ale {distillML}

Constructs an ALE for a model.

Description

Constructs an ALE for a model.

Usage

ale(
  predict_function,
  num_grid_points,
  training_data,
  variable_names,
  center = "zero",
  grid_points,
  window_size
)

Arguments

predict_function

a function taking a single tibble argument and returning the model predictions corresponding to that tibble.

num_grid_points

the number of grid_points at which to construct the ALE

training_data

the training data used to fit the model

variable_names

a vector of feature names in training data for which an ALE is required.

center

one of "uncentered" meaning the plots are not centered, "mean" meaning the plots are centered at their mean and "zero" meaning the ALE passes through the origin. When using center == "zero" we recommend setting window_size because otherwise a smaller and possibly empty set will be used to compute the ALE at zero.

grid_points

The grid points to use for the AlE calculation.

window_size

the fraction of the data (between zero and one) used to compute each ALE point.