pdp2d: Explanation gap visualization

In g-rho/eXplaAInability: Numerical and visual tools to analyse the explainability of partial dependence functions

Description

Visualization of 2D PDP vs. unexplained residual predictions.

Usage

pdp2d(
  model,
  x,
  vnames,
  type = "pdp",
  depth = 21,
  alpha = 2/3,
  right = 0.8,
  top = 0.95,
  digits = 1,
  ...
)

Arguments

model	A model with corresponding predict function that returns numeric values.
x	Data frame.
vnames	Character vector of the variable set for which the patial dependence function is to be computed.
type	Character, either "pdp", "gap" or "both" specifying the meaning of the colours of scatter: either the value of the PD function or the differences between PD and the model's predcitions. In case of "both" two plots are created.
depth	Integer specifiying the number colours in the heat map.
alpha	Numeric value for alpha blending of the points in the scatter plot.
right	Position where to place the legend relative to the range of the x axis.
top	Position where to place the legend relative to the range of the y axis.
digits	Nuber of digits for rounding in the legend.
...	Further arguments to be passed to the predict() method of the model.

Author(s)

[e-mail]

References

• Szepannek, G. (2019): How Much Can We See? A Note on Quantifying Explainability of Machine Learning Models, arXiv:1910.13376 [stat.ML].

Examples

## Not run: 
library(pdp)
library(randomForest)
data(boston)
set.seed(42)
boston.rf <- randomForest(cmedv ~ ., data = boston)
pdp2d(boston.rf, boston, c("lstat", "rm"), type = "both")

## End(Not run)

g-rho/eXplaAInability documentation built on April 6, 2021, 1:42 a.m.