plot.local_importance: Plot Local Variable Importance measure

In ModelOriented/vivo: Variable Importance via Oscillations

Description

Function plot.local_importance plots local importance measure based on Ceteris Paribus profiles.

Usage

## S3 method for class 'local_importance'
plot(
  x,
  ...,
  variables = NULL,
  color = NULL,
  type = NULL,
  title = "Local variable importance"
)

Arguments

x	object returned from local_variable_importance() function
...	other object returned from local_variable_importance() function that shall be plotted together
variables	if not NULL then only variables will be presented
color	a character. How to aggregated measure? Either "_label_method_" or "_label_model_".
type	a character. How variables shall be plotted? Either "bars" (default) or "lines".
title	the plot's title, by default 'Local variable importance'

Value

a ggplot2 object

Examples

library("DALEX")
data(apartments)

library("randomForest")
apartments_rf_model <- randomForest(m2.price ~ construction.year + surface +
                                    floor + no.rooms, data = apartments)

explainer_rf <- explain(apartments_rf_model, data = apartmentsTest[,2:5],
                        y = apartmentsTest$m2.price)

new_apartment <- data.frame(construction.year = 1998, surface = 88, floor = 2L, no.rooms = 3)

profiles <- predict_profile(explainer_rf, new_apartment)

library("vivo")
measure1 <- local_variable_importance(profiles, apartments[,2:5],
                          absolute_deviation = TRUE, point = TRUE, density = FALSE)

plot(measure1)

measure2 <- local_variable_importance(profiles, apartments[,2:5],
                          absolute_deviation = TRUE, point = TRUE, density = TRUE)
plot(measure1, measure2, color = "_label_method_", type = "lines")

ModelOriented/vivo documentation built on Sept. 29, 2020, 10:53 p.m.