Skip to content

Commit

Permalink
Add example for unbiased naive bayes
Browse files Browse the repository at this point in the history
  • Loading branch information
@kyzpanos17 @gciatto
kyzpanos17 authored and gciatto committed Feb 2, 2024
1 parent f42dc29 commit 9e4dd3b
Showing 1 changed file with 354 additions and 0 deletions.
354 changes: 354 additions & 0 deletions examples/mitigation_model_unbiased_naive_bayes.ipynb
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,354 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"import sys\n",
"sys.path.append(\"../\")"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import pandas as pd\n",
"pd.set_option('display.width', 500)\n",
"import aequitas.detection.descriptive_stats as dstats\n",
"import aequitas.detection.metrics as metrics\n",
"import aequitas.mitigation.data as technique\n",
"import aequitas.mitigation.models as model\n",
"import aequitas.tools.data_manip as dm\n",
"import aequitas.tools as tools"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"#Import dataset\n",
"dataset_name=\"Census_Income_Dataset.csv\"\n",
"dataset_directory=\"../datasets/\"+dataset_name\n",
"dataset = pd.read_csv(dataset_directory)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"# Dataset Pre-Processing\n",
"\n",
"# remove fnlwgt column (per instructions)\n",
"dataset = dataset.drop('fnlwgt', axis=1)\n",
"\n",
"# remove education column since there is an educution_num\n",
"dataset = dataset.drop('education', axis=1)\n",
"\n",
"# impute the missing values\n",
"num_data = dataset.shape[0]\n",
"col_names = dataset.columns\n",
"for c in col_names:\n",
"\tdataset[c] = dataset[c].replace(\"?\", np.NaN)\n",
"dataset = dataset.apply(lambda x:x.fillna(x.value_counts().index[0]))"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Dataset:\n",
" Column Name Data Type Column Type (suggestion) Number_Values Values\n",
"0 age int64 Continuous 74 -\n",
"1 workclass text Categorical/Ordinal 8 [Private, Local-gov, Self-emp-not-inc, Federal...\n",
"2 educational-num int64 Categorical/Ordinal 16 [7, 9, 12, 10, 6, 15, 4, 13, 14, 16, 3, 11, 5,...\n",
"3 marital-status text Categorical/Ordinal 7 [Never-married, Married-civ-spouse, Widowed, D...\n",
"4 occupation text Categorical/Ordinal 14 [Machine-op-inspct, Farming-fishing, Protectiv...\n",
"5 relationship text Categorical/Ordinal 6 [Own-child, Husband, Not-in-family, Unmarried,...\n",
"6 race text Categorical/Ordinal 5 [Black, White, Asian-Pac-Islander, Other, Amer...\n",
"7 gender text Binary 2 [Male, Female]\n",
"8 capital-gain int64 Continuous 123 -\n",
"9 capital-loss int64 Continuous 99 -\n",
"10 hours-per-week int64 Continuous 96 -\n",
"11 native-country text Categorical/Ordinal 41 [United-States, Peru, Guatemala, Mexico, Domin...\n",
"12 income text Binary 2 [<=50K, >50K]\n"
]
}
],
"source": [
"# display dataset structure\n",
"dataset_struct=dstats.analyse_dataset(dataset,verbose=True)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
"# split dataset to training and test samples\n",
"training_sample,test_sample = dm.split_dataset(dataset,ratio=0.2,random_state=51)\n",
"\n",
"# Define appropriate transformations for dataset (feature encoding and scaling if required)\n",
"# Define appropriate transformations for dataset (feature encoding and scaling if required)\n",
"transform_dict = {\n",
" \"income\": {\n",
" \"encode\": \"labeling\",\n",
" \"labels\": {\n",
" \"<=50K\": 0,\n",
" \">50K\": 1, \n",
" }\n",
" },\n",
" \"gender\": {\n",
" \"encode\": \"labeling\",\n",
" \"labels\": {\n",
" \"Female\": 0,\n",
" \"Male\": 1, \n",
" }\n",
" },\n",
" \"workclass\": {\n",
" \"encode\": \"labeling\",\n",
" },\n",
" \"race\": {\n",
" \"encode\": \"labeling\", \n",
" },\n",
" \"marital-status\": {\n",
" \"encode\": \"labeling\",\n",
" },\n",
" \"occupation\": {\n",
" \"encode\": \"labeling\", \n",
" },\n",
" \"relationship\": {\n",
" \"encode\": \"labeling\", \n",
" },\n",
" \"native-country\": {\n",
" \"encode\": \"labeling\", \n",
" },\n",
" \"age\":{\n",
" \"scaling\": \"min-max\"\n",
" },\n",
" \"educational-num\":{\n",
" \"scaling\": \"min-max\"\n",
" },\n",
" \"capital-gain\":{\n",
" },\n",
" \"capital-loss\":{\n",
" },\n",
" \"hours-per-week\":{\n",
" \"scaling\": \"min-max\"\n",
" }\n",
"}\n",
"\n",
"# Transform the training sample\n",
"training_sample, transformers = dm.transform_training_data(training_sample, transform_dict)\n",
"\n",
"# Transform the test sample\n",
"test_sample = dm.transform_test_data(test_sample, transform_dict, transformers)"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [],
"source": [
"# define fairness parameters\n",
"class_attribute='income'\n",
"sensitive_attribute='gender'\n",
"outcome=1 # >50K\n",
"priv_group=1 # Male"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Training sample:\n",
"Statistical/Demographic Parity:\n",
"Outcome: 1\n",
" 1 0\n",
"1 0.0 0.196285\n",
"\n",
"\n",
"Test sample:\n",
"Statistical/Demographic Parity:\n",
"Outcome: 1\n",
" 1 0\n",
"1 0.0 0.187467\n",
"\n",
"\n"
]
}
],
"source": [
"# measure statistical parity before classification\n",
"print(\"Training sample:\")\n",
"res=metrics.statistical_parity(training_sample,class_attribute,sensitive_attribute,positive_outcome=outcome,privileged_group=priv_group,verbose=True)\n",
"\n",
"print(\"Test sample:\")\n",
"res=metrics.statistical_parity(test_sample,class_attribute,sensitive_attribute,positive_outcome=outcome,privileged_group=priv_group,verbose=True)"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [],
"source": [
"# define classifier parameters\n",
"classifier_type=\"Naive_Bayes\"\n",
"normal_feat = [\"age\",\"educational-num\",\"hours-per-week\"]\n",
"classifier_params= {\n",
" \"normal_features\":normal_feat\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Classifier Accuracy: 0.86\n"
]
}
],
"source": [
"# Train a classifier on training sample\n",
"clf=tools.train_classifier(training_sample,class_attribute,classifier_type,classifier_params)\n",
"\n",
"# Test classifier on test sample\n",
"predicted_test_sample, _, _, _= tools.test_classifier(clf,test_sample,class_attribute,verbose=True)\n",
"prediction=np.array(predicted_test_sample[class_attribute])"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Statistical/Demographic Parity:\n",
"Outcome: 1\n",
" 1 0\n",
"1 0.0 0.244571\n",
"\n",
"\n"
]
}
],
"source": [
"# measure statistical parity after classification on test sample\n",
"res=metrics.statistical_parity(predicted_test_sample,class_attribute,sensitive_attribute,positive_outcome=outcome,privileged_group=priv_group,verbose=True)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Mitigation unbiased Naive Bayes Model!\n",
"Step: 0 Discrimination: 0.2510332455486799 \n",
"Step: 5 Discrimination: 0.2340343272980231 \n",
"Step: 10 Discrimination: 0.2192761755440438 \n",
"Step: 15 Discrimination: 0.20374534568776193 \n",
"Step: 20 Discrimination: 0.18249669787444095 \n",
"Step: 25 Discrimination: 0.152672653697833 \n",
"Step: 30 Discrimination: 0.11540482643624986 \n",
"Step: 35 Discrimination: 0.060157934079452 \n",
"Final step: 39 Discrimination: 0.0028079451036868153 \n",
"Classifier Accuracy: 0.83\n"
]
}
],
"source": [
"# Train a bias mitigated re-weighting classifier\n",
"print(\"Mitigation unbiased Naive Bayes Model!\")\n",
"clf=model.unbiasedNaiveBayes(training_sample,class_attribute,sensitive_attribute,normal_features=normal_feat)\n",
"\n",
"# Test a classifier\n",
"predicted_test_sample, _, _, _= tools.test_classifier(clf,test_sample,class_attribute,verbose=True)\n",
"prediction=np.array(predicted_test_sample[class_attribute])"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Statistical/Demographic Parity:\n",
"Outcome: 1\n",
" 1 0\n",
"1 0.0 -0.00886\n",
"\n",
"\n"
]
}
],
"source": [
"\n",
"# Measure Discrimination on predicted test sample\n",
"res=metrics.statistical_parity(predicted_test_sample,class_attribute,sensitive_attribute,positive_outcome=outcome,privileged_group=priv_group,verbose=True)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3.11.5 ('aequitas')",
"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": "3.11.5"
},
"orig_nbformat": 4,
"vscode": {
"interpreter": {
"hash": "aafc4311810ab362ec8b20f1ad7cefee81be5411161712b32fbd26f1f27127c5"
}
}
},
"nbformat": 4,
"nbformat_minor": 2
}

0 comments on commit 9e4dd3b

Please sign in to comment.