permutation_importance: Plot permutation importance

In mirka-henninger/InterpretationMethods: Facilitates the implementation of interpretation techniques for social science research

Description

This function allows to plot permutation importance based on mean squared error. Three types are implemented: model-based permutation importance, random forest permutation importnance or conditional random forest variable importance

Usage

permutation_importance(
  object,
  type,
  loss = NULL,
  xlabel = "",
  ylabel = type,
  title = "",
  limits = c(NA, NA)
)

Arguments

object	A prediction object from package iml for model-based permutation importance or an object as returned by randomForest or cforest
type	A character indicating the type of variable importance: model-agnostic, randomforest, or conditional
loss	A character specifying the loss function for type = model-agnostic. See ?iml::FeatureImp for more details
xlabel	An optional character string indicating the x-axis label
ylabel	An optional character string indicating the y-axis label
title	An optional character string indicating the title of the plot
limits	An optional two-entry vector indicating the limits of the y-axis

Value

a plot of type ggplot

Examples

## Not run: 
N <- 1000
x1 <- runif(N, -1, 1)
x2 <- runif(N, -1, 1)
x3 <- x2 + runif(N, -1, 1)
y <- 5 + 5 * x1 + 5 * x2 + 0 * x3 + rnorm(N,1)
dat <- data.frame(x1,x2,x3,y)
rfmod <- randomForest::randomForest(y~., dat)
pred <- iml::Predictor$new(rfmod)
permutation_importance(pred, type = "model-agnostic", limits = c(0,18), loss = "mse")
permutation_importance(rfmod, type = "randomforest", limits = c(0,18))
permutation_importance(rfmod, type = "conditional", limits = c(0,18))

## End(Not run)

mirka-henninger/InterpretationMethods documentation built on Jan. 12, 2022, 4:10 p.m.