notebooks/data_balance_e2e.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# End to End Data Balance and Error Mitigation\n",
    "This Notebook will demonstrate how to use both the Data Balance Analysis capabilities and error mitigation functions together using an example HR dataset which is a tabular dataset with a label column that indicates whether or not a person is promoted based on attributes such as education, gender, number of trainings, and other factors. \n",
    "The steps that we will take in this notebook are \n",
    "\n",
    "1. We will first conduct an analysis on how balanced the data is. \n",
    "2. We will train an example model to see how it performs on the data using metrics and the Responsible AI Error Analysis Dashboard. \n",
    "3. We will try to balance the data to mitigate biases that may have resulted from unbalanced data\n",
    "5. We will train a new model using our data after applying the rebalance\n",
    "4. We will then compare model performance before and after mitigating data imbalances by again using various performance metrics and the Error Analysis Dashboard."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First we import all the dependencies needed in our analysis. This includes the classes to produce the data balance metrics, the sklearn functions to see the model performance and the error mitigation steps like DataRebalance and DataSplit that we apply to the dataset itself. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "from lightgbm import LGBMClassifier\n",
    "from sklearn import metrics\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix, classification_report\n",
    "from sklearn.preprocessing import OrdinalEncoder\n",
    "\n",
    "from raimitigations.utils import split_data\n",
    "import raimitigations.dataprocessing as dp\n",
    "from raimitigations.databalanceanalysis import (\n",
    "    FeatureBalanceMeasure,\n",
    "    AggregateBalanceMeasure,\n",
    "    DistributionBalanceMeasure,\n",
    ")\n",
    "from raimitigations.cohort import CohortManager\n",
    "from download import download_datasets"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train LightGBM Model\n",
    "Now we import the tabular dataset that we will look at in the example, we load it into a pandas dataframe that we can then modify and use for all the other steps. For the data balance analysis portion we need our label columns and a list of sensitive columns that are interested in checking for balance. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_dir = \"../datasets/\"\n",
    "download_datasets(data_dir)\n",
    "df = pd.read_csv(data_dir + 'hr_promotion/train.csv')\n",
    "df.drop(columns=['employee_id'], inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>department</th>\n",
       "      <th>region</th>\n",
       "      <th>education</th>\n",
       "      <th>gender</th>\n",
       "      <th>recruitment_channel</th>\n",
       "      <th>no_of_trainings</th>\n",
       "      <th>age</th>\n",
       "      <th>previous_year_rating</th>\n",
       "      <th>length_of_service</th>\n",
       "      <th>KPIs_met &gt;80%</th>\n",
       "      <th>awards_won?</th>\n",
       "      <th>avg_training_score</th>\n",
       "      <th>is_promoted</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_7</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>f</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>1</td>\n",
       "      <td>35</td>\n",
       "      <td>5.0</td>\n",
       "      <td>8</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>49</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Operations</td>\n",
       "      <td>region_22</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>30</td>\n",
       "      <td>5.0</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>60</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_19</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>1</td>\n",
       "      <td>34</td>\n",
       "      <td>3.0</td>\n",
       "      <td>7</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>50</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_23</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>2</td>\n",
       "      <td>39</td>\n",
       "      <td>1.0</td>\n",
       "      <td>10</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>50</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Technology</td>\n",
       "      <td>region_26</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>45</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>73</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54803</th>\n",
       "      <td>Technology</td>\n",
       "      <td>region_14</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>1</td>\n",
       "      <td>48</td>\n",
       "      <td>3.0</td>\n",
       "      <td>17</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>78</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54804</th>\n",
       "      <td>Operations</td>\n",
       "      <td>region_27</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>f</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>37</td>\n",
       "      <td>2.0</td>\n",
       "      <td>6</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>56</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54805</th>\n",
       "      <td>Analytics</td>\n",
       "      <td>region_1</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>27</td>\n",
       "      <td>5.0</td>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>79</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54806</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_9</td>\n",
       "      <td>NaN</td>\n",
       "      <td>m</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>1</td>\n",
       "      <td>29</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>45</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>54807</th>\n",
       "      <td>HR</td>\n",
       "      <td>region_22</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>27</td>\n",
       "      <td>1.0</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>49</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>54808 rows × 13 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              department     region         education gender  \\\n",
       "0      Sales & Marketing   region_7  Master's & above      f   \n",
       "1             Operations  region_22        Bachelor's      m   \n",
       "2      Sales & Marketing  region_19        Bachelor's      m   \n",
       "3      Sales & Marketing  region_23        Bachelor's      m   \n",
       "4             Technology  region_26        Bachelor's      m   \n",
       "...                  ...        ...               ...    ...   \n",
       "54803         Technology  region_14        Bachelor's      m   \n",
       "54804         Operations  region_27  Master's & above      f   \n",
       "54805          Analytics   region_1        Bachelor's      m   \n",
       "54806  Sales & Marketing   region_9               NaN      m   \n",
       "54807                 HR  region_22        Bachelor's      m   \n",
       "\n",
       "      recruitment_channel  no_of_trainings  age  previous_year_rating  \\\n",
       "0                sourcing                1   35                   5.0   \n",
       "1                   other                1   30                   5.0   \n",
       "2                sourcing                1   34                   3.0   \n",
       "3                   other                2   39                   1.0   \n",
       "4                   other                1   45                   3.0   \n",
       "...                   ...              ...  ...                   ...   \n",
       "54803            sourcing                1   48                   3.0   \n",
       "54804               other                1   37                   2.0   \n",
       "54805               other                1   27                   5.0   \n",
       "54806            sourcing                1   29                   1.0   \n",
       "54807               other                1   27                   1.0   \n",
       "\n",
       "       length_of_service  KPIs_met >80%  awards_won?  avg_training_score  \\\n",
       "0                      8              1            0                  49   \n",
       "1                      4              0            0                  60   \n",
       "2                      7              0            0                  50   \n",
       "3                     10              0            0                  50   \n",
       "4                      2              0            0                  73   \n",
       "...                  ...            ...          ...                 ...   \n",
       "54803                 17              0            0                  78   \n",
       "54804                  6              0            0                  56   \n",
       "54805                  3              1            0                  79   \n",
       "54806                  2              0            0                  45   \n",
       "54807                  5              0            0                  49   \n",
       "\n",
       "       is_promoted  \n",
       "0                0  \n",
       "1                0  \n",
       "2                0  \n",
       "3                0  \n",
       "4                0  \n",
       "...            ...  \n",
       "54803            0  \n",
       "54804            0  \n",
       "54805            0  \n",
       "54806            0  \n",
       "54807            0  \n",
       "\n",
       "[54808 rows x 13 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We do some data transformation on the categorical columns in order to make the training data input in the format that the lightGBM model expects. Although lightGBM can internally deal with categorical columns that the user specifies, we need to encode those categories into integers before we are able to train the lightGBM model on it, so we create a scikit pipeline contains the encoder and the LGBM classifier."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>department</th>\n",
       "      <th>region</th>\n",
       "      <th>education</th>\n",
       "      <th>gender</th>\n",
       "      <th>recruitment_channel</th>\n",
       "      <th>no_of_trainings</th>\n",
       "      <th>age</th>\n",
       "      <th>previous_year_rating</th>\n",
       "      <th>length_of_service</th>\n",
       "      <th>KPIs_met &gt;80%</th>\n",
       "      <th>awards_won?</th>\n",
       "      <th>avg_training_score</th>\n",
       "      <th>is_promoted</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_7</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>f</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>1</td>\n",
       "      <td>35</td>\n",
       "      <td>5.0</td>\n",
       "      <td>8</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>49</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Operations</td>\n",
       "      <td>region_22</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>30</td>\n",
       "      <td>5.0</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>60</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_19</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>1</td>\n",
       "      <td>34</td>\n",
       "      <td>3.0</td>\n",
       "      <td>7</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>50</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_23</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>2</td>\n",
       "      <td>39</td>\n",
       "      <td>1.0</td>\n",
       "      <td>10</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>50</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Technology</td>\n",
       "      <td>region_26</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>other</td>\n",
       "      <td>1</td>\n",
       "      <td>45</td>\n",
       "      <td>3.0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>73</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          department     region         education gender recruitment_channel  \\\n",
       "0  Sales & Marketing   region_7  Master's & above      f            sourcing   \n",
       "1         Operations  region_22        Bachelor's      m               other   \n",
       "2  Sales & Marketing  region_19        Bachelor's      m            sourcing   \n",
       "3  Sales & Marketing  region_23        Bachelor's      m               other   \n",
       "4         Technology  region_26        Bachelor's      m               other   \n",
       "\n",
       "   no_of_trainings  age  previous_year_rating  length_of_service  \\\n",
       "0                1   35                   5.0                  8   \n",
       "1                1   30                   5.0                  4   \n",
       "2                1   34                   3.0                  7   \n",
       "3                2   39                   1.0                 10   \n",
       "4                1   45                   3.0                  2   \n",
       "\n",
       "   KPIs_met >80%  awards_won?  avg_training_score  is_promoted  \n",
       "0              1            0                  49            0  \n",
       "1              0            0                  60            0  \n",
       "2              0            0                  50            0  \n",
       "3              0            0                  50            0  \n",
       "4              0            0                  73            0  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cols_of_interest = [\"education\", \"recruitment_channel\"]\n",
    "categorical_cols = [\n",
    "    \"department\",\n",
    "    \"gender\",\n",
    "    \"education\",\n",
    "    \"region\",\n",
    "    \"recruitment_channel\",\n",
    "]\n",
    "label_col = \"is_promoted\"\n",
    "seed = 42\n",
    "# handle duplicates\n",
    "df = df.drop_duplicates().dropna()\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.pipeline import Pipeline\n",
    "\n",
    "pipeline = Pipeline([\n",
    "                (\"ohe\", dp.EncoderOHE(col_encode=categorical_cols, drop=False)),\n",
    "                (\"model\", LGBMClassifier(n_estimators=50, random_state=42)),\n",
    "            ])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here we do a split on the data, train a LightGBM model using the scikit pipeline we created, and see how this model does on some test data. After this processing, we train the model and we can see that the model does well on false values, getting 97.3% of them correct, but the model does a lot worse on the true values, only identifying approximately a third of the true positives correctly. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def conf_matrix(y, pred):\n",
    "    ((tn, fp), (fn, tp)) = metrics.confusion_matrix(y, pred)\n",
    "    ((tnr, fpr), (fnr, tpr)) = metrics.confusion_matrix(y, pred, normalize=\"true\")\n",
    "    return pd.DataFrame(\n",
    "        [\n",
    "            [f\"TP = {tp} ({tpr:1.2%})\", f\"FN = {fn} ({fnr:1.2%})\"],\n",
    "            [f\"FP = {fp} ({fpr:1.2%})\", f\"TN = {tn} ({tnr:1.2%})\"],\n",
    "        ],\n",
    "        index=[\"True\", \"False\"],\n",
    "        columns=[\"Pred 1\", \"Pred 0\"],\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "number of errors on test dataset: 293\n",
      "                  Pred 1              Pred 0\n",
      "True   TP = 142 (33.57%)   FN = 281 (66.43%)\n",
      "False    FP = 12 (0.27%)  TN = 4426 (99.73%)\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.94      1.00      0.97      4438\n",
      "           1       0.92      0.34      0.49       423\n",
      "\n",
      "    accuracy                           0.94      4861\n",
      "   macro avg       0.93      0.67      0.73      4861\n",
      "weighted avg       0.94      0.94      0.93      4861\n",
      "\n"
     ]
    }
   ],
   "source": [
    "## Train a model and get accuracy numbers\n",
    "# splitting the dataset\n",
    "np.random.seed(42)\n",
    "x_train, x_test, y_train, y_test = split_data(df, label_col, test_size=0.1)\n",
    "# fitting the pipeline and predicting\n",
    "pipeline.fit(x_train, y_train)\n",
    "pred = pipeline.predict(x_test)\n",
    "\n",
    "\n",
    "print(\"number of errors on test dataset: \" + str(sum(pred != y_test)))\n",
    "print(conf_matrix(y_test, pred))\n",
    "print(classification_report(y_test, pred))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Error Analysis on Baseline Model\n",
    "Now that we have a baseline model to work with, we can see how this model is doing overall on the data and see if there are any cohorts within the data that the model performs worse on. Even if the model has a high accuracy, we want to make sure that accuracy is applicable over different sensitive groups. We use the [Error Analysis Dashboard](https://erroranalysis.ai/) to determine which cohorts of data this model performs worse on. To install this library, use the following command:\n",
    "\n",
    "```console\n",
    "> pip install raiwidgets\n",
    "```\n",
    "\n",
    "In the following cell, we set up the parameters required by the [Responsible AI Toolbox](https://github.com/microsoft/responsible-ai-toolbox)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using categorical_feature in Dataset.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n",
      "Passing a set as an indexer is deprecated and will raise in a future version. Use a list instead.\n"
     ]
    }
   ],
   "source": [
    "from raiwidgets import ResponsibleAIDashboard\n",
    "from responsibleai import RAIInsights\n",
    "from raiwidgets.cohort import Cohort, CohortFilter, CohortFilterMethods\n",
    "\n",
    "df_train = x_train.copy()\n",
    "df_train[label_col] = y_train\n",
    "df_test = x_test.copy()\n",
    "df_test[label_col] = y_test\n",
    "\n",
    "rai_insights = RAIInsights(pipeline, df_train, df_test, label_col, 'classification',\n",
    "                           categorical_features=pipeline['ohe'].get_encoded_columns())\n",
    "\n",
    "# Error Analysis\n",
    "rai_insights.error_analysis.add()\n",
    "\n",
    "rai_insights.compute()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now create the dashboard, which will be available in the local host. Open your browser on the indicated local host address and you'll be able to interact with the dashboard."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ResponsibleAI started at http://localhost:5000\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<raiwidgets.responsibleai_dashboard.ResponsibleAIDashboard at 0x7f77dbaf08b0>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ResponsibleAIDashboard(rai_insights)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since the error analysis dashboard is interactive and too large to render on Github, we will include screenshots from our analysis.\n",
    "\n",
    "![screen1](./images/error_analysis_adult_census3.png)\n",
    "\n",
    "![screen2](./images/error_analysis1.png)\n",
    "\n",
    "![screen3](./images/error_analysis2.png)\n",
    "\n",
    "From these screenshots we can see that if we zoom in on certain cohorts that the model is getting more errors on, that region, department and education are all attributes that are involved in those cohorts. For the purpose of this example, we chose remove some of the other columns like KPIs_met from the error analysis since we want to focus on attributes that may lead to biases rather than more measurable attributes. We will focus on analyzing and mitigating errors within the department and education columns for the rest of the analysis. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data Balance Analysis\n",
    "First we can take a look at the feature balance measures. These measures indicate the difference in the label column amongst different feature values. For example the first row here indicates if people with the \"Masters & above\" education has a different proportion of people receiving the promoted outcome than those that have a Bachelor's. Lower values of these measures indicates that the amounts of people with class A vs versus those with class B with a label of 1 is similar. The t-test value can also tell us if the difference we see is statistically significant."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "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>ClassA</th>\n",
       "      <th>ClassB</th>\n",
       "      <th>FeatureName</th>\n",
       "      <th>dp</th>\n",
       "      <th>pmi</th>\n",
       "      <th>sdc</th>\n",
       "      <th>ji</th>\n",
       "      <th>krc</th>\n",
       "      <th>llr</th>\n",
       "      <th>t_test</th>\n",
       "      <th>ttest_pvalue</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>education</td>\n",
       "      <td>0.017814</td>\n",
       "      <td>0.197252</td>\n",
       "      <td>0.004788</td>\n",
       "      <td>0.005594</td>\n",
       "      <td>4.996442</td>\n",
       "      <td>-0.622067</td>\n",
       "      <td>0.262386</td>\n",
       "      <td>0.408789</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>Below Secondary</td>\n",
       "      <td>education</td>\n",
       "      <td>0.022587</td>\n",
       "      <td>0.257448</td>\n",
       "      <td>0.069726</td>\n",
       "      <td>0.076144</td>\n",
       "      <td>-27.035372</td>\n",
       "      <td>3.718893</td>\n",
       "      <td>-2.623174</td>\n",
       "      <td>0.059886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>Below Secondary</td>\n",
       "      <td>education</td>\n",
       "      <td>0.004773</td>\n",
       "      <td>0.060196</td>\n",
       "      <td>0.064938</td>\n",
       "      <td>0.070550</td>\n",
       "      <td>-32.031814</td>\n",
       "      <td>4.340960</td>\n",
       "      <td>-2.885561</td>\n",
       "      <td>0.051023</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>sourcing</td>\n",
       "      <td>other</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.002630</td>\n",
       "      <td>0.030454</td>\n",
       "      <td>-0.000794</td>\n",
       "      <td>-0.000925</td>\n",
       "      <td>1.579734</td>\n",
       "      <td>-0.240590</td>\n",
       "      <td>0.085323</td>\n",
       "      <td>0.469888</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>sourcing</td>\n",
       "      <td>referred</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>-0.038804</td>\n",
       "      <td>-0.366366</td>\n",
       "      <td>0.048196</td>\n",
       "      <td>0.053285</td>\n",
       "      <td>-18.723860</td>\n",
       "      <td>2.639673</td>\n",
       "      <td>-1.733789</td>\n",
       "      <td>0.112546</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             ClassA           ClassB          FeatureName        dp       pmi  \\\n",
       "0  Master's & above       Bachelor's            education  0.017814  0.197252   \n",
       "1  Master's & above  Below Secondary            education  0.022587  0.257448   \n",
       "2        Bachelor's  Below Secondary            education  0.004773  0.060196   \n",
       "0          sourcing            other  recruitment_channel  0.002630  0.030454   \n",
       "1          sourcing         referred  recruitment_channel -0.038804 -0.366366   \n",
       "\n",
       "        sdc        ji        krc       llr    t_test  ttest_pvalue  \n",
       "0  0.004788  0.005594   4.996442 -0.622067  0.262386      0.408789  \n",
       "1  0.069726  0.076144 -27.035372  3.718893 -2.623174      0.059886  \n",
       "2  0.064938  0.070550 -32.031814  4.340960 -2.885561      0.051023  \n",
       "0 -0.000794 -0.000925   1.579734 -0.240590  0.085323      0.469888  \n",
       "1  0.048196  0.053285 -18.723860  2.639673 -1.733789      0.112546  "
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "feature_measures = FeatureBalanceMeasure(cols_of_interest, label_col)\n",
    "\n",
    "train_df = x_train.copy()\n",
    "train_df[label_col] = y_train\n",
    "\n",
    "feat_measures1 = feature_measures.measures(train_df)\n",
    "feat_measures1.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline \n",
    "educations = train_df[\"education\"].unique()\n",
    "education_dp_values = feat_measures1[feat_measures1[\"FeatureName\"] == \"education\"][\n",
    "    [\"ClassA\", \"ClassB\", \"pmi\"]\n",
    "]\n",
    "education_dp_array = np.zeros((len(educations), len(educations)))\n",
    "\n",
    "for idx, row in education_dp_values.iterrows():\n",
    "    class_a = row[0]\n",
    "    class_b = row[1]\n",
    "    dp_value = row[2]\n",
    "    i, j = np.where(educations == class_a)[0][0], np.where(educations == class_b)\n",
    "    dp_value = round(dp_value, 2)\n",
    "    education_dp_array[i, j] = dp_value\n",
    "    education_dp_array[j, i] = -1 * dp_value\n",
    "\n",
    "colormap = \"RdBu\"\n",
    "dp_min, dp_max = -1.0, 1.0\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "im = ax.imshow(education_dp_array, vmin=dp_min, vmax=dp_max, cmap=colormap)\n",
    "\n",
    "cbar = ax.figure.colorbar(im, ax=ax)\n",
    "cbar.ax.set_ylabel(\"Point Mutual Info\", rotation=-90, va=\"bottom\")\n",
    "\n",
    "ax.set_xticks(np.arange(len(educations)))\n",
    "ax.set_yticks(np.arange(len(educations)))\n",
    "ax.set_xticklabels(educations)\n",
    "ax.set_yticklabels(educations)\n",
    "\n",
    "plt.setp(ax.get_xticklabels(), rotation=45, ha=\"right\", rotation_mode=\"anchor\")\n",
    "\n",
    "for i in range(len(educations)):\n",
    "    for j in range(len(educations)):\n",
    "        text = ax.text(\n",
    "            j, i, education_dp_array[i, j], ha=\"center\", va=\"center\", color=\"k\"\n",
    "        )\n",
    "\n",
    "ax.set_title(\"PMI of education in HR Dataset\")\n",
    "fig.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next we can take a look at the distribution balance measures. These measures each of the columns of interest that we selected to the uniform distribution of those values. Values that are closer to zero indicate that the difference between the actual distribution of the data and the uniform distribution of values. We find that these values are pretty far from zero, so they don't align to a uniform distribution well before we make any modifications to the original data. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "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>FeatureName</th>\n",
       "      <th>kl_divergence</th>\n",
       "      <th>js_dist</th>\n",
       "      <th>wasserstein_dist</th>\n",
       "      <th>inf_norm_dist</th>\n",
       "      <th>total_variation_dist</th>\n",
       "      <th>chi_sq_p_value</th>\n",
       "      <th>chi_sq_stat</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>education</td>\n",
       "      <td>0.434946</td>\n",
       "      <td>0.358906</td>\n",
       "      <td>0.236105</td>\n",
       "      <td>0.354158</td>\n",
       "      <td>0.354158</td>\n",
       "      <td>0.0</td>\n",
       "      <td>30343.542312</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.327144</td>\n",
       "      <td>0.315774</td>\n",
       "      <td>0.208248</td>\n",
       "      <td>0.312371</td>\n",
       "      <td>0.312371</td>\n",
       "      <td>0.0</td>\n",
       "      <td>20349.678919</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           FeatureName  kl_divergence   js_dist  wasserstein_dist  \\\n",
       "0            education       0.434946  0.358906          0.236105   \n",
       "1  recruitment_channel       0.327144  0.315774          0.208248   \n",
       "\n",
       "   inf_norm_dist  total_variation_dist  chi_sq_p_value   chi_sq_stat  \n",
       "0       0.354158              0.354158             0.0  30343.542312  \n",
       "1       0.312371              0.312371             0.0  20349.678919  "
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dist_measures = DistributionBalanceMeasure(cols_of_interest)\n",
    "dist_measures1 = dist_measures.measures(train_df)\n",
    "dist_measures1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline \n",
    "measures_of_interest = [\n",
    "    \"kl_divergence\",\n",
    "    \"js_dist\",\n",
    "    \"inf_norm_dist\",\n",
    "    \"total_variation_dist\",\n",
    "    \"wasserstein_dist\",\n",
    "]\n",
    "education_measures = dist_measures1[dist_measures1[\"FeatureName\"] == \"education\"].iloc[\n",
    "    0\n",
    "]\n",
    "recruitment_measures = dist_measures1[\n",
    "    dist_measures1[\"FeatureName\"] == \"recruitment_channel\"\n",
    "].iloc[0]\n",
    "education_array = [\n",
    "    round(education_measures[measure], 4) for measure in measures_of_interest\n",
    "]\n",
    "recruitment_array = [\n",
    "    round(recruitment_measures[measure], 4) for measure in measures_of_interest\n",
    "]\n",
    "\n",
    "x = np.arange(len(measures_of_interest))\n",
    "width = 0.35\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "rects1 = ax.bar(x - width / 2, education_array, width, label=\"education\")\n",
    "rects2 = ax.bar(x + width / 2, recruitment_array, width, label=\"recruitment_channel\")\n",
    "\n",
    "ax.set_xlabel(\"Measure\")\n",
    "ax.set_ylabel(\"Value\")\n",
    "ax.set_title(\n",
    "    \"Distribution Balance Measures of Education and Recruitment Channel in Adult Dataset\"\n",
    ")\n",
    "ax.set_xticks(x)\n",
    "ax.set_xticklabels(measures_of_interest)\n",
    "ax.legend()\n",
    "\n",
    "plt.setp(ax.get_xticklabels(), rotation=20, ha=\"right\", rotation_mode=\"default\")\n",
    "\n",
    "\n",
    "def autolabel(rects):\n",
    "    for rect in rects:\n",
    "        height = rect.get_height()\n",
    "        ax.annotate(\n",
    "            \"{}\".format(height),\n",
    "            xy=(rect.get_x() + rect.get_width() / 2, height),\n",
    "            xytext=(0, 1),  # 1 point vertical offset\n",
    "            textcoords=\"offset points\",\n",
    "            ha=\"center\",\n",
    "            va=\"bottom\",\n",
    "        )\n",
    "\n",
    "\n",
    "autolabel(rects1)\n",
    "autolabel(rects2)\n",
    "\n",
    "fig.tight_layout()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also look at aggregate balance measures which indicate a notion of overall inequality in the data. We can see that the Atkinson Index is 0.79. This means that in order to create a perfectly balanced dataset over these measures we would need to forgo 79.9% of the data. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "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>theil_l_index</th>\n",
       "      <th>theil_t_index</th>\n",
       "      <th>atkinson_index</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.600814</td>\n",
       "      <td>0.762696</td>\n",
       "      <td>0.798268</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   theil_l_index  theil_t_index  atkinson_index\n",
       "0       1.600814       0.762696        0.798268"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "agg_measures = AggregateBalanceMeasure(cols_of_interest)\n",
    "agg_measures1 = agg_measures.measures(train_df)\n",
    "agg_measures1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Error Mitigation: Rebalancing dataset\n",
    "\n",
    "In order to rebalance the data we can choose from three different methods of under or oversampling. These are SMOTE, Tomek and SMOTE-Tomek. SMOTE is an oversampling technique for the less represented class. Tomek is an undersampling technique that would be applied to the more represented class. Smote-Tomek is when both of these methods are applied in conjunction on the dataset. In this example, we will choose to use the SMOTE sampling technique on the columns of interest. The rebalance operation is originally aimed for a single column, that is, the entire dataset is rebalanced based on one column. However, the ``CohortManager`` class allows us to specify a set of cohorts, where the rebalance operation will then be applied to each cohort individually.\n",
    "\n",
    "In the next cell, we will rebalance our dataset using the `recruitment_channel` column as the target column, and the `education` column as the cohort division, that is, the ``CohortManager`` class will first separate the dataset into multiple cohorts, one for each of the existing values in the `education` column. Next, it will run the rebalance operation (over the `recruitment_channel` column) for each cohort separately."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "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>department</th>\n",
       "      <th>region</th>\n",
       "      <th>education</th>\n",
       "      <th>gender</th>\n",
       "      <th>no_of_trainings</th>\n",
       "      <th>age</th>\n",
       "      <th>previous_year_rating</th>\n",
       "      <th>length_of_service</th>\n",
       "      <th>KPIs_met &gt;80%</th>\n",
       "      <th>awards_won?</th>\n",
       "      <th>avg_training_score</th>\n",
       "      <th>is_promoted</th>\n",
       "      <th>recruitment_channel</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Procurement</td>\n",
       "      <td>region_7</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>f</td>\n",
       "      <td>1</td>\n",
       "      <td>29</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>74</td>\n",
       "      <td>0</td>\n",
       "      <td>other</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>HR</td>\n",
       "      <td>region_2</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>f</td>\n",
       "      <td>1</td>\n",
       "      <td>56</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>46</td>\n",
       "      <td>0</td>\n",
       "      <td>other</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>HR</td>\n",
       "      <td>region_7</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>f</td>\n",
       "      <td>2</td>\n",
       "      <td>30</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>6</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>46</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_7</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>2</td>\n",
       "      <td>31</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>6</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>48</td>\n",
       "      <td>0</td>\n",
       "      <td>other</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Finance</td>\n",
       "      <td>region_2</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>m</td>\n",
       "      <td>2</td>\n",
       "      <td>47</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>7</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>56</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72892</th>\n",
       "      <td>Technology</td>\n",
       "      <td>region_2</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>m</td>\n",
       "      <td>1</td>\n",
       "      <td>40</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>13</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>81</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72893</th>\n",
       "      <td>Operations</td>\n",
       "      <td>region_22</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>m</td>\n",
       "      <td>1</td>\n",
       "      <td>37</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>7</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>61</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72894</th>\n",
       "      <td>Operations</td>\n",
       "      <td>region_1</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>m</td>\n",
       "      <td>1</td>\n",
       "      <td>30</td>\n",
       "      <td>3.254223</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>56</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72895</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_22</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>m</td>\n",
       "      <td>1</td>\n",
       "      <td>45</td>\n",
       "      <td>4.595457</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>55</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72896</th>\n",
       "      <td>Sales &amp; Marketing</td>\n",
       "      <td>region_7</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>m</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>3.000000</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>48</td>\n",
       "      <td>0</td>\n",
       "      <td>sourcing</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>72897 rows × 13 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              department     region         education gender  no_of_trainings  \\\n",
       "0            Procurement   region_7        Bachelor's      f                1   \n",
       "1                     HR   region_2        Bachelor's      f                1   \n",
       "2                     HR   region_7        Bachelor's      f                2   \n",
       "3      Sales & Marketing   region_7        Bachelor's      m                2   \n",
       "4                Finance   region_2        Bachelor's      m                2   \n",
       "...                  ...        ...               ...    ...              ...   \n",
       "72892         Technology   region_2  Master's & above      m                1   \n",
       "72893         Operations  region_22  Master's & above      m                1   \n",
       "72894         Operations   region_1  Master's & above      m                1   \n",
       "72895  Sales & Marketing  region_22  Master's & above      m                1   \n",
       "72896  Sales & Marketing   region_7  Master's & above      m                1   \n",
       "\n",
       "       age  previous_year_rating  length_of_service  KPIs_met >80%  \\\n",
       "0       29              3.000000                  3              0   \n",
       "1       56              3.000000                  5              0   \n",
       "2       30              4.000000                  6              0   \n",
       "3       31              4.000000                  6              0   \n",
       "4       47              3.000000                  7              1   \n",
       "...    ...                   ...                ...            ...   \n",
       "72892   40              4.000000                 13              1   \n",
       "72893   37              3.000000                  7              0   \n",
       "72894   30              3.254223                  5              0   \n",
       "72895   45              4.595457                 16              0   \n",
       "72896   28              3.000000                  3              0   \n",
       "\n",
       "       awards_won?  avg_training_score  is_promoted recruitment_channel  \n",
       "0                0                  74            0               other  \n",
       "1                0                  46            0               other  \n",
       "2                0                  46            0            sourcing  \n",
       "3                0                  48            0               other  \n",
       "4                0                  56            0            sourcing  \n",
       "...            ...                 ...          ...                 ...  \n",
       "72892            0                  81            0            sourcing  \n",
       "72893            0                  61            0            sourcing  \n",
       "72894            0                  56            0            sourcing  \n",
       "72895            0                  55            0            sourcing  \n",
       "72896            0                  48            0            sourcing  \n",
       "\n",
       "[72897 rows x 13 columns]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df2 = train_df.copy()\n",
    "\n",
    "rebalance = dp.Rebalance(k_neighbors=4, verbose=False)\n",
    "cht_manager = CohortManager(transform_pipe=rebalance, cohort_col=['education'])\n",
    "smote_df = cht_manager.fit_resample(df=train_df2, rebalance_col='recruitment_channel')\n",
    "\n",
    "smote_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "43746\n",
      "72897\n"
     ]
    }
   ],
   "source": [
    "print(x_train.shape[0])\n",
    "print(smote_df.shape[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After applying the SMOTE Rebalancing mitigation, we get a dataset that is much larger than the original data frame. The SMOTE Method applies oversampling in order to balance the dataset. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### New Model on Rebalanced Datasets\n",
    "After applying the SMOTE Method on the data, we can then train a new lightGBM model on this newly balanced data and see if there are differences in model performance based on this balancing. We compare the results below and find that the new model trained on the data post rebalancing does a better job predicting true positives than the original model and thus has greater recall and overall precision. So not only does data rebalancing help with making sure a model is less biased, it also helps the model actually fit and be able to predict the data outcomes more accurately. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train2 = smote_df.drop(columns=[label_col])\n",
    "y_train2 = smote_df[label_col]\n",
    "\n",
    "pipeline2 = Pipeline([\n",
    "                (\"ohe\", dp.EncoderOHE(col_encode=categorical_cols, drop=False)),\n",
    "                (\"model\", LGBMClassifier(n_estimators=50, random_state=42)),\n",
    "            ])\n",
    "\n",
    "pipeline2.fit(x_train2, y_train2)\n",
    "\n",
    "pred2 = pipeline2.predict(x_test)\n",
    "pred_model1 = pipeline.predict(x_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We compare the number of error that the model 1 that was trained before rebalancing and model 2 that was trained after rebalancing have and we find that overall there are less errors with model 2. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\u001b[95m\u001b[1mBEFORE: \u001b[0mnumber of test dataset instances: \u001b[1m\u001b[92m4861\u001b[0m\n",
      "      : number of errors on test dataset: \u001b[1m\u001b[91m293\u001b[0m\n",
      "\n",
      "\u001b[95m\u001b[1mAFTER:  \u001b[0mnumber of test dataset instances: \u001b[1m\u001b[92m4861\u001b[0m\n",
      "     :  number of errors on test dataset: \u001b[1m\u001b[91m299\u001b[0m\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Compare Results\n",
    "class color:\n",
    "    PURPLE = \"\\033[95m\"\n",
    "    CYAN = \"\\033[96m\"\n",
    "    DARKCYAN = \"\\033[36m\"\n",
    "    BLUE = \"\\033[94m\"\n",
    "    GREEN = \"\\033[92m\"\n",
    "    YELLOW = \"\\033[93m\"\n",
    "    RED = \"\\033[91m\"\n",
    "    BOLD = \"\\033[1m\"\n",
    "    UNDERLINE = \"\\033[4m\"\n",
    "    END = \"\\033[0m\"\n",
    "\n",
    "\n",
    "print(\"\")\n",
    "print(\n",
    "    color.PURPLE\n",
    "    + color.BOLD\n",
    "    + \"BEFORE: \"\n",
    "    + color.END\n",
    "    + \"number of test dataset instances: \"\n",
    "    + color.BOLD\n",
    "    + color.GREEN\n",
    "    + str(len(y_test))\n",
    "    + color.END\n",
    ")\n",
    "print(\n",
    "    \"      : number of errors on test dataset: \"\n",
    "    + color.BOLD\n",
    "    + color.RED\n",
    "    + str(sum(pred_model1 != y_test))\n",
    "    + color.END\n",
    ")\n",
    "print(\"\")\n",
    "print(\n",
    "    color.PURPLE\n",
    "    + color.BOLD\n",
    "    + \"AFTER:  \"\n",
    "    + color.END\n",
    "    + \"number of test dataset instances: \"\n",
    "    + color.BOLD\n",
    "    + color.GREEN\n",
    "    + str(len(y_test))\n",
    "    + color.END\n",
    ")\n",
    "print(\n",
    "    \"     :  number of errors on test dataset: \"\n",
    "    + color.BOLD\n",
    "    + color.RED\n",
    "    + str(sum(pred2 != y_test))\n",
    "    + color.END\n",
    ")\n",
    "print(\"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-----------------------------------------------------------------------\n",
      "\n",
      "\u001b[94m\u001b[1mBEFORE: conf_matrix:\u001b[0m\n",
      "--------------------\n",
      "                  Pred 1              Pred 0\n",
      "True   TP = 142 (33.57%)   FN = 281 (66.43%)\n",
      "False    FP = 12 (0.27%)  TN = 4426 (99.73%)\n",
      "\n",
      "\u001b[94m\u001b[1mAFTER: conf_matrix:\u001b[0m\n",
      "-------------------\n",
      "                  Pred 1              Pred 0\n",
      "True   TP = 136 (32.15%)   FN = 287 (67.85%)\n",
      "False    FP = 12 (0.27%)  TN = 4426 (99.73%)\n",
      "-----------------------------------------------------------------------\n",
      "-----------------------------------------------------------------------\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# compare conf matrices\n",
    "print(\"-----------------------------------------------------------------------\")\n",
    "print(\"\")\n",
    "print(color.BLUE + color.BOLD + \"BEFORE: conf_matrix:\" + color.END)\n",
    "print(\"--------------------\")\n",
    "print(conf_matrix(y_test, pred_model1))\n",
    "print(\"\")\n",
    "print(color.BLUE + color.BOLD + \"AFTER: conf_matrix:\" + color.END)\n",
    "print(\"-------------------\")\n",
    "print(conf_matrix(y_test, pred2))\n",
    "print(\"-----------------------------------------------------------------------\")\n",
    "print(\"-----------------------------------------------------------------------\")\n",
    "print(\"\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[93m\u001b[1mBEFORE: classification_report:\u001b[0m\n",
      "--------------------------------\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.94      1.00      0.97      4438\n",
      "           1       0.92      0.34      0.49       423\n",
      "\n",
      "    accuracy                           0.94      4861\n",
      "   macro avg       0.93      0.67      0.73      4861\n",
      "weighted avg       0.94      0.94      0.93      4861\n",
      "\n",
      "\u001b[93m\u001b[1mAFTER: classification_report:\u001b[0m\n",
      "--------------------------------\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.94      1.00      0.97      4438\n",
      "           1       0.92      0.32      0.48       423\n",
      "\n",
      "    accuracy                           0.94      4861\n",
      "   macro avg       0.93      0.66      0.72      4861\n",
      "weighted avg       0.94      0.94      0.92      4861\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# compare classification report\n",
    "print(color.YELLOW + color.BOLD + \"BEFORE: classification_report:\" + color.END)\n",
    "print(\"--------------------------------\")\n",
    "print(classification_report(y_test, pred_model1))\n",
    "print(color.YELLOW + color.BOLD + \"AFTER: classification_report:\" + color.END)\n",
    "print(\"--------------------------------\")\n",
    "print(classification_report(y_test, pred2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, we didn't manage to improve the performance over the test dataset after rebalancing it, but we can still check to see if we were able to improve the data balance metrics. Now we analyze the data balance analysis on the newly rebalanced data and see what the difference is from before applying SMOTE and after.\n",
    "\n",
    "The feature value measures before rebalancing don't indicate a lot of discrepancy in the outcome within specific features within a class since the values such as demographic parity which is on a 0 to 1 scale are very close to zero. After rebalancing, we still have similarly low values of these measures but there is not a significant improvement since they already started low."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "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>ClassA</th>\n",
       "      <th>ClassB</th>\n",
       "      <th>FeatureName</th>\n",
       "      <th>dp</th>\n",
       "      <th>pmi</th>\n",
       "      <th>sdc</th>\n",
       "      <th>ji</th>\n",
       "      <th>krc</th>\n",
       "      <th>llr</th>\n",
       "      <th>t_test</th>\n",
       "      <th>ttest_pvalue</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>education</td>\n",
       "      <td>0.017814</td>\n",
       "      <td>0.197252</td>\n",
       "      <td>0.004788</td>\n",
       "      <td>0.005594</td>\n",
       "      <td>4.996442</td>\n",
       "      <td>-0.622067</td>\n",
       "      <td>0.262386</td>\n",
       "      <td>0.408789</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>Below Secondary</td>\n",
       "      <td>education</td>\n",
       "      <td>0.022587</td>\n",
       "      <td>0.257448</td>\n",
       "      <td>0.069726</td>\n",
       "      <td>0.076144</td>\n",
       "      <td>-27.035372</td>\n",
       "      <td>3.718893</td>\n",
       "      <td>-2.623174</td>\n",
       "      <td>0.059886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>Below Secondary</td>\n",
       "      <td>education</td>\n",
       "      <td>0.004773</td>\n",
       "      <td>0.060196</td>\n",
       "      <td>0.064938</td>\n",
       "      <td>0.070550</td>\n",
       "      <td>-32.031814</td>\n",
       "      <td>4.340960</td>\n",
       "      <td>-2.885561</td>\n",
       "      <td>0.051023</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>sourcing</td>\n",
       "      <td>other</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.002630</td>\n",
       "      <td>0.030454</td>\n",
       "      <td>-0.000794</td>\n",
       "      <td>-0.000925</td>\n",
       "      <td>1.579734</td>\n",
       "      <td>-0.240590</td>\n",
       "      <td>0.085323</td>\n",
       "      <td>0.469888</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>sourcing</td>\n",
       "      <td>referred</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>-0.038804</td>\n",
       "      <td>-0.366366</td>\n",
       "      <td>0.048196</td>\n",
       "      <td>0.053285</td>\n",
       "      <td>-18.723860</td>\n",
       "      <td>2.639673</td>\n",
       "      <td>-1.733789</td>\n",
       "      <td>0.112546</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>other</td>\n",
       "      <td>referred</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>-0.041434</td>\n",
       "      <td>-0.396820</td>\n",
       "      <td>0.048990</td>\n",
       "      <td>0.054210</td>\n",
       "      <td>-20.303594</td>\n",
       "      <td>2.880263</td>\n",
       "      <td>-1.819112</td>\n",
       "      <td>0.105255</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "             ClassA           ClassB          FeatureName        dp       pmi  \\\n",
       "0  Master's & above       Bachelor's            education  0.017814  0.197252   \n",
       "1  Master's & above  Below Secondary            education  0.022587  0.257448   \n",
       "2        Bachelor's  Below Secondary            education  0.004773  0.060196   \n",
       "0          sourcing            other  recruitment_channel  0.002630  0.030454   \n",
       "1          sourcing         referred  recruitment_channel -0.038804 -0.366366   \n",
       "2             other         referred  recruitment_channel -0.041434 -0.396820   \n",
       "\n",
       "        sdc        ji        krc       llr    t_test  ttest_pvalue  \n",
       "0  0.004788  0.005594   4.996442 -0.622067  0.262386      0.408789  \n",
       "1  0.069726  0.076144 -27.035372  3.718893 -2.623174      0.059886  \n",
       "2  0.064938  0.070550 -32.031814  4.340960 -2.885561      0.051023  \n",
       "0 -0.000794 -0.000925   1.579734 -0.240590  0.085323      0.469888  \n",
       "1  0.048196  0.053285 -18.723860  2.639673 -1.733789      0.112546  \n",
       "2  0.048990  0.054210 -20.303594  2.880263 -1.819112      0.105255  "
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "feat_measures1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "scrolled": true
   },
   "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>ClassA</th>\n",
       "      <th>ClassB</th>\n",
       "      <th>FeatureName</th>\n",
       "      <th>dp</th>\n",
       "      <th>pmi</th>\n",
       "      <th>sdc</th>\n",
       "      <th>ji</th>\n",
       "      <th>krc</th>\n",
       "      <th>llr</th>\n",
       "      <th>t_test</th>\n",
       "      <th>ttest_pvalue</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>Below Secondary</td>\n",
       "      <td>education</td>\n",
       "      <td>0.007317</td>\n",
       "      <td>0.127051</td>\n",
       "      <td>0.049504</td>\n",
       "      <td>0.052789</td>\n",
       "      <td>-39.954385</td>\n",
       "      <td>4.470495</td>\n",
       "      <td>-2.572282</td>\n",
       "      <td>0.061853</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Bachelor's</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>education</td>\n",
       "      <td>-0.010460</td>\n",
       "      <td>-0.157473</td>\n",
       "      <td>-0.003277</td>\n",
       "      <td>-0.003691</td>\n",
       "      <td>-5.858737</td>\n",
       "      <td>0.644341</td>\n",
       "      <td>-0.220763</td>\n",
       "      <td>0.422882</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Below Secondary</td>\n",
       "      <td>Master's &amp; above</td>\n",
       "      <td>education</td>\n",
       "      <td>-0.017777</td>\n",
       "      <td>-0.284524</td>\n",
       "      <td>-0.052781</td>\n",
       "      <td>-0.056479</td>\n",
       "      <td>34.095648</td>\n",
       "      <td>-3.826154</td>\n",
       "      <td>2.351520</td>\n",
       "      <td>0.071519</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>other</td>\n",
       "      <td>sourcing</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.011770</td>\n",
       "      <td>0.148923</td>\n",
       "      <td>0.009862</td>\n",
       "      <td>0.011314</td>\n",
       "      <td>0.067773</td>\n",
       "      <td>0.148923</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>other</td>\n",
       "      <td>referred</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.050002</td>\n",
       "      <td>0.886267</td>\n",
       "      <td>0.041898</td>\n",
       "      <td>0.046479</td>\n",
       "      <td>0.287918</td>\n",
       "      <td>0.886267</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>sourcing</td>\n",
       "      <td>referred</td>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.038232</td>\n",
       "      <td>0.737344</td>\n",
       "      <td>0.032036</td>\n",
       "      <td>0.035165</td>\n",
       "      <td>0.220145</td>\n",
       "      <td>0.737344</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            ClassA            ClassB          FeatureName        dp       pmi  \\\n",
       "0       Bachelor's   Below Secondary            education  0.007317  0.127051   \n",
       "1       Bachelor's  Master's & above            education -0.010460 -0.157473   \n",
       "2  Below Secondary  Master's & above            education -0.017777 -0.284524   \n",
       "0            other          sourcing  recruitment_channel  0.011770  0.148923   \n",
       "1            other          referred  recruitment_channel  0.050002  0.886267   \n",
       "2         sourcing          referred  recruitment_channel  0.038232  0.737344   \n",
       "\n",
       "        sdc        ji        krc       llr    t_test  ttest_pvalue  \n",
       "0  0.049504  0.052789 -39.954385  4.470495 -2.572282      0.061853  \n",
       "1 -0.003277 -0.003691  -5.858737  0.644341 -0.220763      0.422882  \n",
       "2 -0.052781 -0.056479  34.095648 -3.826154  2.351520      0.071519  \n",
       "0  0.009862  0.011314   0.067773  0.148923  0.000000      0.500000  \n",
       "1  0.041898  0.046479   0.287918  0.886267  0.000000      0.500000  \n",
       "2  0.032036  0.035165   0.220145  0.737344  0.000000      0.500000  "
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "feature_measures.measures(smote_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When we compare the distribution measures before and after rebalancing, we find that the data is much more evenly distributed (close to the uniform distribution) for the two columns of interest after rebalancing the data using the SMOTE algorithm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "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>FeatureName</th>\n",
       "      <th>kl_divergence</th>\n",
       "      <th>js_dist</th>\n",
       "      <th>wasserstein_dist</th>\n",
       "      <th>inf_norm_dist</th>\n",
       "      <th>total_variation_dist</th>\n",
       "      <th>chi_sq_p_value</th>\n",
       "      <th>chi_sq_stat</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>education</td>\n",
       "      <td>0.434946</td>\n",
       "      <td>0.358906</td>\n",
       "      <td>0.236105</td>\n",
       "      <td>0.354158</td>\n",
       "      <td>0.354158</td>\n",
       "      <td>0.0</td>\n",
       "      <td>30343.542312</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.327144</td>\n",
       "      <td>0.315774</td>\n",
       "      <td>0.208248</td>\n",
       "      <td>0.312371</td>\n",
       "      <td>0.312371</td>\n",
       "      <td>0.0</td>\n",
       "      <td>20349.678919</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           FeatureName  kl_divergence   js_dist  wasserstein_dist  \\\n",
       "0            education       0.434946  0.358906          0.236105   \n",
       "1  recruitment_channel       0.327144  0.315774          0.208248   \n",
       "\n",
       "   inf_norm_dist  total_variation_dist  chi_sq_p_value   chi_sq_stat  \n",
       "0       0.354158              0.354158             0.0  30343.542312  \n",
       "1       0.312371              0.312371             0.0  20349.678919  "
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# before\n",
    "dist_measures1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "scrolled": true
   },
   "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>FeatureName</th>\n",
       "      <th>kl_divergence</th>\n",
       "      <th>js_dist</th>\n",
       "      <th>wasserstein_dist</th>\n",
       "      <th>inf_norm_dist</th>\n",
       "      <th>total_variation_dist</th>\n",
       "      <th>chi_sq_p_value</th>\n",
       "      <th>chi_sq_stat</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>education</td>\n",
       "      <td>0.43447</td>\n",
       "      <td>0.35937</td>\n",
       "      <td>0.233928</td>\n",
       "      <td>0.350892</td>\n",
       "      <td>0.350892</td>\n",
       "      <td>0.0</td>\n",
       "      <td>50099.684185</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>recruitment_channel</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           FeatureName  kl_divergence  js_dist  wasserstein_dist  \\\n",
       "0            education        0.43447  0.35937          0.233928   \n",
       "1  recruitment_channel        0.00000  0.00000          0.000000   \n",
       "\n",
       "   inf_norm_dist  total_variation_dist  chi_sq_p_value   chi_sq_stat  \n",
       "0       0.350892              0.350892             0.0  50099.684185  \n",
       "1       0.000000              0.000000             1.0      0.000000  "
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# after\n",
    "dist_measures.measures(smote_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The Atkinson index which gives us the overall notion of inequality before and after rebalancing shows us that in order to get a perfectly balanced dataset, we no longer need to forgo any of the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>theil_l_index</th>\n",
       "      <th>theil_t_index</th>\n",
       "      <th>atkinson_index</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.600814</td>\n",
       "      <td>0.762696</td>\n",
       "      <td>0.798268</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   theil_l_index  theil_t_index  atkinson_index\n",
       "0       1.600814       0.762696        0.798268"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# before\n",
    "agg_measures1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "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>theil_l_index</th>\n",
       "      <th>theil_t_index</th>\n",
       "      <th>atkinson_index</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.995942</td>\n",
       "      <td>0.43447</td>\n",
       "      <td>0.630625</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   theil_l_index  theil_t_index  atkinson_index\n",
       "0       0.995942        0.43447        0.630625"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# after\n",
    "agg_measures.measures(smote_df)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.9.13 ('raipub')",
   "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": "3.9.13"
  },
  "vscode": {
   "interpreter": {
    "hash": "98cf402749abf383affb54f23cdde06b52ae2a6e4394659b91d1cafca4224ff4"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}