SHAP plots for a modelblueprint

Calls shap() using the modelblueprint's model, data, and pipeline slots.

Usage

# S3 method for class 'modelblueprint'
shap(
  data,
  vars = NA,
  set = c("train", "test", "holdout"),
  type = c("importance", "dependence"),
  nsim = 50L,
  sample_size = 500L,
  bins = 10L,
  type_agg = c("equal_exposure", "equal_range"),
  ret = c("plot", "data"),
  ...
)

Arguments

data

A modelblueprint.

vars

[character] Features to compute SHAP for. Defaults to data@x_original_inputs when NA.

set

[character(1)] Which dataset to use: "train" (default), "test", or "holdout".

type

[character(1)] "importance" (default) or "dependence". See shap() for details.

nsim

[integer(1)] Monte Carlo permutations per row. Default 50L.

sample_size

[integer(1)] Rows sampled for SHAP computation. Default 500L.

bins

[integer(1)] Number of bins for the dependence plot x-axis. Default 10L.

type_agg

[character(1)] Binning strategy: "equal_exposure" (default) or "equal_range".

ret

[character(1)] "plot" (default) or "data".

...

Further arguments passed to shap().

Value

A plotly object, a named list of plotly objects, or a data.table depending on type and ret.