holisticai.bias.metrics.multiclass_statistical_parity#

holisticai.bias.metrics.multiclass_statistical_parity(p_attr, y_pred, groups=None, classes=None, aggregation_fun='mean')[source]#

Multiclass statistical parity

This function computes statistical parity for a classification task with multiple classes and a protected attribute with multiple groups. For each group compute the vector of success rates for entering each class. Compute all distances (mean absolute deviation) between such vectors. Then aggregate them using the mean, or max strategy.

Interpretation

The accepted values and bounds for this metric are the same as the 1d case. A value of 0 is desired. Values below 0.1 are considered fair.

Parameters

p_attrarray-like: Multiclass protected attribute vector.
y_predarray-like: Prediction vector (categorical).
groupslist, optional: Unique groups from p_attr in order
classeslist, optional: The unique output classes in order
aggregation_funstr, optional: The function to aggregate across groups (‘mean’ or ‘max’)

Returns

float: Multiclass Statistical Parity

Examples

>>> import numpy as np
>>> from holisticai.bias.metrics import multiclass_statistical_parity
>>> p_attr = np.array(["A", "A", "A", "A", "B", "B", "B", "B", "C", "C"])
>>> y_pred = np.array([0, 1, 2, 0, 1, 2, 0, 1, 2, 0])
>>> y_true = np.array([0, 1, 1, 0, 1, 0, 2, 1, 2, 1])
>>> multiclass_statistical_parity(p_attr, y_pred, aggregation_fun="max")
0.5