arthurai.core.bias.threshold_mitigation.ThresholdMitigation#
- class arthurai.core.bias.threshold_mitigation.ThresholdMitigation(arthur_model)#
Bases:
objectMethods
Triggers workflow to train curves for all constraints and for all sensitive attributes.
Fetch curves for the given sensitive attribute and constraint.
Retrieve the thresholds for each group, given a set of curves and aparticular optimization index.
Simple plot of the tradeoff curve for a single sensitive attribute and constraint.
- enable()#
Triggers workflow to train curves for all constraints and for all sensitive attributes. Equivalent to calling arthurmodel.enable_bias_mitigation(); here for convenience.
- get_curves(attribute, constraint='equal_opportunity')#
Fetch curves for the given sensitive attribute and constraint.
- Parameters
attribute (
str) – Name of the sensitive attribute to fetch curves for.constraint (
str) – which constraint to use. “demographic_parity”, “equal_opportunity”, or “equalized_odds”
- Returns
a Curves object which contains information about the attribute, constraint, accuracy-maximizing index, and the actual curves.
- get_thresholds_for_idx(curves, idx=None)#
Retrieve the thresholds for each group, given a set of curves and aparticular optimization index.
- Parameters
curves (
Curves) – one Curves object containing a set of curves corresponding to a single sensitive attribute and constraint.idx (
Optional[int]) – one integer (between 0 and 1000, inclusive) that represents a specific solution option. Defaults to the curve’s accuracy-maximizing index.
- Returns
a dictionary mapping sensitive feature values to the prediction threshold used for that group.