examples/GPA_addl_dataset/gpa_data_preprocessing_addl_datasets_noleakage.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "60c25d09",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "5e1fb8f8",
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_primary_and_addl_dataset(\n",
    "    input_path,\n",
    "    output_path_primary_cand_data,\n",
    "    output_path_primary_safety_data,\n",
    "    output_path_addl_cand_data,\n",
    "    output_path_addl_safety_data,\n",
    "    output_path_primary_metadata,\n",
    "    output_path_addl_metadata,\n",
    "    columns_raw):\n",
    "    \n",
    "    df = pd.read_csv(input_path,header=None,names=columns_raw)\n",
    "    # extract features\n",
    "    features = df.iloc[:,1:-1]\n",
    "\n",
    "    # scale features using standard scaler\n",
    "    scaler = StandardScaler()\n",
    "    scaler.fit(features)\n",
    "    transformed_features = scaler.transform(features)\n",
    "\n",
    "    # replace features in original dataframe with transformed features\n",
    "    df.iloc[:,1:-1] = transformed_features\n",
    "\n",
    "    # convert continuous GPA labels to binary: 0 if GPA < 3, 1 if GPA >= 3\n",
    "    # Make a new column called GPA_class\n",
    "    df.loc[:,'GPA_class'] = df['GPA'].apply(lambda x: float(int(x>=3.0)))\n",
    "\n",
    "    # Drop original GPA column\n",
    "    df = df.drop(columns=['GPA'])\n",
    "    \n",
    "    # Shuffle original dataset\n",
    "    df = df.sample(frac=1.0,replace=False)\n",
    "    \n",
    "    # Take first half for candidate selection, second for safety test\n",
    "    df_cand = df.iloc[0:len(df)//2]\n",
    "    df_safety = df.iloc[len(df)//2:]\n",
    "    \n",
    "    # Remove gender label for primary dataset \n",
    "    df_primary_cand = df_cand.iloc[:,1:]\n",
    "    df_primary_safety = df_safety.iloc[:,1:]\n",
    "    \n",
    "    # Sample 30% of cand df for addl candidate dataset (still has gender label)\n",
    "    # Sample 30% of safety df for addl safety dataset (still has gender label)\n",
    "    df_addl_cand = df_cand.sample(frac=0.3,replace=False,random_state=42)\n",
    "    df_addl_safety = df_safety.sample(frac=0.3,replace=False,random_state=42)\n",
    "\n",
    "    # On both addl datasets, make gender column two columns: \"M\" and \"F\" for men and women, both binary valued\n",
    "    df_addl_cand['M'] = df_addl_cand.loc[:,'Gender']\n",
    "    df_addl_cand['F'] = df_addl_cand.loc[:,'Gender'].apply(lambda x: 0 if x==1 else 1)\n",
    "    \n",
    "    df_addl_safety['M'] = df_addl_safety.loc[:,'Gender']\n",
    "    df_addl_safety['F'] = df_addl_safety.loc[:,'Gender'].apply(lambda x: 0 if x==1 else 1)\n",
    "\n",
    "    # Drop gender column from addl datasets\n",
    "    df_addl_cand = df_addl_cand.drop(columns=[\"Gender\"])\n",
    "    df_addl_safety = df_addl_safety.drop(columns=[\"Gender\"])\n",
    "    # reorder columns so that sensitive attributes M and F are first\n",
    "    cols = df_addl_cand.columns.tolist()\n",
    "    cols_reorder = cols[-2:] + cols[0:-2]\n",
    "    df_addl_cand = df_addl_cand[cols_reorder]\n",
    "    df_addl_safety = df_addl_safety[cols_reorder]\n",
    "        \n",
    "    # Save final dataframes\n",
    "    df_primary_cand.to_csv(output_path_primary_cand_data,index=False,header=False)\n",
    "    df_primary_safety.to_csv(output_path_primary_safety_data,index=False,header=False)\n",
    "\n",
    "    print(f\"Saved primary candidate data file to: {output_path_primary_cand_data}\\n\")\n",
    "    print(f\"Saved primary safety data file to: {output_path_primary_safety_data}\\n\")\n",
    "    \n",
    "    df_addl_cand.to_csv(output_path_addl_cand_data,index=False,header=False)\n",
    "    df_addl_safety.to_csv(output_path_addl_safety_data,index=False,header=False)\n",
    "    print(f\"Saved addl candidate data file to: {output_path_addl_cand_data}\\n\")\n",
    "    print(f\"Saved addl safety data file to: {output_path_addl_safety_data}\\n\")\n",
    "\n",
    "    # Save metadata json files, only need one for primary and one for addl\n",
    "    \n",
    "    primary_metadata_dict = {\n",
    "        \"regime\":\"supervised_learning\",\n",
    "        \"sub_regime\":\"classification\",\n",
    "        \"all_col_names\":list(df_primary_cand.columns),\n",
    "        \"label_col_names\":[\"GPA_class\"],\n",
    "        \"sensitive_col_names\":[]\n",
    "    }\n",
    "    \n",
    "    addl_metadata_dict = {\n",
    "        \"regime\":\"supervised_learning\",\n",
    "        \"sub_regime\":\"classification\",\n",
    "        \"all_col_names\":list(df_addl_cand.columns),\n",
    "        \"label_col_names\":[\"GPA_class\"],\n",
    "        \"sensitive_col_names\":[\"M\",\"F\"]\n",
    "    }\n",
    "    \n",
    "    with open(output_path_primary_metadata,'w') as outfile:\n",
    "        json.dump(primary_metadata_dict,outfile,indent=2)\n",
    "    print(f\"Saved primary metadata file to: {output_path_primary_metadata}\\n\")\n",
    "    \n",
    "    with open(output_path_addl_metadata,'w') as outfile:\n",
    "        json.dump(addl_metadata_dict,outfile,indent=2)\n",
    "    print(f\"Saved addl metadata file to: {output_path_addl_metadata}\\n\")\n",
    "\n",
    "    return "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "adb38a85",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Saved primary candidate data file to: gpa_classification_primary_cand_dataset.csv\n",
      "\n",
      "Saved primary safety data file to: gpa_classification_primary_safety_dataset.csv\n",
      "\n",
      "Saved addl candidate data file to: gpa_classification_addl_cand_dataset.csv\n",
      "\n",
      "Saved addl safety data file to: gpa_classification_addl_safety_dataset.csv\n",
      "\n",
      "Saved primary metadata file to: primary_metadata_classification.json\n",
      "\n",
      "Saved addl metadata file to: addl_metadata_classification.json\n",
      "\n"
     ]
    }
   ],
   "source": [
    "f = \"~/beri/code/notebooks/gpa_data_download/data.csv\"\n",
    "columns_raw = [\"Gender\", \"Physics\", \"Biology\", \n",
    "    \"History\", \"Second_Language\", \"Geography\", \n",
    "    \"Literature\", \"Portuguese_and_Essay\", \n",
    "    \"Math\", \"Chemistry\", \"GPA\"]\n",
    "\n",
    "make_primary_and_addl_dataset(\n",
    "    input_path=f,\n",
    "    output_path_primary_cand_data=\"gpa_classification_primary_cand_dataset.csv\",\n",
    "    output_path_primary_safety_data=\"gpa_classification_primary_safety_dataset.csv\",\n",
    "    output_path_addl_cand_data=\"gpa_classification_addl_cand_dataset.csv\",\n",
    "    output_path_addl_safety_data=\"gpa_classification_addl_safety_dataset.csv\",\n",
    "    output_path_primary_metadata=\"primary_metadata_classification.json\",\n",
    "    output_path_addl_metadata=\"addl_metadata_classification.json\",\n",
    "    columns_raw=columns_raw\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "754a0ddf",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "seldo_py310",
   "language": "python",
   "name": "seldo_py310"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}