fairlearn.metrics.equalized_odds_difference#

fairlearn.metrics.equalized_odds_difference(y_true, y_pred, *, sensitive_features, method='between_groups', sample_weight=None, agg='worst_case')[source]#

Calculate the equalized odds difference.

Depending on the agg parameter, this function computes either:

  • The greater of two metrics: true_positive_rate_difference and false_positive_rate_difference, if agg=’worst_case’. The former is the difference between the largest and smallest of \(P[h(X)=1 | A=a, Y=1]\), across all values \(a\) of the sensitive feature(s). The latter is defined similarly, but for \(P[h(X)=1 | A=a, Y=0]\).

  • The mean of these two metrics, if agg=’mean’.

The equalized odds difference of 0 means that all groups have the same true positive, true negative, false positive, and false negative rates.

Read more in the User Guide.

Added in version 0.4.6.

Parameters:

  • y_true (array_like) – Ground truth (correct) labels.

  • y_pred (array_like) – Predicted labels \(h(X)\) returned by the classifier.

  • sensitive_features (array_like) – The sensitive features over which equalized odds should be assessed

  • method (Literal['between_groups', 'to_overall']) – How to compute the differences. See fairlearn.metrics.MetricFrame.difference() for details.

  • sample_weight (array_like) – The sample weights

  • agg (Literal['worst_case', 'mean']) – The aggregation method. One of “worst_case” or “mean”. If “worst_case”, the greater one of the false positive rate difference and true positive rate difference is returned. If “mean”, the mean of the differences is returned.

Return type:

float

Returns:

The equalized odds difference

Raises:

ValueError – If agg is not one of 'worst_case' or 'mean'.