predict.fasi: Prediction of a FASI Object

In bradleyrava/fasi: Implements the Fair Adjusted Selective Inference (FASI) procedure

Prediction of a FASI Object

Description

After a model is trained with the fasi function, predict estimates the r-scores and classification of all observations in the test data set.

Usage

## S3 method for class 'fasi'
predict(
  object,
  test_data,
  alpha_1,
  alpha_2,
  rscore_plus = T,
  ptd_group_var = "a",
  class_label = "y",
  ranking_score_calibrate,
  ranking_score_test,
  indecision_choice = "2",
  ...
)

Arguments

object	An object of class fasi. It can be created from the fasi function.
test_data	The test data set that contains new observations to be classified.
alpha_1	User specified group and overall FSR control for class 1.
alpha_2	User specified group and overall FSR control for class 2.
rscore_plus	A logical variable that indicates if the r-score or r-score plus is calculated. By default the r-score plus is calculated.
ptd_group_var	The name of the protected group variable in your data set. Defaults to "a".
class_label	The name of the class label variable in your data set. Defaults to "y".
ranking_score_calibrate	A vector of ranking scores for the calibration data set. This should only be used if the built in ranking score algorithms are not used.
ranking_score_test	A vector of ranking scores for the test data set. This should only be used if the built in ranking score algorithms are not used.
indecision_choice	A number, 1, 2, or 3. This determines how the indecisions are treated if we are equally confident in placing them in both class 1 and 2. Defaults to the scenario where class 2 is perfered.
...	Additional arguments

Author(s)

Bradley Rava. PhD Candidate at the University of Southern California's Marshall School of Business. Department of Data Sciences and Operations.

Examples

fasi_object <- fasi(observed_data, model_formula, split_p=0.5, alg="gam", class_label="y")
predict(fasi_object, test_data, alpha_1=0.1, alpha_2=0.1)

bradleyrava/fasi documentation built on May 12, 2024, 6:23 a.m.