.csv - a sample submission file in the correct format.\n",
    "* SolutionTemplateWiDS2021.csv - a list of all the rows (and encounters) that should be in your submissions.\n",
    "* DataDictionaryWiDS2021.csv - supplemental information about the data.\n",
    "* [Files can be downloaded from the kaggle website](https://www.kaggle.com/c/widsdatathon2021/data).\n",
    "\n",
    "\n",
    "### Contents of this notebook\n",
    "* EDA + Binary classification (model used: LightGBM)\n",
    "* Some portion of the feature engineering was inspired by other participants' notebooks published [here](https://www.kaggle.com/c/widsdatathon2021/code).\n",
    "\n",
    "### Final score at the leaderboard: 0.86744 (Top 22%)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style type='text/css'>\n",
       ".datatable table.frame { margin-bottom: 0; }\n",
       ".datatable table.frame thead { border-bottom: none; }\n",
       ".datatable table.frame tr.coltypes td {  color: #FFFFFF;  line-height: 6px;  padding: 0 0.5em;}\n",
       ".datatable .bool    { background: #DDDD99; }\n",
       ".datatable .object  { background: #565656; }\n",
       ".datatable .int     { background: #5D9E5D; }\n",
       ".datatable .float   { background: #4040CC; }\n",
       ".datatable .str     { background: #CC4040; }\n",
       ".datatable .row_index {  background: var(--jp-border-color3);  border-right: 1px solid var(--jp-border-color0);  color: var(--jp-ui-font-color3);  font-size: 9px;}\n",
       ".datatable .frame tr.coltypes .row_index {  background: var(--jp-border-color0);}\n",
       ".datatable th:nth-child(2) { padding-left: 12px; }\n",
       ".datatable .hellipsis {  color: var(--jp-cell-editor-border-color);}\n",
       ".datatable .vellipsis {  background: var(--jp-layout-color0);  color: var(--jp-cell-editor-border-color);}\n",
       ".datatable .na {  color: var(--jp-cell-editor-border-color);  font-size: 80%;}\n",
       ".datatable .footer { font-size: 9px; }\n",
       ".datatable .frame_dimensions {  background: var(--jp-border-color3);  border-top: 1px solid var(--jp-border-color0);  color: var(--jp-ui-font-color3);  display: inline-block;  opacity: 0.6;  padding: 1px 10px 1px 5px;}\n",
       "</style>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Import libraries\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import math\n",
    "from scipy import stats \n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "sns.set_palette('Set3')\n",
    "sns.set_context(rc={'axes.labelsize':18,\n",
    "                    'axes.titlesize':18,\n",
    "                    'font.size':15,\n",
    "                    'legend.fontsize':15,\n",
    "                    'lines.linewidth':2.2})\n",
    "from ipywidgets import interact   # interactive plot\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\") # ignore plot warning messages\n",
    "\n",
    "\n",
    "# Modeling\n",
    "from sklearn.preprocessing import OrdinalEncoder\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "import lightgbm as lgb\n",
    "from sklearn.metrics import classification_report, roc_auc_score, plot_roc_curve, plot_confusion_matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 180)\n",
      "(10234, 179)\n"
     ]
    }
   ],
   "source": [
    "# Import data\n",
    "train = pd.read_csv(\"../input/trainwids2021/TrainingWiDS2021.csv\", index_col=0)\n",
    "test = pd.read_csv(\"../input/testwids2021/UnlabeledWiDS2021.csv\", index_col=0)\n",
    "\n",
    "print(train.shape)\n",
    "print(test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Target distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:xlabel='diabetes_mellitus', ylabel='count'>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Count targets\n",
    "sns.countplot(train['diabetes_mellitus'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* Unbalanced binary target distribution"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Training data quality check"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Missing rows?\n",
    "prop_isnull = train.isnull().mean()\n",
    "prop_isnull = prop_isnull[prop_isnull > 0]\n",
    "plt.title(\"Distribution of missing data\")\n",
    "plt.ylabel(\"# of columns\")\n",
    "plt.xlabel(\"% missing values\")\n",
    "prop_isnull.hist()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<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>encounter_id</th>\n",
       "      <th>hospital_id</th>\n",
       "      <th>age</th>\n",
       "      <th>bmi</th>\n",
       "      <th>elective_surgery</th>\n",
       "      <th>ethnicity</th>\n",
       "      <th>gender</th>\n",
       "      <th>height</th>\n",
       "      <th>hospital_admit_source</th>\n",
       "      <th>icu_admit_source</th>\n",
       "      <th>...</th>\n",
       "      <th>h1_pao2fio2ratio_max</th>\n",
       "      <th>h1_pao2fio2ratio_min</th>\n",
       "      <th>aids</th>\n",
       "      <th>cirrhosis</th>\n",
       "      <th>hepatic_failure</th>\n",
       "      <th>immunosuppression</th>\n",
       "      <th>leukemia</th>\n",
       "      <th>lymphoma</th>\n",
       "      <th>solid_tumor_with_metastasis</th>\n",
       "      <th>diabetes_mellitus</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>0 rows × 180 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "Empty DataFrame\n",
       "Columns: [encounter_id, hospital_id, age, bmi, elective_surgery, ethnicity, gender, height, hospital_admit_source, icu_admit_source, icu_id, icu_stay_type, icu_type, pre_icu_los_days, readmission_status, weight, albumin_apache, apache_2_diagnosis, apache_3j_diagnosis, apache_post_operative, arf_apache, bilirubin_apache, bun_apache, creatinine_apache, fio2_apache, gcs_eyes_apache, gcs_motor_apache, gcs_unable_apache, gcs_verbal_apache, glucose_apache, heart_rate_apache, hematocrit_apache, intubated_apache, map_apache, paco2_apache, paco2_for_ph_apache, pao2_apache, ph_apache, resprate_apache, sodium_apache, temp_apache, urineoutput_apache, ventilated_apache, wbc_apache, d1_diasbp_invasive_max, d1_diasbp_invasive_min, d1_diasbp_max, d1_diasbp_min, d1_diasbp_noninvasive_max, d1_diasbp_noninvasive_min, d1_heartrate_max, d1_heartrate_min, d1_mbp_invasive_max, d1_mbp_invasive_min, d1_mbp_max, d1_mbp_min, d1_mbp_noninvasive_max, d1_mbp_noninvasive_min, d1_resprate_max, d1_resprate_min, d1_spo2_max, d1_spo2_min, d1_sysbp_invasive_max, d1_sysbp_invasive_min, d1_sysbp_max, d1_sysbp_min, d1_sysbp_noninvasive_max, d1_sysbp_noninvasive_min, d1_temp_max, d1_temp_min, h1_diasbp_invasive_max, h1_diasbp_invasive_min, h1_diasbp_max, h1_diasbp_min, h1_diasbp_noninvasive_max, h1_diasbp_noninvasive_min, h1_heartrate_max, h1_heartrate_min, h1_mbp_invasive_max, h1_mbp_invasive_min, h1_mbp_max, h1_mbp_min, h1_mbp_noninvasive_max, h1_mbp_noninvasive_min, h1_resprate_max, h1_resprate_min, h1_spo2_max, h1_spo2_min, h1_sysbp_invasive_max, h1_sysbp_invasive_min, h1_sysbp_max, h1_sysbp_min, h1_sysbp_noninvasive_max, h1_sysbp_noninvasive_min, h1_temp_max, h1_temp_min, d1_albumin_max, d1_albumin_min, d1_bilirubin_max, d1_bilirubin_min, ...]\n",
       "Index: []\n",
       "\n",
       "[0 rows x 180 columns]"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Duplicated rows?\n",
    "train[train.duplicated() == True]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* 160 out of 180 columns have missing information, number of missing records ranging between 3.8 - 87.1%\n",
    "* No duplicated rows"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Distribution of Variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([dtype('int64'), dtype('float64'), dtype('O')], dtype=object)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Data types\n",
    "train.dtypes.unique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 13)\n",
      "(130157, 6)\n",
      "(130157, 161)\n"
     ]
    }
   ],
   "source": [
    "# Separate data by binary, categorical, continous types\n",
    "\n",
    "# Binary\n",
    "two_unique_values = (train.nunique().values == 2) # find columns; n of unique value equals 2\n",
    "binary = train.loc[:,two_unique_values] # select columns\n",
    "binary = binary.loc[:, ~binary.isna().any()] # remove columns when 1 of the unique value is null\n",
    "\n",
    "# Categorical & continuous\n",
    "non_binaries = [col for col in train.columns if col not in binary.columns]\n",
    "train_non_bi = train[non_binaries]\n",
    "categoricals = train_non_bi.select_dtypes(include=[\"object\"])\n",
    "continuous = train_non_bi.select_dtypes(exclude=[\"object\"])\n",
    "\n",
    "print(binary.shape)\n",
    "print(categoricals.shape)\n",
    "print(continuous.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2-1. Binary variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "elective_surgery               0.0\n",
       "apache_post_operative          0.0\n",
       "arf_apache                     0.0\n",
       "intubated_apache               0.0\n",
       "ventilated_apache              0.0\n",
       "aids                           0.0\n",
       "cirrhosis                      0.0\n",
       "hepatic_failure                0.0\n",
       "immunosuppression              0.0\n",
       "leukemia                       0.0\n",
       "lymphoma                       0.0\n",
       "solid_tumor_with_metastasis    0.0\n",
       "diabetes_mellitus              0.0\n",
       "dtype: float64"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Missing data?\n",
    "binary.isnull().mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* No missing rows"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(130157, 12)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Let's drop diabetes_mellitus from binary dataframe to separate features from their target\n",
    "binary.drop('diabetes_mellitus', axis=1, inplace=True)\n",
    "binary.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "0cdc82bcee494a9b889e557d3807c001",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='cols', options=('elective_surgery', 'apache_post_operative', 'arf_…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Interactive plot to check distribution of binary variables\n",
    "@interact(cols=binary.columns)\n",
    "def get_countplot_binary(cols):\n",
    "    plt.figure(figsize=(12,6))\n",
    "    plt.title(f'Distribution of {cols}')\n",
    "    sns.countplot(train[cols])\n",
    "    \n",
    "    # Print normalized counts of each variable\n",
    "    print(binary[cols].value_counts(normalize=True).round(2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9f0037272c2644cd9c1fcdc2994a3162",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='cols', options=('elective_surgery', 'apache_post_operative', 'arf_…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Interactive plot to check relationships\n",
    "# between the binary variables and the target variable\n",
    "@interact(cols=binary.columns)\n",
    "def get_countplot_diabetes(cols):\n",
    "    plt.figure(figsize=(12,6))\n",
    "    sns.countplot(x=cols, hue='diabetes_mellitus', data=train)\n",
    "    plt.title(f'Diabetes mellitus (0, 1) and {cols}')\n",
    "    plt.legend(bbox_to_anchor=(1.15, 1))\n",
    "    \n",
    "    # Print normalized counts of each variable\n",
    "    print(train.groupby('diabetes_mellitus')[cols].value_counts(normalize=True).round(2))\n",
    "    \n",
    "    # This part is for categorical vaviables below\n",
    "    # Rotate the xlabels if more than 4 values available\n",
    "    if train[cols].nunique() > 4:\n",
    "         plt.xticks(rotation=40)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Highlights of binary variables\n",
    "\n",
    "**arf_apache**\n",
    "* When there was no diabetes: positive 2 %\n",
    "* When there was diabetes: positive 6 %\n",
    "\n",
    "<br>\n",
    "* APACHE: Acute Physiology and Chronic Health Evaluation II score to predict hospital mortality"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2-2. Categorical variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ethnicity                0.012193\n",
       "gender                   0.000507\n",
       "hospital_admit_source    0.255061\n",
       "icu_admit_source         0.001844\n",
       "icu_stay_type            0.000000\n",
       "icu_type                 0.000000\n",
       "dtype: float64"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Missing data per each column\n",
    "categoricals.isnull().mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "dfab4975a54241549774ddf13810881f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='cols', options=('ethnicity', 'gender', 'hospital_admit_source', 'i…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Interactive plot to check distribution of categorical variables\n",
    "@interact(cols=categoricals.columns)\n",
    "def get_countplot_cat(cols):\n",
    "    plt.figure(figsize=(12,6))\n",
    "    plt.title(f'Distribution of {cols}')\n",
    "    sns.countplot(train[cols])\n",
    "    \n",
    "    # Rotate the xlabels if more than 4 values available\n",
    "    if categoricals[cols].nunique() > 4:\n",
    "         plt.xticks(rotation=40)\n",
    "            \n",
    "    # Print normalized counts of each variable\n",
    "    print(categoricals[cols].value_counts(normalize=True).round(2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Highlights of categorical data\n",
    "\n",
    "**Ethnity**\n",
    "* Highly unbalanced ratial distribution: Majority of the data (~78%) is from Caucasians.\n",
    "\n",
    "**Gender**\n",
    "* Slightly more male patients(54%) than the females (46%)\n",
    "\n",
    "**Hospital admit source**\n",
    "* More than 52% from Emergercy Department. (25% of the patients lack this information)\n",
    "\n",
    "**ICU admit source**\n",
    "* Over 59% of patients admitted to ICU after Accident & Emergency\n",
    "\n",
    "**ICU stay type**\n",
    "* Most(95%) of the patients stay admitted to the unit, ~4% transferred to other locations, <1% readmitted\n",
    "\n",
    "**ICU type**\n",
    "* ~55% of the patients in Med-Surg ICU (=Medical Surgical Intensive Care Unit)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "09445f4db36f471e9b17c26de2a05a59",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='cols', options=('ethnicity', 'gender', 'hospital_admit_source', 'i…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<function __main__.get_countplot_diabetes(cols)>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Interactive plot to check relationships\n",
    "# between the categorical variables and the target variable\n",
    "interact(get_countplot_diabetes, cols=categoricals.columns)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Ethnity**\n",
    "* Some ethnicity's (e.g. African American) proportion within the two groups differ.\n",
    "\n",
    "**Gender**\n",
    "* Similar gender distribution of patients with/without diabetes mellitus.\n",
    "\n",
    "**Hospital admit source, ICU stay type**\n",
    "* Similar to ethnity, roughly 20-30 % of each group had diabetes mellitus\n",
    "\n",
    "**icu_admit_source and icu_type**\n",
    "* Slight changes in order of frequency when compared negative vs. positive to diabetes mellitus"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2-3. Continuous variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "eae46743c6974736956c4d896b28a5e1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='cols', options=('encounter_id', 'hospital_id', 'age', 'bmi', 'heig…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Histograms to show distribution of continuous values\n",
    "@interact(cols=continuous.columns)\n",
    "def get_histplot(cols):\n",
    "    plt.figure(figsize=(12,6))\n",
    "    plt.title(f'Distribution of {cols}')\n",
    "    sns.histplot(continuous[cols])\n",
    "    \n",
    "    plt.axvline(continuous[cols].median(), color='red')  # Median in red\n",
    "    plt.axvline(continuous[cols].mean(), color='green')  # Mean in green\n",
    "    \n",
    "    print('Mean: ', round(continuous[cols].mean(), 1))\n",
    "    print('Median: ', round(continuous[cols].median(), 1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b2482d72b4df471db0190fa88ffcc0ed",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='cols', options=('age', 'bmi', 'height', 'pre_icu_los_days', 'readm…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Excude ids from plot\n",
    "except_ids = [col for col in continuous.columns if col[-2:] != 'id']\n",
    "\n",
    "# Interactive relplot between continuous variables and the target variable\n",
    "@interact(cols=except_ids + ['diabetes_mellitus'])\n",
    "def get_boxplot(cols):\n",
    "    plt.figure(figsize=(12,6))\n",
    "    sns.boxplot(x=train['diabetes_mellitus'], y=train[cols])\n",
    "    plt.title(f'Diabetes mellitus (0, 1) and {cols}')\n",
    "    \n",
    "    # Print normalized counts of each variable\n",
    "    count_diabetes = train.groupby('diabetes_mellitus')[cols].count()\n",
    "    print('Total counts: ', count_diabetes.sum())\n",
    "    print(' ')\n",
    "    print(count_diabetes)\n",
    "    print(' ')\n",
    "    \n",
    "    # Print mean values\n",
    "    diabetes = train['diabetes_mellitus'] == 1\n",
    "    positive = train[diabetes][cols]\n",
    "    negative = train[~diabetes][cols]\n",
    "    print('No diabetes meallitus(0)', negative.mean())\n",
    "    print('Diabetes meallitus(1)', positive.mean())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "133\n",
      "['age', 'bmi', 'icu_id', 'pre_icu_los_days', 'weight', 'albumin_apache', 'apache_3j_diagnosis', 'bilirubin_apache', 'bun_apache', 'creatinine_apache', 'fio2_apache', 'gcs_eyes_apache', 'gcs_motor_apache', 'gcs_unable_apache', 'gcs_verbal_apache']\n"
     ]
    }
   ],
   "source": [
    "# Calculate ttest and print only the columns with significant difference\n",
    "diabetes = train['diabetes_mellitus'] == 1\n",
    "significant_cols = []\n",
    "\n",
    "for col in continuous.columns:\n",
    "    positive = continuous[diabetes][col]\n",
    "    negative = continuous[~diabetes][col]\n",
    "    tstats , p_value = stats.ttest_ind(negative.values, positive.values, nan_policy='omit')\n",
    "\n",
    "    if p_value < 0.05:\n",
    "        significant_cols.append(col)\n",
    "        \n",
    "print(len(significant_cols))\n",
    "print(significant_cols[:15])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Highlight of continous data\n",
    "\n",
    "**Statistically significant difference between groups**\n",
    "* 133 out of 161 columns including age, bmi, pre_icu_los_days, weight."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Data preprocessing + Feature engineering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 180)\n",
      "(10234, 179)\n"
     ]
    }
   ],
   "source": [
    "train_new = train.copy()\n",
    "test_new = test.copy()\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 179)\n",
      "(10234, 178)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<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>hospital_id</th>\n",
       "      <th>age</th>\n",
       "      <th>bmi</th>\n",
       "      <th>elective_surgery</th>\n",
       "      <th>ethnicity</th>\n",
       "      <th>gender</th>\n",
       "      <th>height</th>\n",
       "      <th>hospital_admit_source</th>\n",
       "      <th>icu_admit_source</th>\n",
       "      <th>icu_id</th>\n",
       "      <th>...</th>\n",
       "      <th>h1_pao2fio2ratio_max</th>\n",
       "      <th>h1_pao2fio2ratio_min</th>\n",
       "      <th>aids</th>\n",
       "      <th>cirrhosis</th>\n",
       "      <th>hepatic_failure</th>\n",
       "      <th>immunosuppression</th>\n",
       "      <th>leukemia</th>\n",
       "      <th>lymphoma</th>\n",
       "      <th>solid_tumor_with_metastasis</th>\n",
       "      <th>diabetes_mellitus</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>encounter_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>147000</th>\n",
       "      <td>163</td>\n",
       "      <td>73.0</td>\n",
       "      <td>28.727377</td>\n",
       "      <td>1</td>\n",
       "      <td>Caucasian</td>\n",
       "      <td>M</td>\n",
       "      <td>182.90</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Operating Room / Recovery</td>\n",
       "      <td>1082</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>147001</th>\n",
       "      <td>159</td>\n",
       "      <td>NaN</td>\n",
       "      <td>24.925526</td>\n",
       "      <td>0</td>\n",
       "      <td>Other/Unknown</td>\n",
       "      <td>M</td>\n",
       "      <td>162.60</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Accident &amp; Emergency</td>\n",
       "      <td>709</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>147002</th>\n",
       "      <td>40</td>\n",
       "      <td>43.0</td>\n",
       "      <td>25.118065</td>\n",
       "      <td>0</td>\n",
       "      <td>Caucasian</td>\n",
       "      <td>M</td>\n",
       "      <td>175.20</td>\n",
       "      <td>Emergency Department</td>\n",
       "      <td>Accident &amp; Emergency</td>\n",
       "      <td>636</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>147003</th>\n",
       "      <td>202</td>\n",
       "      <td>35.0</td>\n",
       "      <td>24.838934</td>\n",
       "      <td>0</td>\n",
       "      <td>Caucasian</td>\n",
       "      <td>M</td>\n",
       "      <td>172.72</td>\n",
       "      <td>Emergency Department</td>\n",
       "      <td>Accident &amp; Emergency</td>\n",
       "      <td>427</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>147004</th>\n",
       "      <td>171</td>\n",
       "      <td>59.0</td>\n",
       "      <td>24.389348</td>\n",
       "      <td>1</td>\n",
       "      <td>Other/Unknown</td>\n",
       "      <td>F</td>\n",
       "      <td>165.00</td>\n",
       "      <td>Operating Room</td>\n",
       "      <td>Operating Room / Recovery</td>\n",
       "      <td>404</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 179 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              hospital_id   age        bmi  elective_surgery      ethnicity  \\\n",
       "encounter_id                                                                  \n",
       "147000                163  73.0  28.727377                 1      Caucasian   \n",
       "147001                159   NaN  24.925526                 0  Other/Unknown   \n",
       "147002                 40  43.0  25.118065                 0      Caucasian   \n",
       "147003                202  35.0  24.838934                 0      Caucasian   \n",
       "147004                171  59.0  24.389348                 1  Other/Unknown   \n",
       "\n",
       "             gender  height hospital_admit_source           icu_admit_source  \\\n",
       "encounter_id                                                                   \n",
       "147000            M  182.90                   NaN  Operating Room / Recovery   \n",
       "147001            M  162.60                   NaN       Accident & Emergency   \n",
       "147002            M  175.20  Emergency Department       Accident & Emergency   \n",
       "147003            M  172.72  Emergency Department       Accident & Emergency   \n",
       "147004            F  165.00        Operating Room  Operating Room / Recovery   \n",
       "\n",
       "              icu_id  ... h1_pao2fio2ratio_max h1_pao2fio2ratio_min  aids  \\\n",
       "encounter_id          ...                                                   \n",
       "147000          1082  ...                  NaN                  NaN     0   \n",
       "147001           709  ...                  NaN                  NaN     0   \n",
       "147002           636  ...                  NaN                  NaN     0   \n",
       "147003           427  ...                  NaN                  NaN     0   \n",
       "147004           404  ...                  NaN                  NaN     0   \n",
       "\n",
       "              cirrhosis  hepatic_failure  immunosuppression  leukemia  \\\n",
       "encounter_id                                                            \n",
       "147000                0                0                  0         0   \n",
       "147001                0                0                  0         0   \n",
       "147002                0                0                  0         0   \n",
       "147003                0                0                  0         0   \n",
       "147004                0                0                  0         0   \n",
       "\n",
       "              lymphoma  solid_tumor_with_metastasis  diabetes_mellitus  \n",
       "encounter_id                                                            \n",
       "147000               0                            0                  0  \n",
       "147001               0                            0                  0  \n",
       "147002               0                            0                  0  \n",
       "147003               0                            0                  0  \n",
       "147004               0                            1                  0  \n",
       "\n",
       "[5 rows x 179 columns]"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Sort by encounter_id and set it as an index\n",
    "train_new = train_new.sort_values('encounter_id')\n",
    "test_new = test_new.sort_values('encounter_id')\n",
    "\n",
    "train_new.set_index('encounter_id', inplace=True)\n",
    "test_new.set_index('encounter_id', inplace=True)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)\n",
    "train_new.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "817d53f48fd54f199dd21e7bde759fcf",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "interactive(children=(Dropdown(description='col', options=('hospital_id', 'age', 'bmi', 'elective_surgery', 'e…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "@interact(col=train_new.columns)\n",
    "def another_plot(col):\n",
    "    plt.figure(figsize=(14,6))\n",
    "    if col in continuous.columns:\n",
    "        train_new.groupby('diabetes_mellitus')[col].hist()\n",
    "    else:\n",
    "        sns.countplot(x=col, hue='diabetes_mellitus', data=train_new)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* Type 2 diabetes is an adult-onset metabolic disease linked with excessive body weight, insufficient exercise\n",
    "    * related columns: age, height-weight, bmi\n",
    "    * ethnity (Excess body fat is associated with 30% of cases in Chinese and Japanese descent, 60–80% of cases in those of European and African descent, and 100% of Pima Indians and Pacific Islanders)\n",
    "\n",
    "<br>\n",
    "\n",
    "* High blood sugar level -> 'Hyperglycemia': an excessive amount of glucose circulates in the blood plasma.\n",
    "    * glucose-apache, d1_glucose_max, d1_glucose_min"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 183)\n",
      "(10234, 182)\n"
     ]
    }
   ],
   "source": [
    "# age >= 18 -> adult?\n",
    "# bmi >= 30 -> obesity\n",
    "\n",
    "def flag_obese_adult(df):\n",
    "    df['obese_adult'] = pd.Series()\n",
    "    adult = df['age'] >= 18\n",
    "    obese = df['bmi'] >= 30\n",
    "    df['obese_adult'] = df[adult & obese]['obese_adult'].apply(lambda x: 1)\n",
    "    df['obese_adult'].replace(np.nan, 0, inplace=True)\n",
    "    df['obese_adult_diagnosis'] = df['obese_adult'] * df['apache_3j_diagnosis'] + df['obese_adult'] * df['apache_2_diagnosis']\n",
    "\n",
    "flag_obese_adult(train_new)\n",
    "flag_obese_adult(test_new)\n",
    "\n",
    "def metrics_obesity(df):\n",
    "    df['obese_age'] = df['bmi'] / df['age']\n",
    "    df['height_weight_ratio'] = df['weight'] / df['height']\n",
    "\n",
    "metrics_obesity(train_new)\n",
    "metrics_obesity(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 471)\n",
      "(10234, 470)\n"
     ]
    }
   ],
   "source": [
    "def after_first_hour(df):\n",
    "    lab_col = [c for c in df.columns if((c.startswith(\"h1\")) | (c.startswith(\"d1\")))]\n",
    "    lab_col_names = list(set(list(map(lambda i: i[ 3 : -4], lab_col))))\n",
    "    \n",
    "    first_h = []\n",
    "    for v in lab_col_names:\n",
    "        first_h.append(v+\"_started_after_firstHour\")\n",
    "        df[v+\"_d1_value_range\"] = df[f\"d1_{v}_max\"].subtract(df[f\"d1_{v}_min\"])\n",
    "        df[v+\"_d1_value_avg\"] = df[[f\"d1_{v}_max\", f\"d1_{v}_min\"]].mean(axis=1)\n",
    "        df[v+\"_d1_value_ratio\"] = df[f\"d1_{v}_max\"].div(df[f\"d1_{v}_min\"])\n",
    "\n",
    "        df[v+\"_h1_value_range\"] = df[f\"h1_{v}_max\"].subtract(df[f\"h1_{v}_min\"])\n",
    "        df[v+\"_h1_value_avg\"] = df[[f\"h1_{v}_max\", f\"h1_{v}_min\"]].mean(axis=1)\n",
    "        df[v+\"_h1_value_ratio\"] = df[f\"h1_{v}_max\"].div(df[f\"h1_{v}_min\"])\n",
    "        \n",
    "        df[v+\"_tot_change_value_range_normed\"] = abs((df[v+\"_d1_value_range\"].div(df[v+\"_h1_value_range\"])))\n",
    "        df[v+\"_started_after_firstHour\"] = ((df[f\"h1_{v}_max\"].isna()) & (df[f\"h1_{v}_min\"].isna())) & (~df[f\"d1_{v}_max\"].isna())\n",
    "        df[v+\"_day_more_extreme\"] = ((df[f\"d1_{v}_max\"]>df[f\"h1_{v}_max\"]) | (df[f\"d1_{v}_min\"]<df[f\"h1_{v}_min\"]))\n",
    "        df[v+\"_day_more_extreme\"].fillna(False)    \n",
    "\n",
    "after_first_hour(train_new)\n",
    "after_first_hour(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 471)\n",
      "(10234, 470)\n"
     ]
    }
   ],
   "source": [
    "def metrics_apache(df):\n",
    "    apache_cols = [c for c in df.columns if c[-6:] == 'apache']\n",
    "    for col in apache_cols:\n",
    "        if col in binary.columns:\n",
    "            df[col] = df[col]*5 # to have effect on further combination with other columns\n",
    "        else:\n",
    "            df[col] = df[col]**2\n",
    "            \n",
    "metrics_apache(train_new)\n",
    "metrics_apache(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 481)\n",
      "(10234, 480)\n"
     ]
    }
   ],
   "source": [
    "# Normal daily glucose level max: 7.8  min: 4 mmol/L\n",
    "# (ref: https://www.diabetes.co.uk/diabetes_care/blood-sugar-level-ranges.html)\n",
    "def metrics_glucose(df):  \n",
    "    high_glucose = df['d1_glucose_max'] > 7.8\n",
    "    low_glucose = df['d1_glucose_min'] < 4\n",
    "    df['glucose_imbalance'] = pd.Series()\n",
    "    df['glucose_imbalance'] = df[high_glucose | low_glucose]['glucose_imbalance'].apply(lambda x:1)\n",
    "    df['glucose_imbalance'] = df['glucose_imbalance'].replace(np.nan, 0)\n",
    "    df['glucose_imbalance_obese_adult'] = df['obese_adult'] * df['glucose_imbalance']\n",
    "    df['glucose_imbalance_diagonsis'] = df['glucose_imbalance'] * df['apache_3j_diagnosis'] + df['glucose_imbalance'] * df['apache_2_diagnosis']\n",
    "\n",
    "    df['d1_glucose_max_2'] = df['d1_glucose_max']**2\n",
    "    df['d1_glucose_min_2'] = df['d1_glucose_min']**2\n",
    "    df['glucose_age'] = df['glucose_apache'] * df['age']\n",
    "    df['glucose_imbalace_age'] = df['glucose_imbalance'] * df['age']**2\n",
    "    df['glucose_bmi'] = df['glucose_apache'] * df['bmi']\n",
    "    df['glucose_imbalace_bmi'] = df['glucose_imbalance'] * df['bmi']**2\n",
    "    df['glucose_peril'] = (df['d1_glucose_max'] - df['d1_glucose_min'])/df['d1_glucose_max']\n",
    "\n",
    "metrics_glucose(train_new)\n",
    "metrics_glucose(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 486)\n",
      "(10234, 485)\n"
     ]
    }
   ],
   "source": [
    "def metrics_renal(df):\n",
    "    df['arf_apache_diagonsis'] = df['arf_apache'] * df['apache_3j_diagnosis'] + df['arf_apache'] * df['apache_2_diagnosis']\n",
    "    df['bun_creatinine_max'] = df['d1_bun_max'] * df['d1_creatinine_max']\n",
    "    df['bun_creatinine_min'] = df['d1_bun_min'] * df['d1_creatinine_min']\n",
    "    df['metrics_renal'] = df['bun_apache'] * df['creatinine_apache']\n",
    "    df['metrics_renal_arf'] = df['metrics_renal'] * df['arf_apache'] \n",
    "    \n",
    "metrics_renal(train_new)\n",
    "metrics_renal(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 488)\n",
      "(10234, 487)\n"
     ]
    }
   ],
   "source": [
    "def metrics_gcs(df):\n",
    "    df['metrics_gcs'] = df['gcs_motor_apache'] * df['gcs_eyes_apache']\n",
    "    df['metrics_gcs_2'] = df['metrics_gcs'] * df['gcs_verbal_apache']\n",
    "\n",
    "metrics_gcs(train_new)\n",
    "metrics_gcs(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 494)\n",
      "(10234, 493)\n"
     ]
    }
   ],
   "source": [
    "def possible_dka(df):\n",
    "# diabetic ketoacidosis (DKA): disturbance of the mechanisms to maintain blood pH\n",
    "# an acute and life-threatening complication of the chronic metabolic disease\n",
    "# https://www.ncbi.nlm.nih.gov/pmc/articles/PMC151994/\n",
    "\n",
    "# normal anion gap: Na+ – (Cl– + HCO3–) = 12 mmol/L\n",
    "# In DKA, anion gap increased & the sum of bicarbonate and chloride concentrations is reduced\n",
    "# And potassium, lactate increased\n",
    "\n",
    "    for val in ['sodium', 'hco3', 'potassium', 'calcium']:\n",
    "        df[f'd1_{val}_max'] = df[f'd1_{val}_max'].replace(np.nan, df[f'd1_{val}_max'].mean())\n",
    "        df[f'd1_{val}_min'] = df[f'd1_{val}_min'].replace(np.nan, df[f'd1_{val}_min'].mean())\n",
    "        df[f'med_{val}'] = df[[f'd1_{val}_max', f'd1_{val}_min']].median(axis=1)\n",
    "\n",
    "# Also, arterial pH reduces (<=7.30)\n",
    "    acidic_blood_gas = (df[['d1_arterial_ph_max', 'd1_arterial_ph_min']].mean(axis=1) <= 7.30)\n",
    "    df['arterial_ph_low'] = acidic_blood_gas*1\n",
    "    df['arterial_ph_low_diagonsis'] = df['arterial_ph_low'] * df['apache_3j_diagnosis'] + df['arterial_ph_low'] * df['apache_2_diagnosis']\n",
    "\n",
    "possible_dka(train_new)\n",
    "possible_dka(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 503)\n",
      "(10234, 502)\n"
     ]
    }
   ],
   "source": [
    "def metrics_heart(df):    \n",
    "    mbp = [f for f in df.columns if f[3:6] == 'mbp']\n",
    "    df['metrics_mbp'] = df[mbp].sum(axis=1)\n",
    "    df['metrics_mbp'] = df['metrics_mbp'].fillna(df['metrics_mbp'].mean())\n",
    "    \n",
    "    sysbp = [f for f in df.columns if f[3:8] == 'sysbp']\n",
    "    df['metrics_sysbp'] = df[sysbp].sum(axis=1)\n",
    "    df['metrics_sysbp'] = df['metrics_sysbp'].fillna(df['metrics_sysbp'].mean())\n",
    "    df['d1_sysbp_max_2'] = df['d1_sysbp_max'].apply(lambda x: x*x)\n",
    "\n",
    "    diasbp = [f for f in df.columns if f[3:9] == 'diasbp']\n",
    "    df['metrics_diasbp'] = df[diasbp].sum(axis=1)\n",
    "    df['metrics_diasbp'] = df['metrics_diasbp'].fillna(df['metrics_diasbp'].mean())\n",
    "    df['d1_diasbp_max_2'] = df['d1_diasbp_max'].apply(lambda x: x*x)\n",
    "    \n",
    "    # In hypertensive diabetes mellitus, lower K & lower Ca & higher Na can be observed\n",
    "    # https://pubmed.ncbi.nlm.nih.gov/18488152/\n",
    "    # Hypertension: systolic > 129 mmHg or diastolic > 79 mmHg\n",
    "    d1_systolic_avg = df[['d1_sysbp_max', 'd1_sysbp_min']].mean(axis=1) > 129\n",
    "    d1_diastolic_avg = df[['d1_diasbp_max', 'd1_diasbp_min']].mean(axis=1) > 79\n",
    "    hypertensive = [d1_systolic_avg | d1_diastolic_avg]\n",
    "    df['hypertensive'] = hypertensive[0]*1\n",
    "    df['obese_adult_hypertensive'] = df['obese_adult'] * df['hypertensive']\n",
    "    \n",
    "    df['metrics_ions_max'] = (df['d1_calcium_max']+df['d1_potassium_max']) / df['d1_sodium_max']\n",
    "    df['metrics_ions_max'].replace(np.nan, df['metrics_ions_max'].mean(), inplace=True)\n",
    "    df['metrics_ions_min'] = (df['d1_calcium_min']+df['d1_potassium_min']) / df['d1_sodium_min']\n",
    "    df['metrics_ions_min'].replace(np.nan, df['metrics_ions_min'].mean(), inplace=True)\n",
    "\n",
    "metrics_heart(train_new)\n",
    "metrics_heart(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 506)\n",
      "(10234, 505)\n"
     ]
    }
   ],
   "source": [
    "def metrics_liver(df):\n",
    "    df['metrics_liver'] = df['cirrhosis'].fillna(0) * df['hepatic_failure'].fillna(0)\n",
    "    df['obese_adult_liver'] = df['obese_adult'] * df['metrics_liver']\n",
    "    df['obese_adult_glucose_imbalace_liver'] = df['glucose_imbalance'] + df['obese_adult_liver']\n",
    "\n",
    "metrics_liver(train_new)\n",
    "metrics_liver(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 509)\n",
      "(10234, 508)\n"
     ]
    }
   ],
   "source": [
    "def metrics_immune(df):\n",
    "    immune_cols = ['wbc_apache', 'immunosuppression', 'leukemia', 'lymphoma', 'aids', 'solid_tumor_with_metastasis']\n",
    "    df['metrics_immune'] = df[immune_cols].sum(axis=1)\n",
    "    df['metrics_immune'] = df['metrics_immune'].fillna(df['metrics_immune'].mode())\n",
    "    df['obese_adult_immunity'] = df['obese_adult'] * df['metrics_immune']\n",
    "    df['obese_adult_glucose_imbalace_immune'] = df['glucose_imbalance'] + df['obese_adult_immunity']\n",
    "    \n",
    "metrics_immune(train_new)\n",
    "metrics_immune(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 510)\n",
      "(10234, 509)\n"
     ]
    }
   ],
   "source": [
    "def pre_icu_days(df):\n",
    "    df['pre_icu_los_days'] = df['pre_icu_los_days']**2\n",
    "    df['pre_icu_day_obeseadult_gluimbalance'] = df['obese_adult'] * df['pre_icu_los_days'] + df['glucose_imbalance'] * df['pre_icu_los_days']\n",
    "\n",
    "\n",
    "pre_icu_days(train_new)\n",
    "pre_icu_days(test_new)\n",
    "\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0\n",
      "encounter_id\n",
      "147000    2.0\n",
      "147001    5.0\n",
      "147002    2.0\n",
      "147003    2.0\n",
      "147004    5.0\n",
      "Name: ethnicity, dtype: float64\n",
      "(130157, 510)\n",
      "(10234, 509)\n"
     ]
    }
   ],
   "source": [
    "# Ordinal encoding for categorical variables\n",
    "cols_to_encode = [col for col in test_new.columns if col in categoricals]\n",
    "\n",
    "oe = OrdinalEncoder()\n",
    "for col in cols_to_encode:\n",
    "    if train_new[col].mode().values[0] != np.nan:\n",
    "        train_new[col].replace(np.nan, train_new[col].mode().values[0], inplace=True)\n",
    "        test_new[col].replace(np.nan, test_new[col].mode().values[0], inplace=True)\n",
    "        \n",
    "    else:\n",
    "        train_new[col].replace(np.nan, 'NA', inplace=True)\n",
    "        test_new[col].replace(np.nan, 'NA', inplace=True)\n",
    "    \n",
    "    oe.fit(train_new[[col]])\n",
    "    train_new[col] = oe.transform(train_new[[col]])\n",
    "    test_new[col] = oe.transform(test_new[[col]])\n",
    "\n",
    "print(train_new.ethnicity.isnull().mean())\n",
    "print(train_new.ethnicity[:5])\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 515)\n",
      "(10234, 514)\n"
     ]
    }
   ],
   "source": [
    "def engineer_catcols(df):\n",
    "    df['ethnicity_bmi'] = df['ethnicity'] * df['bmi']\n",
    "    df['ethnicity_obeseadult_gluimbalance'] = df['ethnicity'] * df['obese_adult'] + df['glucose_imbalance']\n",
    "    df['ethnicity_obesehypertensive_gluimbalance'] = df['ethnicity']*df['obese_adult'] \\\n",
    "                                                    +  df['ethnicity']*df['hypertensive']  \\\n",
    "                                                    +  df['ethnicity']*df['glucose_imbalance']\n",
    "    df['icu_admit_source_pre_icu_days'] = df['icu_admit_source'] * df['pre_icu_los_days']\n",
    "    df['icu_type_pre_icu_days'] = df['icu_type'] * df['pre_icu_los_days']\n",
    "    \n",
    "engineer_catcols(train_new)\n",
    "engineer_catcols(test_new)\n",
    "print(train_new.shape)\n",
    "print(test_new.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 4. Modelling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(130157, 513)\n",
      "(10234, 512)\n"
     ]
    }
   ],
   "source": [
    "# Remove features which don't help improving the predictions\n",
    "cols_to_remove = ['readmission_status', 'hospital_id']\n",
    "\n",
    "# Define features and the target\n",
    "final_features = [c for c in test_new.columns if (c in train_new.columns) and (c not in cols_to_remove)]\n",
    "target = 'diabetes_mellitus'\n",
    "\n",
    "train_df = train_new[final_features + [target]].replace([np.inf, -np.inf], np.nan)\n",
    "test_df = test_new[final_features].replace([np.inf, -np.inf], np.nan)\n",
    "\n",
    "print(train_df.shape)\n",
    "print(test_df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "param = {\n",
    "       'boosting_type': 'gbdt',\n",
    "        'objective': 'binary',\n",
    "        'metric': 'auc',\n",
    "        'learning_rate': 0.01,\n",
    "        'subsample': .8,\n",
    "        'colsample_bytree': 0.08,\n",
    "        'reg_alpha': .2,\n",
    "        'reg_lambda': .2,\n",
    "        'scale_pos_weight': 2,\n",
    "        'n_estimators': 10000,\n",
    "        'verbose': 1,\n",
    "        'max_depth': -1,\n",
    "        'seed':100, \n",
    "        'force_col_wise': True\n",
    "        }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "fold 1\n",
      "[LightGBM] [Info] Number of positive: 22521, number of negative: 81604\n",
      "[LightGBM] [Info] Total Bins 77161\n",
      "[LightGBM] [Info] Number of data points in the train set: 104125, number of used features: 509\n",
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.216288 -> initscore=-1.287430\n",
      "[LightGBM] [Info] Start training from score -1.287430\n",
      "Training until validation scores don't improve for 1000 rounds\n",
      "[500]\tvalid_0's auc: 0.854534\n",
      "[1000]\tvalid_0's auc: 0.864687\n",
      "[1500]\tvalid_0's auc: 0.868711\n",
      "[2000]\tvalid_0's auc: 0.870845\n",
      "[2500]\tvalid_0's auc: 0.871907\n",
      "[3000]\tvalid_0's auc: 0.872584\n",
      "[3500]\tvalid_0's auc: 0.873041\n",
      "[4000]\tvalid_0's auc: 0.873278\n",
      "[4500]\tvalid_0's auc: 0.873431\n",
      "[5000]\tvalid_0's auc: 0.873579\n",
      "[5500]\tvalid_0's auc: 0.873626\n",
      "[6000]\tvalid_0's auc: 0.873708\n",
      "[6500]\tvalid_0's auc: 0.873774\n",
      "[7000]\tvalid_0's auc: 0.873818\n",
      "[7500]\tvalid_0's auc: 0.873868\n",
      "[8000]\tvalid_0's auc: 0.873865\n",
      "[8500]\tvalid_0's auc: 0.873852\n",
      "Early stopping, best iteration is:\n",
      "[7565]\tvalid_0's auc: 0.873913\n",
      "\n",
      "fold 2\n",
      "[LightGBM] [Info] Number of positive: 22520, number of negative: 81605\n",
      "[LightGBM] [Info] Total Bins 77134\n",
      "[LightGBM] [Info] Number of data points in the train set: 104125, number of used features: 510\n",
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.216279 -> initscore=-1.287487\n",
      "[LightGBM] [Info] Start training from score -1.287487\n",
      "Training until validation scores don't improve for 1000 rounds\n",
      "[500]\tvalid_0's auc: 0.853618\n",
      "[1000]\tvalid_0's auc: 0.8636\n",
      "[1500]\tvalid_0's auc: 0.867824\n",
      "[2000]\tvalid_0's auc: 0.870112\n",
      "[2500]\tvalid_0's auc: 0.871248\n",
      "[3000]\tvalid_0's auc: 0.871711\n",
      "[3500]\tvalid_0's auc: 0.872114\n",
      "[4000]\tvalid_0's auc: 0.872321\n",
      "[4500]\tvalid_0's auc: 0.872544\n",
      "[5000]\tvalid_0's auc: 0.872521\n",
      "[5500]\tvalid_0's auc: 0.872579\n",
      "[6000]\tvalid_0's auc: 0.872696\n",
      "[6500]\tvalid_0's auc: 0.87278\n",
      "[7000]\tvalid_0's auc: 0.872841\n",
      "[7500]\tvalid_0's auc: 0.87281\n",
      "[8000]\tvalid_0's auc: 0.872739\n",
      "Early stopping, best iteration is:\n",
      "[7019]\tvalid_0's auc: 0.872854\n",
      "\n",
      "fold 3\n",
      "[LightGBM] [Info] Number of positive: 22521, number of negative: 81605\n",
      "[LightGBM] [Info] Total Bins 77089\n",
      "[LightGBM] [Info] Number of data points in the train set: 104126, number of used features: 509\n",
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.216286 -> initscore=-1.287442\n",
      "[LightGBM] [Info] Start training from score -1.287442\n",
      "Training until validation scores don't improve for 1000 rounds\n",
      "[500]\tvalid_0's auc: 0.857649\n",
      "[1000]\tvalid_0's auc: 0.866814\n",
      "[1500]\tvalid_0's auc: 0.870578\n",
      "[2000]\tvalid_0's auc: 0.87261\n",
      "[2500]\tvalid_0's auc: 0.873474\n",
      "[3000]\tvalid_0's auc: 0.874052\n",
      "[3500]\tvalid_0's auc: 0.874437\n",
      "[4000]\tvalid_0's auc: 0.874419\n",
      "[4500]\tvalid_0's auc: 0.874492\n",
      "[5000]\tvalid_0's auc: 0.874505\n",
      "[5500]\tvalid_0's auc: 0.874391\n",
      "Early stopping, best iteration is:\n",
      "[4659]\tvalid_0's auc: 0.874525\n",
      "\n",
      "fold 4\n",
      "[LightGBM] [Info] Number of positive: 22521, number of negative: 81605\n",
      "[LightGBM] [Info] Total Bins 77138\n",
      "[LightGBM] [Info] Number of data points in the train set: 104126, number of used features: 509\n",
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.216286 -> initscore=-1.287442\n",
      "[LightGBM] [Info] Start training from score -1.287442\n",
      "Training until validation scores don't improve for 1000 rounds\n",
      "[500]\tvalid_0's auc: 0.853854\n",
      "[1000]\tvalid_0's auc: 0.863741\n",
      "[1500]\tvalid_0's auc: 0.867578\n",
      "[2000]\tvalid_0's auc: 0.869748\n",
      "[2500]\tvalid_0's auc: 0.870844\n",
      "[3000]\tvalid_0's auc: 0.871349\n",
      "[3500]\tvalid_0's auc: 0.871731\n",
      "[4000]\tvalid_0's auc: 0.871851\n",
      "[4500]\tvalid_0's auc: 0.871826\n",
      "[5000]\tvalid_0's auc: 0.871883\n",
      "[5500]\tvalid_0's auc: 0.871897\n",
      "[6000]\tvalid_0's auc: 0.871832\n",
      "Early stopping, best iteration is:\n",
      "[5108]\tvalid_0's auc: 0.87192\n",
      "\n",
      "fold 5\n",
      "[LightGBM] [Info] Number of positive: 22521, number of negative: 81605\n",
      "[LightGBM] [Info] Total Bins 77080\n",
      "[LightGBM] [Info] Number of data points in the train set: 104126, number of used features: 510\n",
      "[LightGBM] [Info] [binary:BoostFromScore]: pavg=0.216286 -> initscore=-1.287442\n",
      "[LightGBM] [Info] Start training from score -1.287442\n",
      "Training until validation scores don't improve for 1000 rounds\n",
      "[500]\tvalid_0's auc: 0.857885\n",
      "[1000]\tvalid_0's auc: 0.867947\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "nfold = 5\n",
    "\n",
    "predictors = final_features\n",
    "\n",
    "skf = StratifiedKFold(n_splits=nfold, shuffle=True, random_state=1001)\n",
    "\n",
    "oof = np.zeros(len(train_df))\n",
    "predictions = np.zeros(len(test_df))\n",
    "\n",
    "i = 1\n",
    "for train_index, valid_index in skf.split(train_df, train_df[target].values):\n",
    "    print(\"\\nfold {}\".format(i))\n",
    "    xg_train = lgb.Dataset(train_df.iloc[train_index][predictors].values,\n",
    "                           label=train_df.iloc[train_index][target].values,\n",
    "                           feature_name=predictors,\n",
    "                           free_raw_data = False\n",
    "                           )\n",
    "    xg_valid = lgb.Dataset(train_df.iloc[valid_index][predictors].values,\n",
    "                           label=train_df.iloc[valid_index][target].values,\n",
    "                           feature_name=predictors,\n",
    "                           free_raw_data = False\n",
    "                           )   \n",
    "\n",
    "    \n",
    "    clf = lgb.train(param, xg_train, valid_sets = [xg_valid], verbose_eval=500, early_stopping_rounds = 1000)\n",
    "    oof[valid_index] = clf.predict(train_df.iloc[valid_index][predictors].values, num_iteration=clf.best_iteration) \n",
    "    \n",
    "    predictions += clf.predict(test_df[predictors], num_iteration=clf.best_iteration) / nfold\n",
    "    i = i + 1\n",
    "\n",
    "print(\"\\n\\nCV AUC: {:<0.4f}\".format(roc_auc_score(train_df[target].values, oof)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 5. Predict and Submit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "template = pd.read_csv(\"../input/solutiontemplatewids2021/SolutionTemplateWiDS2021.csv\")\n",
    "template = template.sort_values('encounter_id')\n",
    "print(template.shape)\n",
    "template.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "template['diabetes_mellitus'] = predictions\n",
    "print(template.isnull().sum())\n",
    "template.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Convert DataFrame to a csv file that can be uploaded\n",
    "#This is saved in the same directory as your notebook\n",
    "filename = 'Diabetes Mellitus Predictions.csv'\n",
    "template.to_csv(filename, index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}