break_down_uncertainty: Explanation Level Uncertainty of Sequential Variable Attribution

Description

This function calculates the break down algorithm for B random orderings. Then it calculates the distribution of attributions for these different orderings. Note that the shap() function is just a simplified interface to the break_down_uncertainty() function with a default value set to B=25.

Usage

break_down_uncertainty(x, ..., keep_distributions = TRUE, B = 10)
# S3 method for explainer
break_down_uncertainty(
  x,
  new_observation,
  ...,
  keep_distributions = TRUE,
  B = 10
)
# S3 method for default
break_down_uncertainty(
  x,
  data,
  predict_function = predict,
  new_observation,
  label = class(x)[1],
  ...,
  path = NULL,
  keep_distributions = TRUE,
  B = 10
)
shap(x, ..., B = 25)

Value

an object of the break_down_uncertainty class.

Arguments

x: an explainer created with function explain or a model.
...: other parameters.
keep_distributions: if TRUE then we will keep distribution for predicted values. It's needed by the describe function.
B: number of random paths
new_observation: a new observation with columns that correspond to variables used in the model.
data: validation dataset, will be extracted from x if it is an explainer.
predict_function: predict function, will be extracted from x if it is an explainer.
label: name of the model. By default it's extracted from the 'class' attribute of the model.
path: if specified, then this path will be highlighed on the plot. Use average in order to show an average effect

References

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

Examples

Run this code

library("DALEX")
library("iBreakDown")
set.seed(1313)
model_titanic_glm <- glm(survived ~ gender + age + fare,
                       data = titanic_imputed, family = "binomial")
explain_titanic_glm <- explain(model_titanic_glm,
                               data = titanic_imputed,
                               y = titanic_imputed$survived,
                           label = "glm")

# there is no explanation level uncertanity linked with additive models
bd_glm <- break_down_uncertainty(explain_titanic_glm, titanic_imputed[1, ])
bd_glm
plot(bd_glm)

if (FALSE) {
## Not run:
library("randomForest")
set.seed(1313)
model <- randomForest(status ~ . , data = HR)
new_observation <- HR_test[1,]

explainer_rf <- explain(model,
                        data = HR[1:1000, 1:5])

bd_rf <- break_down_uncertainty(explainer_rf,
                           new_observation)
bd_rf
plot(bd_rf)

# example for regression - apartment prices
# here we do not have intreactions
model <- randomForest(m2.price ~ . , data = apartments)
explainer_rf <- explain(model,
                        data = apartments_test[1:1000, 2:6],
                        y = apartments_test$m2.price[1:1000])

bd_rf <- break_down_uncertainty(explainer_rf, apartments_test[1,])
bd_rf
plot(bd_rf)

bd_rf <- break_down_uncertainty(explainer_rf, apartments_test[1,], path = 1:5)
plot(bd_rf)

bd_rf <- break_down_uncertainty(explainer_rf,
                                     apartments_test[1,],
                                     path = c("floor", "no.rooms", "district",
                                         "construction.year", "surface"))
plot(bd_rf)

bd <- break_down(explainer_rf,
                    apartments_test[1,])
plot(bd)

s <- shap(explainer_rf,
                   apartments_test[1,])
plot(s)
}