plot.individual_variable_profile_explainer: Plots Individual Variable Profile Explanations

In pbiecek/ceterisParibus2: Ceteris Paribus Profiles

Description

Function 'plot.individual_variable_profile_explainer' plots Individual Variable Profiles for selected observations. Various parameters help to decide what should be plotted, profiles, aggregated profiles, points or rugs.

Usage

## S3 method for class 'individual_variable_profile_explainer'
plot(x, ..., size = 1,
  alpha = 0.3, color = "black", size_points = 2, alpha_points = 1,
  color_points = color, size_rugs = 0.5, alpha_rugs = 1,
  color_rugs = color, size_residuals = 1, alpha_residuals = 1,
  color_residuals = color, only_numerical = TRUE,
  show_profiles = TRUE, show_observations = TRUE, show_rugs = FALSE,
  show_residuals = FALSE, aggregate_profiles = NULL,
  facet_ncol = NULL, selected_variables = NULL)

Arguments

x	a ceteris paribus explainer produced with function 'individual_variable_profile()'
...	other explainers that shall be plotted together
size	a numeric. Size of lines to be plotted
alpha	a numeric between 0 and 1. Opacity of lines
color	a character. Either name of a color or name of a variable that should be used for coloring
size_points	a numeric. Size of points to be plotted
alpha_points	a numeric between 0 and 1. Opacity of points
color_points	a character. Either name of a color or name of a variable that should be used for coloring
size_rugs	a numeric. Size of rugs to be plotted
alpha_rugs	a numeric between 0 and 1. Opacity of rugs
color_rugs	a character. Either name of a color or name of a theme_mi2variable that should be used for coloring
size_residuals	a numeric. Size of line and points to be plotted for residuals
alpha_residuals	a numeric between 0 and 1. Opacity of points and lines for residuals
color_residuals	a character. Either name of a color or name of a variable that should be used for coloring for residuals
only_numerical	a logical. If TRUE then only numerical variables will be plotted. If FALSE then only categorical variables will be plotted.
show_profiles	a logical. If TRUE then profiles will be plotted. Either individual or aggregate (see 'aggregate_profiles')
show_observations	a logical. If TRUE then individual observations will be marked as points
show_rugs	a logical. If TRUE then individual observations will be marked as rugs
show_residuals	a logical. If TRUE then residuals will be plotted as a line ended with a point
aggregate_profiles	function. If NULL (default) then individual profiles will be plotted. If a function (e.g. mean or median) then profiles will be aggregated and only the aggregate profile will be plotted
facet_ncol	number of columns for the 'facet_wrap()'
selected_variables	if not NULL then only 'selected_variables' will be presented

Value

a ggplot2 object

Examples

library("DALEX2")
 ## Not run: 
library("randomForest")
set.seed(59)

apartments_rf <- randomForest(m2.price ~ construction.year + surface + floor +
                                no.rooms + district, data = apartments)
explainer_rf <- explain(apartments_rf,
                        data = apartments_test[,2:6], y = apartments_test$m2.price)

apartments_lm <- lm(m2.price ~ construction.year + surface + floor +
                                no.rooms + district, data = apartments)
explainer_lm <- explain(apartments_lm,
                        data = apartments_test[,2:6], y = apartments_test$m2.price)

library("e1071")
apartments_svm <- svm(m2.price ~ construction.year + surface + floor +
                                no.rooms + district, data = apartments)
explainer_svm <- explain(apartments_svm,
                        data = apartments_test[,2:6], y = apartments_test$m2.price)

# individual explanations
my_apartment <- apartments_test[1, ]

# for random forest
lp_rf <- individual_variable_profile(explainer_rf, my_apartment)
lp_rf

plot(lp_rf)

# for others
lp_lm <- individual_variable_profile(explainer_lm, my_apartment)
plot(lp_rf, lp_lm, color = "_label_")

# for others
lp_svm <- individual_variable_profile(explainer_svm, my_apartment)
plot(lp_rf, lp_lm, lp_svm, color = "_label_")

# more parameters
plot(lp_rf, lp_lm, lp_svm, color = "_label_",
   show_profiles = TRUE, show_observations = TRUE,
   show_rugs = TRUE,
   alpha = 0.3, alpha_points = 1,
   size_points = 5, size_rugs = 0.5)

plot(lp_rf, show_profiles = TRUE, show_observations = TRUE,
   color = "black",
   selected_variables = c("construction.year", "surface"),
   alpha = 0.3, alpha_points = 1, alpha_residuals = 0.5,
   size_points = 5, size_rugs = 0.5)

plot(lp_rf, show_profiles = TRUE, show_observations = TRUE,
   color = "surface",
   selected_variables = c("construction.year", "surface"),
   alpha = 0.3, alpha_points = 1, alpha_residuals = 0.5,
   size_points = 5, size_rugs = 0.5)

plot(lp_rf, show_profiles = TRUE, show_observations = TRUE,
   color = "darkblue", aggregate_profiles = mean,
   selected_variables = c("construction.year", "surface"),
   alpha = 0.3, alpha_points = 1, alpha_residuals = 0.5,
   size_points = 5, size_rugs = 0.5)

# --------
# multiclass

HR_rf <- randomForest(status ~ . , data = HR)
explainer_rf <- explain(HR_rf, data = HRTest, y = HRTest)

my_HR <- HRTest[1, ]

lp_rf <- individual_variable_profile(explainer_rf, my_HR)
lp_rf

plot(lp_rf, color = "_label_")

## End(Not run)

pbiecek/ceterisParibus2 documentation built on Sept. 16, 2019, 6:26 p.m.