Skip to contents

Plots the PDP, ALE, or ICE plots for an Interpreter object

Usage

# S3 method for Interpreter
plot(
  x,
  method = "pdp+ice",
  features = NULL,
  features.2d = NULL,
  clusters = NULL,
  clusterType = "preds",
  smooth = FALSE,
  smooth.bandwidth = NULL,
  smooth.kernel = "normal",
  smooth.npoints = 2 * x$grid.size,
  ...
)

Arguments

x

Interpreter object to generate plots from

method

The type of plot that we want to generate. Must be one of "ice", "pdp+ice", "pdp", or "ale"

features

a vector of feature names that we want to produce 1-D plots for.

features.2d

2-D features that we want to produce plots for arguments. A two-column dataframe of pairs of features to make local surrogates for. Each row represents a pair of features, with the names of features as the entries.If the 'method' parameter is set to "ale", this argument should not be used.

clusters

A number of clusters to cluster the ICE predictions with. If this is not NULL, one must use the method "ice".

clusterType

An indicator specifying what method to use for the clustering. The possible options are "preds", and "gradient". If "preds" is used, the clusters will be determined by running K means on the predictions of the ICE functions. If the "gradient" option is used, the clusters will be determined by running K means on the numerical gradient of the predictions of the ICE functions. If this is not NULL, one must use the method "ice".

smooth

A binary variable to determine whether to smoothen the plots of the PDP, ICE, or ALE curves for continuous variables.

smooth.bandwidth

The bandwidth for the kernels. They are scaled such that their quartiles are at 0.25 * bandwidth. By default, this is set as the maximum difference between the minimum and maximum of the grid points.

smooth.kernel

The type of kernel to be used. Users can input either strings "box" or "normal". The default is "normal".

smooth.npoints

The number of points returned when using the kernel method. By default, this is twice the number of grid points for that feature.

...

Additiional parameters to pass to the plot function

Value

A list of plots with 1-d features and 2-d features. For 2-d features with one continuous and one categorical feature, the plot is a linear plot of the continuous feature with group colors representing the categorical feature. For two continuous features, the plot is a heatmap with the shade representing the value of the outcome.