intro_to_ai_ethics/04-ai-fairness.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "ccc9bed1",
   "metadata": {
    "papermill": {
     "duration": 0.02183,
     "end_time": "2022-03-06T22:14:51.882445",
     "exception": false,
     "start_time": "2022-03-06T22:14:51.860615",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "**This notebook is an exercise in the [AI Ethics](https://www.kaggle.com/learn/ai-ethics) course.  You can reference the tutorial at [this link](https://www.kaggle.com/alexisbcook/ai-fairness).**\n",
    "\n",
    "---\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b76304ee",
   "metadata": {
    "papermill": {
     "duration": 0.020721,
     "end_time": "2022-03-06T22:14:51.924049",
     "exception": false,
     "start_time": "2022-03-06T22:14:51.903328",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "In the tutorial, you learned about different ways of measuring fairness of a machine learning model.  In this exercise, you'll train a few models to approve (or deny) credit card applications and analyze fairness.  Don't worry if you're new to coding: this exercise assumes no programming knowledge.\n",
    "\n",
    "# Introduction\n",
    "\n",
    "We work with a **synthetic** dataset of information submitted by credit card applicants. \n",
    "\n",
    "To load and preview the data, run the next code cell.  When the code finishes running, you should see a message saying the data was successfully loaded, along with a preview of the first five rows of the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "6a70a255",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:51.969945Z",
     "iopub.status.busy": "2022-03-06T22:14:51.968806Z",
     "iopub.status.idle": "2022-03-06T22:14:53.579263Z",
     "shell.execute_reply": "2022-03-06T22:14:53.579795Z",
     "shell.execute_reply.started": "2022-03-06T21:51:33.905342Z"
    },
    "papermill": {
     "duration": 1.63535,
     "end_time": "2022-03-06T22:14:53.580057",
     "exception": false,
     "start_time": "2022-03-06T22:14:51.944707",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data successfully loaded!\n",
      "\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>Num_Children</th>\n",
       "      <th>Group</th>\n",
       "      <th>Income</th>\n",
       "      <th>Own_Car</th>\n",
       "      <th>Own_Housing</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>288363</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>40690</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>64982</th>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>75469</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>227641</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>70497</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>137672</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>61000</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12758</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>56666</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Num_Children  Group  Income  Own_Car  Own_Housing\n",
       "288363             1      1   40690        0            1\n",
       "64982              2      0   75469        1            0\n",
       "227641             1      1   70497        1            1\n",
       "137672             1      1   61000        0            0\n",
       "12758              1      1   56666        1            1"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Set up feedback system\n",
    "from learntools.core import binder\n",
    "binder.bind(globals())\n",
    "from learntools.ethics.ex4 import *\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "# Load the data, separate features from target\n",
    "data = pd.read_csv(\"../input/synthetic-credit-card-approval/synthetic_credit_card_approval.csv\")\n",
    "X = data.drop([\"Target\"], axis=1)\n",
    "y = data[\"Target\"]\n",
    "\n",
    "# Break into training and test sets\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.8, test_size=0.2, random_state=0)\n",
    "\n",
    "# Preview the data\n",
    "print(\"Data successfully loaded!\\n\")\n",
    "X_train.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aa7f22ca",
   "metadata": {
    "papermill": {
     "duration": 0.0218,
     "end_time": "2022-03-06T22:14:53.624644",
     "exception": false,
     "start_time": "2022-03-06T22:14:53.602844",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    " The dataset contains, for each applicant:\n",
    "- income (in the `Income` column),\n",
    "- the number of children (in the `Num_Children` column),\n",
    "- whether the applicant owns a car (in the `Own_Car` column, the value is `1` if the applicant owns a car, and is else `0`), and\n",
    "- whether the applicant owns a home (in the `Own_Housing` column, the value is `1` if the applicant owns a home, and is else `0`)\n",
    "\n",
    "When evaluating fairness, we'll check how the model performs for users in different groups, as identified by the `Group` column: \n",
    "- The `Group` column breaks the users into two groups (where each group corresponds to either `0` or `1`).  \n",
    "- For instance, you can think of the column as breaking the users into two different races, ethnicities, or gender groupings.  If the column breaks users into different ethnicities, `0` could correspond to a non-Hispanic user, while `1` corresponds to a Hispanic user. \n",
    "\n",
    "\n",
    "Run the next code cell without changes to train a simple model to approve or deny individuals for a credit card.  The output shows the performance of the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "e22b451b",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:53.671919Z",
     "iopub.status.busy": "2022-03-06T22:14:53.671221Z",
     "iopub.status.idle": "2022-03-06T22:14:54.786876Z",
     "shell.execute_reply": "2022-03-06T22:14:54.786348Z",
     "shell.execute_reply.started": "2022-03-06T21:52:51.333919Z"
    },
    "papermill": {
     "duration": 1.140167,
     "end_time": "2022-03-06T22:14:54.787013",
     "exception": false,
     "start_time": "2022-03-06T22:14:53.646846",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total approvals: 38246\n",
      "Group A: 8028 (20.99% of approvals)\n",
      "Group B: 30218 (79.01% of approvals)\n",
      "\n",
      "Overall accuracy: 94.79%\n",
      "Group A: 94.56%\n",
      "Group B: 95.02%\n",
      "\n",
      "Sensitivity / True positive rate:\n",
      "Group A: 77.23%\n",
      "Group B: 98.03%\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWgAAAEWCAYAAABLzQ1kAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAArLklEQVR4nO3deZwVxbnG8d8zM2yyyI4EUVGJuKPiFo0XNwRNgnq9iomCxoiJokbNoiY3Rs2iN4vGGE00IhIX4hqJokgUNJqIgCKKK8EFEEVElE0QeO8fXYMHnOUMzHLOzPPNpz/T5+3q7urBvFOnurpaEYGZmRWekoaugJmZVcwJ2sysQDlBm5kVKCdoM7MC5QRtZlagnKDNzAqUE7SZWYFygrZ6JWmIpMmSlklakNbPlKSGrls5SadICkknNHRdrGlzgrZ6I+kC4HfAr4AtgG7At4EDgOaV7FNabxX8zDBgETC0Ac5tto4TtNULSZsDlwFnRsTdEbEkMs9FxDciYmUqN0rS9ZLGSVoGHCxpR0mTJC2WNFPS13KOO0nSt3I+nyLpyZzPIekcSbMlLZT0K0mV/ncvaWvgv4DhwBGStqj934ZZfpygrb7sD7QA7s+j7NeBnwNtgcnA34FHgK7A2cBtknaowbmPAfoBewKDgW9WUXYoMDUi7gFeBr5Rg/OY1SonaKsvnYGFEbG6PCDpX6lVvELSQTll74+IpyJiLdAXaANcERGrIuIx4AHgxBqc+8qIWBQRbwNXV7PvUOD2tH477uawBuQEbfXlA6CzpLLyQER8KSLap225/y3OyVn/AjAnJetybwE9anDu3OO9lY75OZIOAHoBY1LodmBXSX1rcC6zWuMEbfXl38BKsi6G6uROsfgO0HODfuOtgHlpfRmwWc62ivqMe26w7zuVnHcYIGC6pHfJulfK42b1zgna6kVELAYuBa6TdJyktpJKUuu0dRW7TgaWAz+Q1ExSf+CrfNbKnQ4cK2kzSdsDp1VwjO9L6iCpJ3Au8NcNC0hqCRxPdnOwb85yNvD13Ja/WX1xgrZ6ExH/B5wP/AB4Ly1/An4I/KuSfVaRJeRBwELgOmBoRLySilwFrErHugW4rYLD3A9MI0vmDwI3VVDmaGAFMDoi3i1fgJFAGTCwZldrtunkCfutMZMUQO+ImNXQdTGrKbegzcwKlBO0mVmBcheHmVmBcgvazKxAeehQBVTWKtS8bUNXw2pgjx23augqWA09++y0hRHRZVOOUdpu64jVK/IqGyveHx8RRTUaxwm6AmrelhY7HN/Q1bAaeGrytQ1dBauhVs301qYeI1Z/Qos+Q/Iq+8lzv++8qeerb07QZla8BBTOVOK1zgnazIpb5bPHFj0naDMrbm5Bm5kVIkFJQ7x0p344QZtZ8RLu4jAzK0xyF4eZWcFyC9rMrEC5BW1mVojkFrSZWUESHsVhZlaY3II2MytcJe6DNjMrPB4HbWZWwDyKw8ysEPlRbzOzwuUuDjOzAiQ/6m1mVrjcgjYzK1BuQZuZFSI/qGJmVpga+aPejfdPj5k1AakFnc9S1VGklpKekfS8pJmSLk3xUZLekDQ9LX1TXJKukTRL0gxJe+Yca5ik19MyLCe+l6QX0j7XSNX3zbgFbWbFrXb6oFcCh0TEUknNgCclPZS2fT8i7t6g/CCgd1r2Ba4H9pXUEbgE6AcEME3S2Ij4MJU5HZgMjAMGAg9RBbegzay41UILOjJL08dmaYkqdhkMjE77PQ20l9QdOAKYEBGLUlKeAAxM29pFxNMREcBo4OjqLs0J2syKW/lY6OqWag+jUknTgQVkSXZy2vTz1I1xlaQWKdYDmJOz+9wUqyo+t4J4lZygzax4qUZ90J0lTc1ZhuceKiLWRERfYEtgH0m7ABcBfYC9gY7AD+vz8twHbWZFTSV5tzMXRkS/6gpFxGJJE4GBEfHrFF4p6Wbge+nzPKBnzm5bptg8oP8G8UkpvmUF5avkFrSZFS0BkvJaqjyO1EVS+7TeCjgceCX1HZNGXBwNvJh2GQsMTaM59gM+ioj5wHhggKQOkjoAA4DxadvHkvZLxxoK3F/d9bkFbWbFS2nZdN2BWySVkjVc74yIByQ9JqlLOst04Nup/DjgSGAWsBw4FSAiFkm6HJiSyl0WEYvS+pnAKKAV2eiNKkdwgBO0mRW16lvH+YiIGcAeFcQPqaR8AGdVsm0kMLKC+FRgl5rUywnazIpabSToQuUEbWZFrST/m4RFxwnazIpX7fVBFyQnaDMrWqqlPuhC5QRtZkXNCdrMrEA5QZuZFSgnaDOzQiRQiRO0mVnB8U1CM7MC5gRtZlaoGm9+doI2syImt6DNzAqWE7SZWQES8lwcZmYFq/E2oJ2gzayIuQ/azKxwOUGbmRUoJ2gzswLlR72toLRoXsaDN3yXFs3KKC0rZeyjz3HFDeP4cr8vcvm5x9C8WSnTX57D2T+7jTVr1nL2SYfyP4P2BqCstIQvbrMF2w+4kNatmnP9T4fSpWNbArjlvqf405hJAFz87aM48qDdWBvB+4uWcNalt/Luwo8a7qIbsd2+9hPabNaC0pISyspKmDj6h3z40TK+efFI3p6/iK26d+TmX55G+3abERFc+Ju7mfDUTFq1bM51l5zM7n16NvQlNJh83tid53FaAk8ALcjy4t0RcYmkXsAYoBMwDTg5IlZJagGMBvYCPgBOiIg307EuAk4D1gDnRMT4FB8I/A4oBf4cEVdUW6/s3YeFRdIa4AWgGbCa7BdxVUSsrY/zl2zWNVrscHx9nGqjtW7VnGUrVlFWWsJDfz6fH111DyN/8U0Gn/l7/vP2Ai464yjmzF/ErWP/vd5+A7+8C9858WAGn/l7unVqR7fO7Zjx6lzabNaCiaN/yEnfv4FX33iXtq1bsmTZJwAMP+G/6NOrO+dfMaYhLjUvH065tqGrsNF2+9pPmDj6B3Rq32Zd7CfX/I0O7TbjvFMGcNWoR1i8ZDmXnn00jzw1kxv++jh3/e47TH3xTS76zd38Y9T3G7D2G69VM02LiH6bcowW3XpH9yFX5VX2rWu+Wun5lGX51hGxVFIz4EngXOB84N6IGCPpj8DzEXG9pDOB3SLi25KGAMdExAmSdgLuAPYBvgD8A/hiOs1rwOHAXLK3fp8YES9VVedCHUC4IiL6RsTOZBc0CLikgetUUJatWAVAs7JSmpWVsmbNWlZ9upr/vL0AgEmTX+Frh/T93H7/PaAf9zwyDYD3PviYGa/OBWDp8pW89ua7dO/SHmBdcgZo3aoFhfiHvDF76PEZnPiVfQE48Sv7Mm7SDADGPT6DIUftgyT23rUXHy1Z0eS/2ZS3oqtbqhKZpeljs7QEcAhwd4rfAhyd1genz6Tth6YkPxgYExErI+INYBZZst4HmBURsyNiFVmrfHB111aoCXqdiFgADAdGKFMq6VeSpkiaIekMAEn9JU2SdLekVyTdlsofIulv5ceTdLik+xrocmpNSYl44rYLee2RK5g0+RWmzXyLstJS+u64FQBfO7QvPbp1WG+fVi2acej+OzL2semfO17P7h3ZbYctmTbzzXWxH3/nq7z4wOX8z8B+/OJPD9bl5TRpkjh2xLX0P/lKRt37JAALFi1hi86bA9CtUzsWLFoCwPz3F6/37/qFru2Zv2Bxvde5oCjPBTpLmpqzDF/vMFlumQ4sACYA/wEWR8TqVGQu0COt9wDmAKTtH5F1g6yLb7BPZfEqFUUfdETMllQKdCX7q/NRROyd+oGekvRIKroHsDPwDvAUcAAwEbhOUpeIeB84FRi54TnSP1b2D9aszYabC87atcFB37iCdm1aceuvTmfH7bpz2o9u5hfnHUvz5mVMfPoV1qxdv0do4EG7MnnGbBZ/vHy9eOtWzRl95be46Lf3rNdy/tn1f+dn1/+d804ZwOnHH8QVN4yrl2trah668Ty+0LU97y9awjEjrqX3Nlustz1rATZQ5YpADfqgF1bVpRIRa4C+ktoD9wF9Nr12m6bgW9AVGAAMTX/pJpP91eqdtj0TEXNTX/V0YJvIvpv/BTgp/eL3Bx7a8KARcUNE9IuIfiprVfdXUUs+XrqCf057jUP334kpL7zBkcOv5rBTfs2/npvFrLcWrFf22MP34p7x09aLlZWWcMuVp3PXw1N5YOLzFZ7jroemVNhdYrXjC13bA9ClY1u+0n83np35Jl07tl3XdfHuwo/o0qEtAN27tGfeex+u2/edBYvpnvZviqTs22Q+S74iYjFZw25/oL2k8obslsC8tD4P6JnVQWXA5mQ3C9fFN9insniViiJBS9qW7I7oArIvK2enPuq+EdErIspb0CtzdlvDZ98QbgZOAk4E7sr5ylKUOrVvQ7s22R+Rli2acfA+fXj9zffo3CFr+TdvVsa5ww7n5vR1GaBd65YcsOf2jHt8xnrH+v3/foPX3nyX625/bL34tj27rFsf9F+78dqb79XV5TRpy1asXPetZdmKlTz29CvsuN0XGHjQrtzxwGQA7nhgMoP+azcABh20K2MefIaIYMoLb9CuTat1XSFNU379z9W1siV1SQ04JLUiu/f1MlmiPi4VGwbcn9bHps+k7Y+lxuBYYIikFmkESG/gGbKbgr0l9ZLUHBiSylap4Ls4JHUB/ghcGxEhaTzwHUmPRcSnkr5INX+JIuIdSe8APwYOq/ta160tOrfjup+eTGlJCSUl4r5/PMv4J1/ksnOOZsCBu1BSIkbe80/+OfW1dfscdfDuTJz8Css/WbUutt/u2zLkqH2Z+fo8nrjtQgAu/8NYJvzrJS4ZMZjeW3dl7dpgzruLOP+XhTuCo5i9/8ESTvrBjQCsWb2G/x7Yj8O+tBN77rQVp140klvH/pueW3Tk5l9+E4ABB+zMhKdmsucxl9KqZTP+8JOTGrL6BaGWun+6A7ekrtQS4M6IeEDSS8AYST8DngNuSuVvAv4iaRawiCzhEhEzJd0JvEQ2Au2s1HWCpBHAeLJhdiMjYma111aId+crGGb3F+C3EbFWUgnwM+CrZK3p98nurO4BfC8ivpKOcS0wNSJGpc9DgO9GxH7Vnb8YhtnZ+op5mF1TVRvD7Fpu8cXYetjv8yr72v8N3OTz1beCbEFHRGkV29YCF6cl16S0lJcbscH2A4Eba6eGZlYQVGst6IJUkAm6tkmaBiwDLmjouphZ7RHU6AZgsWkSCToi9mroOphZ3XCCNjMrRO7iMDMrTMLTjZqZFajamc2uUDlBm1lRa8T52QnazIqYfJPQzKwguQ/azKyANeL87ARtZsXNLWgzswLViPOzE7SZFTG5BW1mVpBEzSbjLzZO0GZW1BpxA9oJ2syKm7s4zMwKkSdLMjMrTH5QxcysgDXmBF0Ub/U2M6tMSYnyWqoiqaekiZJekjRT0rkp/lNJ8yRNT8uROftcJGmWpFclHZETH5hisyRdmBPvJWlyiv81vd276mvbqN+ImVkhSH3Q+SzVWA1cEBE7AfsBZ0naKW27KiL6pmUcQNo2BNgZGAhcJ6k0vRX8D8AgYCfgxJzjXJmOtT3wIXBadZVygjazoqU0H3Q+S1UiYn5EPJvWlwAvAz2q2GUwMCYiVkbEG8AsYJ+0zIqI2RGxChgDDFZWgUOAu9P+twBHV3d9TtBmVtRq0ILuLGlqzjK84uNpG2APYHIKjZA0Q9JISR1SrAcwJ2e3uSlWWbwTsDgiVm8Qr5JvEppZUSvJ/ybhwojoV1UBSW2Ae4DvRsTHkq4HLgci/fwN8M1NqG6NOEGbWdFSLU7YL6kZWXK+LSLuBYiI93K23wg8kD7OA3rm7L5lilFJ/AOgvaSy1IrOLV8pd3GYWVErUX5LVVIf8U3AyxHx25x495xixwAvpvWxwBBJLST1AnoDzwBTgN5pxEZzshuJYyMigInAcWn/YcD91V2bW9BmVtRqaRz0AcDJwAuSpqfYxWSjMPqSdXG8CZwBEBEzJd0JvEQ2AuSsiFiT6jMCGA+UAiMjYmY63g+BMZJ+BjxH9gehSpUmaEm/T5WqUEScU93BzczqWm3k54h4kuzBxA2Nq2KfnwM/ryA+rqL9ImI22SiPvFXVgp5akwOZmdU3kQ21a6wqTdARcUvuZ0mbRcTyuq+SmVn+GvF00NXfJJS0v6SXgFfS590lXVfnNTMzq47ye8y7WCf1z2cUx9XAEWTDRIiI54GD6rBOZmZ5Edk46HyWYpTXKI6ImLPBndI1dVMdM7OaKdLcm5d8EvQcSV8CIg3kPpfsOXUzswbX1Kcb/TZwFtlz4+8AfdNnM7MGle88HMWaw6ttQUfEQuAb9VAXM7MaKy3W7JuHfEZxbCvp75Lel7RA0v2Stq2PypmZVac2phstVPl0cdwO3Al0B74A3AXcUZeVMjPLRzaKY9Pn4ihU+STozSLiLxGxOi23Ai3rumJmZtXKs/VcrC3oqubi6JhWH0rv1RpDNjfHCVTxfLqZWX0q0tybl6puEk4jS8jll39GzrYALqqrSpmZ5atYW8f5qGoujl71WREzs5oSUFqsHcx5yOtJQkm7kL2hdl3fc0SMrqtKmZnlq/Gm5zwStKRLgP5kCXoc2evEnwScoM2sQUk1eidh0clnFMdxwKHAuxFxKrA7sHmd1srMLE9N+klCYEVErJW0WlI7YAHrvxTRzKzBNMmbhDmmSmoP3Eg2smMp8O+6rJSZWb4acX6uvosjIs6MiMUR8UfgcGBY6uowM2tQkigtyW+p5jg9JU2U9JKkmZLOTfGOkiZIej397JDiknSNpFmSZkjaM+dYw1L51yUNy4nvJemFtM81yqPpX2mClrTnhgvQESjLrYyZWUOqpScJVwMXRMROwH7AWZJ2Ai4EHo2I3sCj6TNkgyV6p2U4cH2qS0fgEmBfshfEXlKe1FOZ03P2G1hdparq4vhNFdsCOKS6gxer3ftsxWNP/q6hq2E18J/3ljZ0FayB5DPSoToRMR+Yn9aXSHqZbIrlwWSj2ABuASYBP0zx0RERwNOS2kvqnspOiIhFAJImAAMlTQLaRcTTKT4aOBp4qKp6VfWgysEbcZ1mZvVG1OgmYWdJU3M+3xARN3zumNI2wB7AZKBbSt4A7wLd0noPYE7ObnNTrKr43AriVcrrQRUzs0JVgwcJF0ZEv6oKSGoD3AN8NyI+zk3+ERGSYmPruTFq49uBmVmDkKiVm4TZsdSMLDnfFhH3pvB7qeuC9HNBis9j/eHGW6ZYVfEtK4hXyQnazIpabcwHnUZU3AS8HBG/zdk0FigfiTEMuD8nPjSN5tgP+Ch1hYwHBkjqkG4ODgDGp20fS9ovnWtozrEqlc+j3iJ75dW2EXGZpK2ALSLimer2NTOra7U0DvoA4GTgBUnTU+xi4ArgTkmnAW8Bx6dt44AjgVnAcuBUgIhYJOlyYEoqd1n5DUPgTGAU0Irs5mCVNwghvz7o64C1ZKM2LgOWkH0N2DuPfc3M6kz2RpVNz9AR8SSVz7t0aAXlg0penh0RI4GRFcSnArvUpF75JOh9I2JPSc+lk3woqXlNTmJmVlcacz9tPgn6U0mlZGOfkdSFrEVtZtbgGvOj3vkk6GuA+4Cukn5ONrvdj+u0VmZmeSh/1LuxqjZBR8RtkqaR9cMIODoiXq7zmpmZ5aER5+e8RnFsRXaX8u+5sYh4uy4rZmZWndq6SVio8unieJDPXh7bEugFvArsXIf1MjPLSyPOz3l1ceya+znNZHdmndXIzCxfeTyEUsxqPBdHRDwrad+6qIyZWU2pEb82Np8+6PNzPpYAewLv1FmNzMzyJKCsEQ+EzqcF3TZnfTVZn/Q9dVMdM7OaabLvJEwPqLSNiO/VU33MzPKWjeJo6FrUnUoTtKSyiFgt6YD6rJCZWd7UdEdxPEPW3zxd0ljgLmBZ+cac+VLNzBpMUx8H3RL4gGw2u/Lx0AE4QZtZgxJQ2kRvEnZNIzhe5LPEXK5eX/tiZlYxUdJEh9mVAm2oeI5UJ2gza3DZS2MbuhZ1p6oEPT8iLqu3mpiZ1VQTfpKwEV+2mTUWjfkmYVXd6597zYuZWSEp7+LIZ6n2WNJISQskvZgT+6mkeZKmp+XInG0XSZol6VVJR+TEB6bYLEkX5sR7SZqc4n/N581UlSbonBcdmpkVrNIS5bXkYRQwsIL4VRHRNy3jACTtBAwhm9VzIHCdpNL0cN8fgEHATsCJqSzAlelY2wMfAqdVV6FGPEDFzBo7kSWxfJbqRMQTQL4N08HAmIhYGRFvkL3de5+0zIqI2RGxChgDDFb2PPohwN1p/1uAo6s7iRO0mRUvZXNx5LMAnSVNzVmG53mWEZJmpC6QDinWA5iTU2ZuilUW7wQsjojVG8Sr5ARtZkVNeS7Awojol7PckMfhrwe2A/oC84Hf1HL1q1Tj+aDNzApFXb/yKiLeW3cu6UbggfRxHtAzp+iWKUYl8Q+A9uVzHG1QvlJuQZtZUatBC7rmx5a653w8huzJaoCxwBBJLST1AnqTzV80BeidRmw0J7uRODYiApgIHJf2HwbcX9353YI2syImSmrpSRVJdwD9yfqq5wKXAP0l9SV7evpN4AyAiJgp6U7gJbJ58s+KiDXpOCOA8WRPY4+MiJnpFD8Exkj6GfAccFN1dXKCNrOiVT6KozZExIkVhCtNohHxc+DnFcTHAeMqiM8mG+WRNydoMytqTfaNKmZmha7xpmcnaDMrZnIL2sysIAkodYI2MytMjTc9O0GbWZFrxA1oJ2gzK17ZMLvGm6GdoM2sqLkFbWZWkITcgjYzKzwexWFmVqjyfJ1VsXKCNrOi5gRtZlag3AdtZlaAsgn7G7oWdccJ2syKWl2+UaWhOUGbWVFzF4cVrHnvfci5l9/K+x8uQYiTBu/Pt47vz2XX3s+Ep16kebNStu7Rmasu/jqbt92MRR8tY/iPRjL9lbc5ftC+/OKC49Yd6/5/PMs1ox9hzZrgsAN25sdnfq0Br6zxenPu+1x85e3rPs97dxFnnHQ4S5au4G/jp9Bh89YAnDn0CA7cuw9PP/c61456mE9Xr6ZZWRnnfnMQe+++PQAPPz6dm++ciCS6dGzH5RecQPu0f1PQ2Ls4lL0qq44OLh0N3AfsGBGv1NmJatkee/aLx56c3NDVyMt7Cz/ivQ8+ZrcderJ02ScMPO3XjPzlt3hnwWIO3Ks3ZWWl/Oy6sQD8+MyvsXzFSl54bS6vzp7PK7PfXZegF320jAGn/h/jb/o+nTq04dzLb+W4QXvz5X47NOTl5e2dD1c0dBU2ypo1azly2C8Y9duzGDthKpu1asHJxx60XplX/jOPTu3b0qVTO2a9+S5n/2QkD42+mNVr1jBo6C+467rzab95a343chwtWzTjjG8c3kBXUzO7bNl2WkT025Rj9Nllj7jx3sfyKnvQDh03+Xz1ra5fGnsi8GT6uckkucW/gW6dN2e3HbKXCLdp3ZLtt+7G/PcX03/fPpSVlQKw185bM3/BYgA2a9WCfXffjhbNm613nLffWci2W3ahU4c2AHx57x0YN+n5+ruQJmrK87Po0b0T3bt2qLRMn+160KVTOwC227obK1d9yqpPV0NABKxYuYqIYNnylevKNRlpHHQ+SzGqswQtqQ1wIHAa2ZttkdRf0hOSHpT0qqQ/SipJ25ZKukrSTEmPSuqS4pMkXS1pKnCupEMlPSfpBUkj01t1B0q6K+fc/SU9kNYHSPq3pGcl3ZXq1SjNmf8BL74+lz133ma9+B0PTuaQ/Xesct9tenThP28vYM78D1i9eg0PPzGDeSmpW90Z/8TzHHHQ7us+3/nAvxgy4mouvfouPl66/HPlH33qRfps14PmzcooKyvlwjOPZshZVzNw6C94Y857DD587/qsfkGorbd6p3yyQNKLObGOkiZIej397JDiknSNpFmSZkjaM2efYan865KG5cT3SnlrVtq32mrVZQt6MPBwRLwGfCBprxTfBzgb2AnYDjg2xVsDUyNiZ+Bxsjfqlmuevpr8ARgFnBARu5L1oX8H+Aewr6TyzrcTyN6e2xn4MXBYROwJTAXOr6iykoZLmipp6sKF72/61dezZctX8q0fjeSyc46lbeuW6+K/u+URykpLOHZA1d/s2rfbjF9+73i+/ZNbOObMa+jZvSOlJXX9Batp+/TT1TzxzMscduCuABx35H787cYfcPs159C5Yzuu+vOD65X/z1vv8ftRD3HxiGMAWL16DfeMe5rbrjmHh0dfzPbbdOfmuybW+3U0pPJHvfNZ8jAKGLhB7ELg0YjoDTyaPgMMAnqnZThwPWQJnSx37UuW6y4pT+qpzOk5+214rs+py/8HngiMSetj+Kyb45mImJ1eUX4HWSsbYC3w17R+a06cnPgOwBsp6QPcAhwUEauBh4Gvpm6Qo4D7gf3I/hA8JWk6MAzYuqLKRsQNEdEvIvp17txlIy+5YXy6eg3f+tFIjh3QjyP7f9Ya++uDk/nHUzO59pKheb0WaMCBu/Dgjefz9xvOY7uturFtz+L6PRSbp6a9Sp/tetCpQ1sAOnVoS2lpCSUlJRxzxN7MfG3uurLvLfyI7//8L1x6/vFs2b0TAK/OfgeALbt3QhKHf3lXZrz8dv1fSEOrpSZ0RDwBLNogPJgsz5B+Hp0THx2Zp4H2kroDRwATImJRRHwITAAGpm3tIuLpyG78jc45VqXqpE83/RU5BNhVUgClQAAPpp+5KrtLmRtflsdpxwAjyH7BUyNiSfoKMaGS16k3ChHBBb+8g95bd+OMIQevi098+mWuu/1R7r32HDZr2TyvYy38cAmdO7Rl8cfLGXXvk/zp8lPqqNYGMP7x9bs3Fi76mM4dsz7kif+eyXZbdwNgydIVfPenNzPilIH03WmbdeW7dtqc2XMW8OFHS+mweRsmPzeLXk3wj2oNhtl1Tl2l5W6IiBuq2adbRMxP6+8C3dJ6D2BOTrm5KVZVfG4F8SrV1U2344C/RMQZ5QFJjwNfBvaR1At4i6wrovwXVJL2GwN8nezm4oZeBbaRtH1EzAJOJusOIf0cSfYVorzl/jTwh/LyqQukR04LvOg9M2M2dz88hR23685hw/4PgIvOOIr/vfpeVn66mhO+ex2Q3Si88gcnALDPf1/K0mWfsGr1asb/cwZ3XHUmX+y1Bf979b28NGseAOedOpDtturaMBfVBKz4ZBXPTJ/Fj0Ycuy72u5sf4rXZ7yCJ7l078KPUlfHXB/7FnPkf8Oc7HuXPdzwKwLWXn0aXTu04/cRDOf2Hf6KsrJTuXdpzyXn/0yDX05BqcANw4aaM4oiISA3OelMnw+wkTQSujIiHc2LnkPUXvw8sAbYHJgJnRsRaSUvJkvUAYAFZP/P7kiYB34uIqek4hwK/JvvjMgX4TkSsTNuuBU4BukbE8hQ7BLgSaJGq8uOIGFtV/YtpmJ1linWYXVNWG8Psdtx1jxh9/6S8yu6zXftqzydpG+CBiNglfX4V6B8R81M3xaSI2EHSn9L6Hbnlypfyxml5ubRMjIg+KX5ibrnK1EkLOiIOriB2jaQZZMn2K5Xs97kbeBHRf4PPjwJ7VLL/CLJujtzYY0DTu7Vt1lTU7RC6sWT3rq5IP+/PiY+QNIbshuBHKYmPB36Rc2NwAHBRRCyS9LGk/YDJwFDg99Wd3OOKzaxoSbU3F4ekO8hawJ0lzSUbjXEFcKek08i6ZY9PxccBRwKzgOXAqQApEV9O9u0e4LKIKL/xeCbZSJFWwENpqVK9JuiImETW1K9oW6Mdn2xmdae2GtBVDCY4tIKyAZxVyXFGkt0P2zA+FdilJnVyC9rMiluRPiWYDydoMytifmmsmVnBKtZ5NvLhBG1mRUs4QZuZFSx3cZiZFSi3oM3MClQjzs9O0GZWxPKd7LlIOUGbWVFzH7SZWQFq7C+NdYI2s+LmBG1mVpjcxWFmVqA8zM7MrEA14vzsBG1mRa4RZ2gnaDMrWrU5YX8hcoI2s6LWeNOzE7SZFbtGnKGdoM2siDXuCftLGroCZmabQspvqf44elPSC5KmS5qaYh0lTZD0evrZIcUl6RpJsyTNkLRnznGGpfKvSxq2KdfmBG1mRat8wv7aSNDJwRHRNyL6pc8XAo9GRG/g0fQZYBDQOy3DgeshS+hkbwPfF9gHuKQ8qW8MJ2gzK2rK838baTBwS1q/BTg6Jz46Mk8D7SV1B44AJkTEooj4EJgADNzYkztBm1lRq0ELurOkqTnL8A0OFcAjkqblbOsWEfPT+rtAt7TeA5iTs+/cFKssvlF8k9DMiloN2sYLc7ouKnJgRMyT1BWYIOmV3I0REZJi42q5cdyCNrPilWfrOZ8+6IiYl34uAO4j60N+L3VdkH4uSMXnAT1zdt8yxSqLbxQnaDMrcspzqeIIUmtJbcvXgQHAi8BYoHwkxjDg/rQ+FhiaRnPsB3yUukLGAwMkdUg3Bwek2EZxF4eZFa1anLC/G3CfsqZ2GXB7RDwsaQpwp6TTgLeA41P5ccCRwCxgOXAqQEQsknQ5MCWVuywiFm1spZygzayo1cZUHBExG9i9gvgHwKEVxAM4q5JjjQRGbnqtnKDNrMg15icJnaDNrLg13vzsBG1mxa0R52cnaDMrXjV8jLvoOEGbWVFTI87QTtBmVtQab3p2gjazIteIG9BO0GZWzBr3hP1O0GZWtMrng26snKDNrKg5QZuZFSh3cZiZFSKPgzYzK0zVTyRa3Jygzay4NeIM7QRtZkXNfdBmZgWqlibsL0hO0GZW3JygzcwKk7s4zMwKUGN/klDZq7Usl6T3yV4Q2Rh1BhY2dCWsRhrrv9nWEdFlUw4g6WGy308+FkbEwE05X31zgm5iJE2NiH4NXQ/Ln//Nmq6Shq6AmZlVzAnazKxAOUE3PTc0dAWsxvxv1kS5D9rMrEC5BW1mVqCcoM3MCpQTdJGTtEbSdEkzJT0v6QJJ/ndtYJKOlhSS+jR0Xax4+f/IxW9FRPSNiJ2Bw4FBwCUNXCeDE4En089NJslP/TZBTtCNSEQsAIYDI5QplfQrSVMkzZB0BoCk/pImSbpb0iuSbkvlD5H0t/LjSTpc0n0NdDlFS1Ib4EDgNGBIivWX9ISkByW9KumP5d90JC2VdFX6FvSopC4pPknS1ZKmAudKOlTSc5JekDRSUgtJAyXdlXPu/pIeSOsDJP1b0rOS7kr1siLiBN3IRMRsoBToSpYgPoqIvYG9gdMl9UpF9wC+C+wEbAscAEwE+pQnCOBUYGT91b7RGAw8HBGvAR9I2ivF9wHOJvudbwccm+KtganpW9DjrP8NqHl6ivAPwCjghIjYlWwene8A/wD2ldQ6lT8BGCOpM/Bj4LCI2BOYCpxfFxdrdccJunEbAAyVNB2YDHQCeqdtz0TE3IhYC0wHtolszOVfgJMktQf2Bx6q70o3AicCY9L6GD7r5ngmImZHxBrgDrJWNsBa4K9p/dacODnxHYA3UtIHuAU4KCJWAw8DX03dIEcB9wP7kf0heCr9+w8Dtq61K7R64X6tRkbStsAaYAHZZF9nR8T4Dcr0B1bmhNbw2X8LNwN/Bz4B7koJwPIkqSNwCLCrpCD7NhPAg+lnrsoeQsiNL8vjtGOAEcAispb4EkkCJkRErfSBW8NwC7oRSV0TfwSuTa3h8cB3JDVL27+Y81W4QhHxDvAO2dfjm+u4yo3RccBfImLriNgmInoCbwBfBvaR1Cv1PZ9AdhMRsv8fHpfWv54Tz/UqsI2k7dPnk8m6Q0g/9wRO57OW+9PAAeXlJbWW9MXaukirH07Qxa9V+TA7sv7IR4BL07Y/Ay8Bz0p6EfgT+X1rug2YExEv10WFG7kTgQ1vrN6T4lOAa4GXyZJ2ebllZMn7RbLW92UbHjQiPiG7J3CXpBfIukX+mLatAR4gG8HzQIq9D5wC3CFpBvBvwEP+iowf9bbPkXQt8FxE3NTQdWksUrfS9yLiKxVsWxoRHmFhn+M+aFuPpGlkLboLGrouZk2dW9BmZgXKfdBmZgXKCdrMrEA5QZuZFSgnaNsoObPovZjmedhsE441StJxaf3Pknaqomx/SV/aiHO8mR5/ziu+QZmlNTzXTyV9r6Z1NNuQE7RtrPJZ9HYBVgHfzt24sbOvRcS3IuKlKor0B2qcoM2KkRO01YZ/Atun1u0/JY0FXqpiNj1JujbN6vYPsomdSNsmSeqX1gemmdieT7O8bUP2h+C81Hr/sqQuku5J55gi6YC0bydJj6QZ4v5M9th7lST9TdK0tM/wDbZVNNvcdpIeTvv8U5772WqZx0HbJkkt5UFkE/ZA9sjxLhHxRkpyH0XE3pJakE3c8wjZTHo7kE3m043saceRGxy3C3Aj2YRAb0jqGBGLJP0RWBoRv07lbgeuiognJW1F9nj7jmQzwj0ZEZdJOopsZr/qfDOdoxUwRdI9EfEBn802d56kn6RjjyB7meu3I+J1SfsC15E9CWhWK5ygbWO1SrOkQdaCvoms6+GZiHgjxQcAu5X3LwObk82mdxBwR3pE+R1Jj1Vw/P2AJ8qPFRGLKqnHYcBO2dxAALRL8x4fRJrOMyIelPRhHtd0jqRj0nrPVNcP+Pxsc/emc3yJ7NHr8v1b5HEOs7w5QdvGWhERfXMDKVHlzr5W2Wx6R9ZiPUqA/dJcFRvWJW/pUezDgP0jYrmkSUDLSopHOu/iDX8HZrXJfdBWlyqbTe8J4ITUR90dOLiCfZ8GDlJ6wUCaxhNgCdA2p9wjZJPgk8r1TatPkM0Mh6RBQIdq6ro58GFKzn3IWvDlPjfbXER8DLwh6X/SOSRp92rOYVYjTtBWlyqbTe8+4PW0bTTZTGvrSbOxDSfrTniez7oY/g4cU36TEDgH6JduQr7EZ6NJLiVL8DPJujrerqauDwNlkl4GriD7A1GustnmvgGcluo3k+xNKma1xnNxmJkVKLegzcwKlBO0mVmBcoI2MytQTtBmZgXKCdrMrEA5QZuZFSgnaDOzAvX/CPou5lHKpIQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn import tree\n",
    "from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# Train a model and make predictions\n",
    "model_baseline = tree.DecisionTreeClassifier(random_state=0, max_depth=3)\n",
    "model_baseline.fit(X_train, y_train)\n",
    "preds_baseline = model_baseline.predict(X_test)\n",
    "\n",
    "# Function to plot confusion matrix\n",
    "def plot_confusion_matrix(estimator, X, y_true, y_pred, display_labels=[\"Deny\", \"Approve\"],\n",
    "                          include_values=True, xticks_rotation='horizontal', values_format='',\n",
    "                          normalize=None, cmap=plt.cm.Blues):\n",
    "    cm = confusion_matrix(y_true, y_pred, normalize=normalize)\n",
    "    disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=display_labels)\n",
    "    return cm, disp.plot(include_values=include_values, cmap=cmap, xticks_rotation=xticks_rotation,\n",
    "                     values_format=values_format)\n",
    "\n",
    "# Function to evaluate the fairness of the model\n",
    "def get_stats(X, y, model, group_one, preds):\n",
    "        \n",
    "    y_zero, preds_zero, X_zero = y[group_one==False], preds[group_one==False], X[group_one==False]\n",
    "    y_one, preds_one, X_one = y[group_one], preds[group_one], X[group_one]\n",
    "    \n",
    "    print(\"Total approvals:\", preds.sum())\n",
    "    print(\"Group A:\", preds_zero.sum(), \"({}% of approvals)\".format(round(preds_zero.sum()/sum(preds)*100, 2)))\n",
    "    print(\"Group B:\", preds_one.sum(), \"({}% of approvals)\".format(round(preds_one.sum()/sum(preds)*100, 2)))\n",
    "    \n",
    "    print(\"\\nOverall accuracy: {}%\".format(round((preds==y).sum()/len(y)*100, 2)))\n",
    "    print(\"Group A: {}%\".format(round((preds_zero==y_zero).sum()/len(y_zero)*100, 2)))\n",
    "    print(\"Group B: {}%\".format(round((preds_one==y_one).sum()/len(y_one)*100, 2)))\n",
    "    \n",
    "    cm_zero, disp_zero = plot_confusion_matrix(model, X_zero, y_zero, preds_zero)\n",
    "    disp_zero.ax_.set_title(\"Group A\")\n",
    "    cm_one, disp_one = plot_confusion_matrix(model, X_one, y_one, preds_one)\n",
    "    disp_one.ax_.set_title(\"Group B\")\n",
    "    \n",
    "    print(\"\\nSensitivity / True positive rate:\")\n",
    "    print(\"Group A: {}%\".format(round(cm_zero[1,1] / cm_zero[1].sum()*100, 2)))\n",
    "    print(\"Group B: {}%\".format(round(cm_one[1,1] / cm_one[1].sum()*100, 2)))\n",
    "    \n",
    "# Evaluate the model    \n",
    "get_stats(X_test, y_test, model_baseline, X_test[\"Group\"]==1, preds_baseline)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "95dc06d1",
   "metadata": {
    "papermill": {
     "duration": 0.024107,
     "end_time": "2022-03-06T22:14:54.835683",
     "exception": false,
     "start_time": "2022-03-06T22:14:54.811576",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "The confusion matrices above show how the model performs on some test data. We also print additional information (calculated from the confusion matrices) to assess fairness of the model. For instance,\n",
    "- The model approved 38246 people for a credit card. Of these individuals, 8028 belonged to Group A, and 30218 belonged to Group B.\n",
    "- The model is 94.56% accurate for Group A, and 95.02% accurate for Group B. These percentages can be calculated directly from the confusion matrix; for instance, for Group A, the accuracy is (39723+7528)/(39723+500+2219+7528).\n",
    "- The true positive rate (TPR) for Group A is 77.23%, and the TPR for Group B is 98.03%. These percentages can be calculated directly from the confusion matrix; for instance, for Group A, the TPR is 7528/(7528+2219).\n",
    "\n",
    "# 1) Varieties of fairness\n",
    "\n",
    "Consider three different types of fairness covered in the tutorial:\n",
    "- **Demographic parity**: Which group has an unfair advantage, with more representation in the group of approved applicants?  (Roughly 50% of applicants are from Group A, and 50% of applicants are from Group B.)\n",
    "- **Equal accuracy**: Which group has an unfair advantage, where applicants are more likely to be correctly classified? \n",
    "- **Equal opportunity**:  Which group has an unfair advantage, with a higher true positive rate?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "9b50aff4",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:54.889316Z",
     "iopub.status.busy": "2022-03-06T22:14:54.888249Z",
     "iopub.status.idle": "2022-03-06T22:14:54.892828Z",
     "shell.execute_reply": "2022-03-06T22:14:54.893410Z",
     "shell.execute_reply.started": "2022-03-06T21:58:35.023902Z"
    },
    "papermill": {
     "duration": 0.032994,
     "end_time": "2022-03-06T22:14:54.893572",
     "exception": false,
     "start_time": "2022-03-06T22:14:54.860578",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 4, \"questionId\": \"1_VarietiesOfFairness\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Solution:</span> All of the fairness criteria are unfairly biased in favor of Group B. 79.01% of applicants are from Group B -- if demographic parity is our fairness criterion, fair would mean that 50% of the applicants are from Group B. The model is also slightly more accurate for applicants from Group B (with accuracy of 95.02%, vs 94.56% for Group A). The true positive rate is very high for Group B (98.03%, vs. 77.23% for Group A). In other words, for Group B, almost all people who should be approved are actually approved. For Group A, if you should be approved, your chances of actually being approved are much lower."
      ],
      "text/plain": [
       "Solution: All of the fairness criteria are unfairly biased in favor of Group B. 79.01% of applicants are from Group B -- if demographic parity is our fairness criterion, fair would mean that 50% of the applicants are from Group B. The model is also slightly more accurate for applicants from Group B (with accuracy of 95.02%, vs 94.56% for Group A). The true positive rate is very high for Group B (98.03%, vs. 77.23% for Group A). In other words, for Group B, almost all people who should be approved are actually approved. For Group A, if you should be approved, your chances of actually being approved are much lower."
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check your answer (Run this code cell to get credit!)\n",
    "q_1.check()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3670ceb5",
   "metadata": {
    "papermill": {
     "duration": 0.026416,
     "end_time": "2022-03-06T22:14:54.946385",
     "exception": false,
     "start_time": "2022-03-06T22:14:54.919969",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "Run the next code cell without changes to visualize the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "339f69e7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:55.005344Z",
     "iopub.status.busy": "2022-03-06T22:14:55.004662Z",
     "iopub.status.idle": "2022-03-06T22:14:55.811893Z",
     "shell.execute_reply": "2022-03-06T22:14:55.811429Z",
     "shell.execute_reply.started": "2022-03-06T22:00:18.823791Z"
    },
    "papermill": {
     "duration": 0.838762,
     "end_time": "2022-03-06T22:14:55.812014",
     "exception": false,
     "start_time": "2022-03-06T22:14:54.973252",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def visualize_model(model, feature_names, class_names=[\"Deny\", \"Approve\"], impurity=False):\n",
    "    plot_list = tree.plot_tree(model, feature_names=feature_names, class_names=class_names, impurity=impurity)\n",
    "    [process_plot_item(item) for item in plot_list]\n",
    "\n",
    "def process_plot_item(item):\n",
    "    split_string = item.get_text().split(\"\\n\")\n",
    "    if split_string[0].startswith(\"samples\"):\n",
    "        item.set_text(split_string[-1])\n",
    "    else:\n",
    "        item.set_text(split_string[0])\n",
    "\n",
    "plt.figure(figsize=(20, 6))\n",
    "plot_list = visualize_model(model_baseline, feature_names=X_train.columns)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "652a7378",
   "metadata": {
    "papermill": {
     "duration": 0.019043,
     "end_time": "2022-03-06T22:14:55.850397",
     "exception": false,
     "start_time": "2022-03-06T22:14:55.831354",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "The flowchart shows how the model makes decisions:\n",
    "- `Group <= 0.5` checks what group the applicant belongs to: if the applicant belongs to Group A, then `Group <= 0.5` is true.\n",
    "- Entries like `Income <= 80210.5` check the applicant's income.\n",
    "\n",
    "To follow the flow chart, we start at the top and trace a path depending on the details of the applicant.  If the condition is true at a split, then we move down and to the left branch.  If it is false, then we move to the right branch.\n",
    "\n",
    "For instance, consider an applicant in Group B, who has an income of 75k.  Then, \n",
    "- We start at the top of the flow chart.  the applicant has an income of 75k, so `Income <= 80210.5` is true, and we move to the left.\n",
    "- Next, we check the income again. Since `Income <= 71909.5` is false, we move to the right.\n",
    "- The last thing to check is what group the applicant belongs to.  The applicant belongs to Group B, so `Group <= 0.5` is false, and we move to the right, where the model has decided to approve the applicant.\n",
    "\n",
    "# 2) Understand the baseline model\n",
    "\n",
    "Based on the visualization, how can you explain one source of unfairness in the model?\n",
    "\n",
    "**Hint**: Consider the example applicant, but change the group membership from Group B to Group A (leaving all other characteristics the same).  Is this slightly different applicant approved or denied by the model?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "5849b8e8",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:55.890596Z",
     "iopub.status.busy": "2022-03-06T22:14:55.890134Z",
     "iopub.status.idle": "2022-03-06T22:14:55.895549Z",
     "shell.execute_reply": "2022-03-06T22:14:55.895992Z",
     "shell.execute_reply.started": "2022-03-06T22:04:31.942744Z"
    },
    "papermill": {
     "duration": 0.02709,
     "end_time": "2022-03-06T22:14:55.896165",
     "exception": false,
     "start_time": "2022-03-06T22:14:55.869075",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 4, \"questionId\": \"2_BaselineModel\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Solution:</span> For all applicants with income between 71909.5 and 88440.5, the model's decision depends on group membership: the model approves applicants from Group B and denies applicants from Group A.  Since we want the model to treat individuals from Group A and Group B fairly, this is clearly a bad model.  Although this data is very simple, in practice, with much larger data, visualizing the how a model makes decisions can be very useful to more deeply understand what might be going wrong with a model."
      ],
      "text/plain": [
       "Solution: For all applicants with income between 71909.5 and 88440.5, the model's decision depends on group membership: the model approves applicants from Group B and denies applicants from Group A.  Since we want the model to treat individuals from Group A and Group B fairly, this is clearly a bad model.  Although this data is very simple, in practice, with much larger data, visualizing the how a model makes decisions can be very useful to more deeply understand what might be going wrong with a model."
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check your answer (Run this code cell to get credit!)\n",
    "q_2.check()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b13a321",
   "metadata": {
    "papermill": {
     "duration": 0.019395,
     "end_time": "2022-03-06T22:14:55.935581",
     "exception": false,
     "start_time": "2022-03-06T22:14:55.916186",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "Next, you decide to remove group membership from the training data and train a new model.  Do you think this will make the model treat the groups more equally?\n",
    "\n",
    "Run the next code cell to see how this new **group unaware** model performs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "e5aaa511",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:55.978630Z",
     "iopub.status.busy": "2022-03-06T22:14:55.978183Z",
     "iopub.status.idle": "2022-03-06T22:14:56.886499Z",
     "shell.execute_reply": "2022-03-06T22:14:56.886138Z",
     "shell.execute_reply.started": "2022-03-06T22:06:18.056444Z"
    },
    "papermill": {
     "duration": 0.93108,
     "end_time": "2022-03-06T22:14:56.886606",
     "exception": false,
     "start_time": "2022-03-06T22:14:55.955526",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total approvals: 36670\n",
      "Group A: 11624 (31.7% of approvals)\n",
      "Group B: 25046 (68.3% of approvals)\n",
      "\n",
      "Overall accuracy: 92.66%\n",
      "Group A: 93.61%\n",
      "Group B: 91.72%\n",
      "\n",
      "Sensitivity / True positive rate:\n",
      "Group A: 93.24%\n",
      "Group B: 86.21%\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create new dataset with gender removed\n",
    "X_train_unaware = X_train.drop([\"Group\"],axis=1)\n",
    "X_test_unaware = X_test.drop([\"Group\"],axis=1)\n",
    "\n",
    "# Train new model on new dataset\n",
    "model_unaware = tree.DecisionTreeClassifier(random_state=0, max_depth=3)\n",
    "model_unaware.fit(X_train_unaware, y_train)\n",
    "\n",
    "# Evaluate the model\n",
    "preds_unaware = model_unaware.predict(X_test_unaware)\n",
    "get_stats(X_test_unaware, y_test, model_unaware, X_test[\"Group\"]==1, preds_unaware)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b41bbe1",
   "metadata": {
    "papermill": {
     "duration": 0.019956,
     "end_time": "2022-03-06T22:14:56.926611",
     "exception": false,
     "start_time": "2022-03-06T22:14:56.906655",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 3) Varieties of fairness, part 2\n",
    "\n",
    "How does this model compare to the first model you trained, when you consider **demographic parity**, **equal accuracy**, and **equal opportunity**?  Once you have an answer, run the next code cell."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "84fd4746",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:56.969073Z",
     "iopub.status.busy": "2022-03-06T22:14:56.968409Z",
     "iopub.status.idle": "2022-03-06T22:14:56.972466Z",
     "shell.execute_reply": "2022-03-06T22:14:56.972959Z",
     "shell.execute_reply.started": "2022-03-06T22:07:39.063365Z"
    },
    "papermill": {
     "duration": 0.026867,
     "end_time": "2022-03-06T22:14:56.973149",
     "exception": false,
     "start_time": "2022-03-06T22:14:56.946282",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 4, \"questionId\": \"3_VarietiesOfFairnessTwo\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Solution:</span> When we consider demographic parity, the new model is still biased in favor of Group B, but is now a bit more fair than the original model. But now, if you consider either equal accuracy or equal opportunity, the model is biased in favor of Group A! It's also important to note that the overall accuracy of the model has dropped -- for each group, the model is making slightly less accurate decisions."
      ],
      "text/plain": [
       "Solution: When we consider demographic parity, the new model is still biased in favor of Group B, but is now a bit more fair than the original model. But now, if you consider either equal accuracy or equal opportunity, the model is biased in favor of Group A! It's also important to note that the overall accuracy of the model has dropped -- for each group, the model is making slightly less accurate decisions."
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check your answer (Run this code cell to get credit!)\n",
    "q_3.check()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "41e760ea",
   "metadata": {
    "papermill": {
     "duration": 0.032665,
     "end_time": "2022-03-06T22:14:57.038573",
     "exception": false,
     "start_time": "2022-03-06T22:14:57.005908",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "You decide to train a third potential model, this time with the goal of having each group have even representation in the group of approved applicants.  (This is an implementation of group thresholds, which you can optionally read more about [here](https://pair-code.github.io/what-if-tool/ai-fairness.html).)  \n",
    "\n",
    "Run the next code cell without changes to evaluate this new model.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "b23dcf8e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:57.107133Z",
     "iopub.status.busy": "2022-03-06T22:14:57.106474Z",
     "iopub.status.idle": "2022-03-06T22:14:57.572136Z",
     "shell.execute_reply": "2022-03-06T22:14:57.572494Z",
     "shell.execute_reply.started": "2022-03-06T22:10:10.761937Z"
    },
    "papermill": {
     "duration": 0.501149,
     "end_time": "2022-03-06T22:14:57.572626",
     "exception": false,
     "start_time": "2022-03-06T22:14:57.071477",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total approvals: 38241\n",
      "Group A: 19869 (51.96% of approvals)\n",
      "Group B: 18372 (48.04% of approvals)\n",
      "\n",
      "Overall accuracy: 79.38%\n",
      "Group A: 79.74%\n",
      "Group B: 79.02%\n",
      "\n",
      "Sensitivity / True positive rate:\n",
      "Group A: 100.0%\n",
      "Group B: 63.64%\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Change the value of zero_threshold to hit the objective\n",
    "zero_threshold = 0.11\n",
    "one_threshold = 0.99\n",
    "\n",
    "# Evaluate the model\n",
    "test_probs = model_unaware.predict_proba(X_test_unaware)[:,1]\n",
    "preds_approval = (((test_probs>zero_threshold)*1)*[X_test[\"Group\"]==0] + ((test_probs>one_threshold)*1)*[X_test[\"Group\"]==1])[0]\n",
    "get_stats(X_test, y_test, model_unaware, X_test[\"Group\"]==1, preds_approval)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "10a624b5",
   "metadata": {
    "papermill": {
     "duration": 0.024226,
     "end_time": "2022-03-06T22:14:57.620944",
     "exception": false,
     "start_time": "2022-03-06T22:14:57.596718",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 4) Varieties of fairness, part 3\n",
    "\n",
    "How does this final model compare to the previous models, when you consider **demographic parity**, **equal accuracy**, and **equal opportunity**?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "8c61d199",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-03-06T22:14:57.672269Z",
     "iopub.status.busy": "2022-03-06T22:14:57.671795Z",
     "iopub.status.idle": "2022-03-06T22:14:57.677776Z",
     "shell.execute_reply": "2022-03-06T22:14:57.678218Z",
     "shell.execute_reply.started": "2022-03-06T22:10:53.983884Z"
    },
    "papermill": {
     "duration": 0.033057,
     "end_time": "2022-03-06T22:14:57.678341",
     "exception": false,
     "start_time": "2022-03-06T22:14:57.645284",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "parent.postMessage({\"jupyterEvent\": \"custom.exercise_interaction\", \"data\": {\"outcomeType\": 1, \"valueTowardsCompletion\": 0.25, \"interactionType\": 1, \"questionType\": 4, \"questionId\": \"4_VarietiesOfFairnessThree\", \"learnToolsVersion\": \"0.3.4\", \"failureMessage\": \"\", \"exceptionClass\": \"\", \"trace\": \"\"}}, \"*\")"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/markdown": [
       "<span style=\"color:#33cc33\">Solution:</span> This model acheives nearly equal representation in the pool of approved applicants from each group -- if demographic parity is what we care about, then this model is much more fair than the first two models. Accuracy is roughly the same for each group, but there is a substantial drop in overall accuracy for each group. If we examine the model for equal opportunity fairness, the model is biased in favor of Group A: all individuals from Group A who should be approved are approved, whereas only 63% of individuals from Group B who should be approved are approved.  (This is similar to the dynamic in the first model, with the favored group switched -- that is, in the first model, nearly all individuals from Group B who should be approved were approved by the model."
      ],
      "text/plain": [
       "Solution: This model acheives nearly equal representation in the pool of approved applicants from each group -- if demographic parity is what we care about, then this model is much more fair than the first two models. Accuracy is roughly the same for each group, but there is a substantial drop in overall accuracy for each group. If we examine the model for equal opportunity fairness, the model is biased in favor of Group A: all individuals from Group A who should be approved are approved, whereas only 63% of individuals from Group B who should be approved are approved.  (This is similar to the dynamic in the first model, with the favored group switched -- that is, in the first model, nearly all individuals from Group B who should be approved were approved by the model."
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Check your answer (Run this code cell to get credit!)\n",
    "q_4.check()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "19e22a1d",
   "metadata": {
    "papermill": {
     "duration": 0.025561,
     "end_time": "2022-03-06T22:14:57.729097",
     "exception": false,
     "start_time": "2022-03-06T22:14:57.703536",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "This is only a short exercise to explore different types of fairness, and to illustrate the tradeoff that can occur when you optimize for one type of fairness over another.  We have focused on model training here, but in practice, to really mitigate bias, or to make ML systems fair, we need to take a close look at every step in the process, from data collection to releasing a final product to users. \n",
    "\n",
    "For instance, if you take a close look at the data, you'll notice that on average, individuals from Group B tend to have higher income than individuals from Group A, and are also more likely to own a home or a car.  Knowing this will prove invaluable to deciding what fairness criterion you should use, and to inform ways to achieve fairness.  (*For instance, it would likely be a bad aproach, if you did not remove the historical bias in the data and then train the model to get equal accuracy for each group.*)\n",
    "\n",
    "In this course, we intentionally avoid taking an opinionated stance on how exactly to minimize bias and ensure fairness in specific projects.  This is because the correct answers continue to evolve, since AI fairness is an active area of research.  This lesson was a hands-on introduction to the topic, and you can continue your learning by reading blog posts from the [Partnership on AI](https://www.partnershiponai.org/research-lander/) or by following conferences like the [ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT)](https://facctconference.org/)."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "998e54f8",
   "metadata": {
    "papermill": {
     "duration": 0.025754,
     "end_time": "2022-03-06T22:14:57.780637",
     "exception": false,
     "start_time": "2022-03-06T22:14:57.754883",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Keep going\n",
    "\n",
    "Continue to **[learn how to use model cards](https://www.kaggle.com/var0101/model-cards)** to make machine learning models transparent to large audiences."
   ]
  }
 ],
 "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": "3.7.12"
  },
  "papermill": {
   "default_parameters": {},
   "duration": 17.653723,
   "end_time": "2022-03-06T22:14:58.617526",
   "environment_variables": {},
   "exception": null,
   "input_path": "__notebook__.ipynb",
   "output_path": "__notebook__.ipynb",
   "parameters": {},
   "start_time": "2022-03-06T22:14:40.963803",
   "version": "2.3.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}