Partial dependence plot for a modelblueprint

Calls pdp() using the modelblueprint's model, target, exposure, and data slots.

Usage

# S3 method for class 'modelblueprint'
pdp(
  data,
  var = NA,
  set = c("train", "test", "holdout"),
  bins = 10L,
  sample_size = 10000L,
  type_agg = c("equal_exposure", "equal_range"),
  ret = c("plot", "data"),
  ...
)

Arguments

data

A modelblueprint.

var

[character(1)] Feature to compute the PDP for.

set

[character(1)] Dataset to use: "train", "test", or "holdout". Default "train".

bins

[integer(1)] Number of bins. Default 10L.

sample_size

[integer(1)] Rows to sample. Default 10000L.

type_agg

[character(1)] "equal_exposure" or "equal_range".

ret

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

...

Further arguments passed to pdp().

Value

A plotly object or data.table depending on ret.

Performance

When @x_original_inputs is set, the working dataset is narrowed to those columns (plus the target and exposure) before scoring, which avoids copying unused columns on wide frames. This assumes feat_eng_fun only consumes the declared @x_original_inputs; if your feature engineering reads other columns, leave @x_original_inputs unset so the full frame is used.