|View source on GitHub|
Absolute correlation between predictions on two groups of examples.
model_remediation.min_diff.losses.AbsoluteCorrelationLoss( name: Optional[Text] = None )
Name used for logging or tracking. Defaults to
Absolute correlation measures how correlated predictions are with membership (regardless of direction). The metric guarantees that the result is 0 if and only if the two distributions it is comparing are indistinguishable.
sensitive_group_labels input is used to determine whether each example
is part of the sensitive group. This currently only supports hard membership
For more details, see the paper.