2_Exp_OUE.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e9bccd79",
   "metadata": {},
   "source": [
    "## Static parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "6810ab1b",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "nb_seed = 20\n",
    "dataset = 'LSAC' # select one in ['adult', 'ACSCoverage', 'LSAC']\n",
    "split_strategy = 'uniform' # select one in ['uniform', 'k_based']\n",
    "mechanism = \"OUE\"\n",
    "\n",
    "# read LGBM hyparameters of non-private model\n",
    "params = np.load('results/' + dataset + '/non_private' + '/LGBM_hyperparameters.npy', allow_pickle='TRUE').item()\n",
    "\n",
    "# set mechanism folder\n",
    "folder_name = 'results/' + dataset + \"/\" + mechanism + \"/\" +  split_strategy\n",
    "\n",
    "# for ML\n",
    "test_size = 0.2 # test proportion for train_test_split\n",
    "if dataset == 'adult':\n",
    "    target = 'income'\n",
    "    protected_attribute = 'gender'\n",
    "    \n",
    "elif dataset == 'ACSCoverage':\n",
    "    target = 'PUBCOV'\n",
    "    protected_attribute = 'DIS'\n",
    "    \n",
    "elif dataset == 'LSAC':\n",
    "    target = 'pass_bar'\n",
    "    protected_attribute = 'race1' \n",
    "\n",
    "# for privacy\n",
    "lst_eps = [0.25, 0.5, 1, 2, 4, 8, 10, 20, 50] # epsilon-LDP values\n",
    "if dataset == 'adult':\n",
    "    lst_sensitive_att = [protected_attribute, 'race', 'native-country', 'age']\n",
    "    \n",
    "elif dataset == 'ACSCoverage':\n",
    "    lst_sensitive_att = [protected_attribute, 'AGEP', 'SEX', 'SCHL']\n",
    "\n",
    "elif dataset == 'LSAC':\n",
    "    lst_sensitive_att = [protected_attribute, 'fam_inc', 'gender', 'fulltime']"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1895b9d1",
   "metadata": {},
   "source": [
    "## Writing function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "662cbe9e",
   "metadata": {},
   "outputs": [],
   "source": [
    "def write(folder_name, values, mechanism, epsilon):\n",
    "    with open(folder_name + \"/LGBM_results_\"+mechanism+\"_eps_\"+str(epsilon)+\".csv\", mode='a', newline='') as scores_file:\n",
    "        scores_writer = csv.writer(scores_file, delimiter=',', quotechar='\"', quoting=csv.QUOTE_MINIMAL)\n",
    "        scores_writer.writerow(values)\n",
    "    scores_file.close()    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a5abd282",
   "metadata": {},
   "source": [
    "## Importing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "7c0a04b3",
   "metadata": {},
   "outputs": [],
   "source": [
    "# General imports\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import time\n",
    "import copy\n",
    "import csv\n",
    "\n",
    "# sklearn imports\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "from lightgbm import LGBMClassifier\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score, confusion_matrix, f1_score, roc_auc_score, recall_score\n",
    "\n",
    "# designed functions\n",
    "from functions import fairness_metrics\n",
    "\n",
    "# multi-freq-ldpy import\n",
    "from multi_freq_ldpy.pure_frequency_oracles.UE import UE_Client\n",
    "from numba import jit\n",
    "\n",
    "@jit(nopython=True)\n",
    "def setting_seed(seed):\n",
    "    \"\"\" Function to set seed for reproducibility.\n",
    "    Calling numpy.random.seed() from interpreted code will \n",
    "    seed the NumPy random generator, not the Numba random generator.\n",
    "    Check: https://numba.readthedocs.io/en/stable/reference/numpysupported.html\"\"\"\n",
    "    \n",
    "    np.random.seed(seed)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "98f616e9",
   "metadata": {},
   "source": [
    "## Reading dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "afdf45b3",
   "metadata": {
    "scrolled": false
   },
   "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>fam_inc</th>\n",
       "      <th>gender</th>\n",
       "      <th>fulltime</th>\n",
       "      <th>race1</th>\n",
       "      <th>lsat</th>\n",
       "      <th>ugpa</th>\n",
       "      <th>pass_bar</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>32</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>17</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>24</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>27</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>36</td>\n",
       "      <td>4</td>\n",
       "      <td>1</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",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20422</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20423</th>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20424</th>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>24</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20425</th>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>32</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20426</th>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>18</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>20427 rows × 7 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       fam_inc  gender  fulltime  race1  lsat  ugpa  pass_bar\n",
       "0            4       0         0      1    32     4         1\n",
       "1            3       0         0      1    17     4         1\n",
       "2            0       1         0      1    24     4         1\n",
       "3            3       1         0      1    27     4         1\n",
       "4            3       1         0      1    36     4         1\n",
       "...        ...     ...       ...    ...   ...   ...       ...\n",
       "20422        1       1         0      0    14     1         0\n",
       "20423        2       1         0      0     8     1         0\n",
       "20424        2       1         1      0    24     1         1\n",
       "20425        2       1         1      1    32     0         1\n",
       "20426        2       1         0      1    18     0         1\n",
       "\n",
       "[20427 rows x 7 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "if dataset == 'adult':\n",
    "    df = pd.read_csv('datasets/db_adult_processed_26k.csv')\n",
    "    \n",
    "elif dataset == 'ACSCoverage':\n",
    "    df = pd.read_csv('datasets/db_ACSCoverage.csv')\n",
    "\n",
    "elif dataset == 'LSAC':\n",
    "    df = pd.read_csv('datasets/db_LSAC.csv')\n",
    "\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d55fd99",
   "metadata": {},
   "source": [
    "## Run LGBM on DP data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "d7137bd3",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.25\n",
      "0.5\n",
      "1\n",
      "2\n",
      "4\n",
      "8\n",
      "10\n",
      "20\n",
      "50\n",
      "That took 980.9397497177124 seconds\n"
     ]
    }
   ],
   "source": [
    "header = [\"seed\", \n",
    "          \"acc\", \"f1\", \"auc\", \"recall\", \"cm\",\n",
    "          \"SP_a_1\", \"SP_a_0\", \"SPD\", \"DI\", \n",
    "          \"EO_a_1\", \"EO_a_0\", \"EOD\", \n",
    "          \"OA_a_1\", \"OA_a_0\", \"OAD\",\n",
    "         ]\n",
    "\n",
    "starttime = time.time()\n",
    "\n",
    "# domain size of sensitive attributes\n",
    "lst_k = {att: len(set(df[att])) for att in lst_sensitive_att}\n",
    "\n",
    "for epsilon in lst_eps:\n",
    "    print(epsilon)\n",
    "    \n",
    "    # write head of csv file\n",
    "    write(folder_name, header, mechanism, epsilon)\n",
    "    \n",
    "    for seed in range(nb_seed):\n",
    "        setting_seed(seed) # for reproducibility\n",
    "        \n",
    "        # Use original datasets\n",
    "        X = copy.deepcopy(df.drop(target, axis=1))\n",
    "        y = copy.deepcopy(df[target])\n",
    "        \n",
    "        # Train test splitting\n",
    "        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size, shuffle=True, stratify=y, random_state=seed)\n",
    "        y_train.reset_index(inplace=True, drop=True)\n",
    "        y_test.reset_index(inplace=True, drop=True)\n",
    "        \n",
    "        # One-Hot-Encoding + LDP randomization\n",
    "        lst_df_train = []\n",
    "        lst_df_test = []\n",
    "        for col in X_train.columns:\n",
    "\n",
    "            lst_col_name = [col+\"_{}\".format(val) for val in range(len(set(df[col])))]\n",
    "            k = len(set(df[col]))\n",
    "            OHE = np.eye(k)\n",
    "\n",
    "            if col in lst_sensitive_att: # LDP randomization\n",
    "                eps_att = epsilon / len(lst_sensitive_att) if split_strategy=='uniform' else epsilon * k / sum(lst_k.values())\n",
    "                df_ohe = pd.DataFrame(np.stack(X_train[col].apply(lambda x: UE_Client(x, k, eps_att, optimal=True))), columns=lst_col_name)\n",
    "\n",
    "            else: # just one-hot-encoding\n",
    "                df_ohe = pd.DataFrame([OHE[val] for val in X_train[col]], columns=lst_col_name)\n",
    "\n",
    "            lst_df_train.append(df_ohe)\n",
    "            \n",
    "            # test set is original, i.e., just one-hot-encoding\n",
    "            df_ohe_test = pd.DataFrame([OHE[val] for val in X_test[col]], columns=lst_col_name)\n",
    "            lst_df_test.append(df_ohe_test)\n",
    "\n",
    "        # concat one-hot-encoded train/test sets\n",
    "        X_train = pd.concat(lst_df_train, axis=1)\n",
    "        X_test = pd.concat(lst_df_test, axis=1)\n",
    "        \n",
    "        # instantiate and train model\n",
    "        model = LGBMClassifier(random_state=seed, n_jobs=2, objective=\"binary\")\n",
    "        model.set_params(**params)\n",
    "        model.fit(X_train, y_train)\n",
    "        y_pred = model.predict(X_test)\n",
    "        \n",
    "        # performance metrics        \n",
    "        acc = accuracy_score(y_test, y_pred)\n",
    "        f1 = f1_score(y_test, y_pred)\n",
    "        auc = roc_auc_score(y_test, y_pred)\n",
    "        recall = recall_score(y_test, y_pred)\n",
    "        cm = confusion_matrix(y_test, y_pred)\n",
    "\n",
    "        # prepare dataset for fairness analysis\n",
    "        df_fm = pd.concat([X_test, y_test], axis=1)\n",
    "        df_fm['y_pred'] = y_pred\n",
    "\n",
    "        # fairness metrics\n",
    "        fair_met = fairness_metrics(df_fm, protected_attribute, target)\n",
    "\n",
    "        # write results to csv\n",
    "        write(folder_name, \n",
    "              [str(seed),\n",
    "              acc, f1, auc, recall, cm,\n",
    "              fair_met[\"SP_a_1\"], fair_met[\"SP_a_0\"], fair_met[\"SPD\"], fair_met[\"DI\"], \n",
    "              fair_met[\"EO_a_1\"], fair_met[\"EO_a_0\"], fair_met[\"EOD\"], \n",
    "              fair_met[\"OA_a_1\"], fair_met[\"OA_a_0\"], fair_met[\"OAD\"]], \n",
    "              mechanism, epsilon)\n",
    "        \n",
    "print('That took {} seconds'.format(time.time() - starttime))  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8fbf3967",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c311717a",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2ec7cf02",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}