Source code for holisticai.bias.mitigation.preprocessing.reweighing
from __future__ import annotations
from typing import Optional
import numpy as np
import pandas as pd
from holisticai.utils.transformers.bias import BMPreprocessing as BMPre
from holisticai.utils.transformers.bias import SensitiveGroups
[docs]
class Reweighing(BMPre):
"""
Reweighing preprocessing [1]_ weights the examples in each group-label combination to ensure fairness before\
classification.
Examples
--------
>>> from holisticai.bias.mitigation import Reweighing
>>> mitigator = Reweighing()
>>> train_data_transformed = mitigator.fit_transform(train_data, group_a, group_b)
>>> test_data_transformed = mitigator.transform(test_data, group_a, group_b)
References
----------
.. [1] Kamiran, Faisal, and Toon Calders. "Data preprocessing techniques for classification\
without discrimination." Knowledge and information systems 33.1 (2012): 1-33.
"""
def __init__(self):
self._sensgroups = SensitiveGroups()
[docs]
def fit(
self,
y: np.ndarray,
group_a: np.ndarray,
group_b: np.ndarray,
sample_weight: Optional[np.ndarray] = None,
):
"""
Fit the Reweighing model to the data. This method calculates the sample weights to ensure that the \
data is fair with respect to the specified sensitive groups before classification.
Parameters
----------
y : array-like
Target vector
group_a : array-like
Group membership vector (binary)
group_b : array-like
Group membership vector (binary)
sample_weight : array-like, optional
Samples weights vector. Default is None.
Returns
-------
Self
"""
params = self._load_data(y=y, sample_weight=sample_weight, group_a=group_a, group_b=group_b)
y = params["y"]
sample_weight = params["sample_weight"]
group_a = params["group_a"]
group_b = params["group_b"]
group_lbs = self._sensgroups.fit_transform(np.stack([group_a, group_b], axis=1))
classes = np.unique(y)
df = pd.DataFrame()
df["LABEL"] = pd.Series(y)
df["GROUP_ID"] = group_lbs
df["COUNT"] = 1
for g in self._sensgroups.group_names:
for c in classes:
df[f"{g}-{c}"] = (df["GROUP_ID"] == g) & (df["LABEL"] == c)
df_group_values = df.groupby(["GROUP_ID", "LABEL"])["COUNT"].sum()
df_values = df_group_values.groupby(level="LABEL").sum()
df_groups = df_group_values.groupby(level="GROUP_ID").sum()
df_group_values_prob = df_group_values / df_groups
df_values_prob = df_values / df_values.sum()
df_group_values_weights = df_values_prob / df_group_values_prob
self.sample_weight = np.ones_like(y, dtype=np.float32)
for g in self._sensgroups.group_names:
for c in classes:
mask = df[f"{g}-{c}"]
self.sample_weight[mask] = df_group_values_weights.at[g, c]
self._update_estimator_param("sample_weight", self.sample_weight)
return self
