holisticai.bias.mitigation.DebiasingExposure#
- class holisticai.bias.mitigation.DebiasingExposure(group_col: str, query_col='query_id', doc_col='doc_id', score_col='judgment', feature_cols=None, gamma: float = 1.0, number_of_iterations=3000, learning_rate=0.001, lambdaa=0.001, init_var=0.01, standardize=False, verbose=0)[source]#
Disparate Exposure Learning to Rank (DELTR) [1] incorporates a measure of performance and a measure of disparate exposure into its loss function. Trains a linear model based on performance and fairness for a protected group.
Parameters
- group_colstr
Name of the column in data that contains protected attribute.
- query_colstr
Name of the column in data that contains query ids (optional).
- doc_colstr
List of name of the column in data that contains document ids (optional).
- score_colstr
Name of the column in data that contains judgment values (optional).
- feature_cols :
Name of the columns in data that contains feature values (optional).
- gammafloat
Gamma parameter for the cost calculation in the training phase (recommended to be around 1).
- number_of_iterationsint
Number of iteration in gradient descent (optional).
- learning_ratefloat
Learning rate in gradient descent (optional).
- lambdaafloat
Regularization constant (optional).
- init_varfloat
Range of values for initialization of weights (optional).
- standardizebool
Boolean indicating whether the data should be standardized or not (optional).
- verboseint
If > 0, print progress.
References