fairmodels (version 1.1.0)

pre_process_data: Pre-process data

Description

Function aggregates all pre-processing algorithms for bias mitigation. User passes unified arguments and specifies type to receive transformed data.frame

Usage

pre_process_data(data, protected, y, type = "resample_uniform", ...)

Arguments

data

data.frame

protected

factor, protected attribute (sensitive variable) containing information about gender, race etc...

y

numeric, numeric values of predicted variable. 1 should denote favorable outcome.

type

character, type of pre-processing algorithm to be used, one of:

• resample_uniform

• resample_preferential

• reweight

• disparate_impact_remover

...

other parameters passed to pre-processing algorithms

Value

modified data (data.frame). In case of type = 'reweight' data has feature `_weights_` containing weights that need to be passed to model. In other cases data is ready to be passed as training data to a model.

Examples

# NOT RUN {
data("german")

pre_process_data(german,
                 german$Sex,
                 as.numeric(german$Risk)-1,
                 type = "disparate_impact_remover",
                 features_to_transform = 'Age')
# }