fairlearn.reductions.ErrorRate

class fairlearn.reductions.ErrorRate(*, costs=None)

Misclassification error as a moment.

A classifier \(h(X)\) has the misclassification error equal to

\[P[h(X) \ne Y]\]

It is also possible to specify costs for false positives and false negatives. The error then evaluates to

\[c_{FP} P[h(X)=1, Y=0] + c_{FN} P[h(X)=0, Y=1]\]

where \(c_{FP}\) and \(c_{FN}\) are the costs of false positive and false negative errors respectively. The default misclassification error corresponds to \(c_{FP}=c_{FN}=1.0\).

Read more in the User Guide.

Added in version 0.3.0.

Parameters:

costs (dict[Literal['fp', 'fn'], float] | None) – Dictionary with keys 'fp' and 'fn' containing the costs of false positives and false negatives. If none are provided costs of 1.0 are assumed.

bound()

Return vector of fairness bound constraint the length of gamma.

Return type:

Series

default_objective()

Return the default objective for the moment.

Return type:

Moment

gamma(predictor)

Calculate a vector of moments.

When ErrorRate() is used as a constraint, then gamma[j]≤0 for all j is used as the set of constraints. When ErrorRate() is used as an objective, then gamma[0] is used as the objective.

Parameters:

predictor (Callable) – bound function returning predictions

Return type:

Series

Returns:

error (pandas.Series) – gamma value for the predictor

load_data(X, y, *, sensitive_features)

Load the specified data into the object.

Parameters:

• X (array of shape (n_samples, n_features)) – The feature array

• y (array_like of shape (n_samples,)) – The label vector

• sensitive_features (array_like of shape (n_samples, n_sensitive_features)) – The sensitive feature vector

Return type:

None

project_lambda(lambda_vec)

Return the lambda values.

Return type:

Series

signed_weights(lambda_vec=None)

Return the signed weights.

Return type:

Series

property index: Index

Return the index listing the constraints.

property total_samples: int

Return the number of samples in the data.