ingredients (version 2.3.0)

plot.ceteris_paribus_2d_explainer: Plot Ceteris Paribus 2D Explanations

Description

This function plots What-If Plots for a single prediction / observation.

Usage

# S3 method for ceteris_paribus_2d_explainer
plot(
  x,
  ...,
  facet_ncol = NULL,
  add_raster = TRUE,
  add_contour = TRUE,
  bins = 3,
  add_observation = TRUE,
  pch = "+",
  size = 6
)

Value

a ggplot2 object

Arguments

x

a ceteris paribus explainer produced with the ceteris_paribus_2d() function

...

currently will be ignored

facet_ncol

number of columns for the facet_wrap

add_raster

if TRUE then geom_raster will be added to present levels with diverging colors

add_contour

if TRUE then geom_contour will be added to present contours

bins

number of contours to be added

add_observation

if TRUE then geom_point will be added to present observation that is explained

pch

character, symbol used to plot observations

size

numeric, size of individual datapoints

References

Explanatory Model Analysis. Explore, Explain, and Examine Predictive Models. https://ema.drwhy.ai/

Examples

library("DALEX")
library("ingredients")
library("ranger")

# \donttest{
apartments_rf_model <- ranger(m2.price ~., data = apartments)

explainer_rf <- explain(apartments_rf_model,
                        data = apartments_test[,-1],
                        y = apartments_test[,1],
                        verbose = FALSE)

new_apartment <- apartments_test[1,]
new_apartment

wi_rf_2d <- ceteris_paribus_2d(explainer_rf, observation = new_apartment)
head(wi_rf_2d)

plot(wi_rf_2d)
plot(wi_rf_2d, add_contour = FALSE)
plot(wi_rf_2d, add_observation = FALSE)
plot(wi_rf_2d, add_raster = FALSE)

# HR data
model <- ranger(status ~ gender + age + hours + evaluation + salary, data = HR,
                probability = TRUE)

pred1 <- function(m, x)   predict(m, x)$predictions[,1]

explainer_rf_fired <- explain(model,
                              data = HR[,1:5],
                              y = as.numeric(HR$status == "fired"),
                              predict_function = pred1,
                              label = "fired")
new_emp <- HR[1,]
new_emp

wi_rf_2d <- ceteris_paribus_2d(explainer_rf_fired, observation = new_emp)
head(wi_rf_2d)

plot(wi_rf_2d)
# }