label_artificial_data_regr: Label a dataset using a supplied learner.

In pfistfl/distillery: Compress arbitrary mlr3 models into keras models

Description

Uses the provided learner in order to predict on the provided task's features. 'prob = TRUE' implements Dark Knowledge as implmented by Hinton, 2015. If 'prob', predicts probabilities, else response. The mixing parameter 'alpha' corresponds to a trade-off between the predicted target and the actual target, where 0 only uses predicted targets (default) and 1 only uses the true label.

Usage

label_artificial_data_regr(task, teacher, alpha = 0)

Arguments

task	['Task'] Task to create labels for.
teacher	['Learner'] Trained learner to compress into a neural network.
alpha	['numeric'] Specifies a convex-combination of labels predicted by the learner (alpha = 0) and labels of the observation before swapping.

Value

A ['vector'] containing the response for each observation

pfistfl/distillery documentation built on April 17, 2021, 10 p.m.