genSurface: Generate response surface for 2D interaction

In karlkumbier/rf_surface: Random Forest Response Surface

Generate response surface for 2D interaction

Usage

genSurface(
  x,
  int,
  y = NULL,
  fit = NULL,
  read.forest = NULL,
  rectangles = NULL,
  wt.node = "size",
  varnames = colnames(x),
  nbin = 100,
  bins = NULL,
  binFun = NULL,
  yscale = 1,
  yFun = function(x) return(x),
  filter.rules = NULL
)

Arguments

x	numeric feature matrix, with replicate features grouped
int	signed interaction to plot. Formatted as 'X1+_X2+_X3-_...'
y	response vector to be visualzed. If NULL, genSurface will use random forest predictions associated with each decision rule
fit	a fitted random forest, from packages randomForest or ranger.
read.forest	output of readForest.
rectangles	a list of hyperrectangles corresponding to leaf nodes in an RF, as retuned by forestHR. If both rectangles and read.forest are supplied, read.forest will be ignored.
wt.node	indicator for how nodes are to be weighted in response surfaces. One of 'size' - weighting proportional to leaf node size or 'none' - indicating uniform weighting.
varnames	character vector indicating feature names. If NULL, colnames(x) are used as feature names.
nbin:	number of bins to plot surface map over
bins:	user generated grid to plot over. If supplied, nbin is ignored
filter.rules:	a list of filtering functions to be applied to rf decision paths. If NULL, default rules will filter to a random sample of 10

Response surface give E(Y|X) over regions learned by an iRF.

karlkumbier/rf_surface documentation built on April 17, 2025, 5:07 p.m.