Getting Started

Installation

Install with pip directly:

pip install fair-scoring

Introductory Example

The following example shows how compute the equal opportunity bias of the compas dataset

import pandas as pd
from fairscoring.metrics import bias_eo

# Load compas data
dataURL = 'https://raw.githubusercontent.com/propublica/compas-analysis/master/compas-scores-two-years.csv'
df = pd.read_csv(dataURL)

# Relevant data
scores = 11 - df['decile_score']
target = df['two_year_recid']
attribute = df['race']

# Compute the bias
bias = bias_eo(scores, target, attribute, groups=['African-American', 'Caucasian'],favorable_target=0)

To get a more detailed result, we can call

result = bias_eo.bias(scores, target, attribute, groups=['African-American', 'Caucasian'],favorable_target=0, n_permute=1000)

print(f"Bias: {result.bias:.3f}")
print(f"Pos: {100*result.pos_component:.0f}%")
print(f"Neg: {100*result.neg_component:.0f}%")
print(f"p-value: {result.p_value:.2f}")

Note

The information of the result of a bias-computation depends on the metric and also on the call. In this case setting n_permute=1000 leads to a permutation test, which results in p_value.