model_profile: Dataset Level Variable Profile as Partial Dependence or...

In DALEX: moDel Agnostic Language for Exploration and eXplanation

Dataset Level Variable Profile as Partial Dependence or Accumulated Local Dependence Explanations

Description

This function calculates explanations on a dataset level set that explore model response as a function of selected variables. The explanations can be calulated as Partial Dependence Profile or Accumulated Local Dependence Profile. Find information how to use this function here: https://ema.drwhy.ai/partialDependenceProfiles.html. The variable_profile function is a copy of model_profile.

Usage

model_profile(
  explainer,
  variables = NULL,
  N = 100,
  ...,
  groups = NULL,
  k = NULL,
  center = TRUE,
  type = "partial"
)

variable_profile(
  explainer,
  variables = NULL,
  N = 100,
  ...,
  groups = NULL,
  k = NULL,
  center = TRUE,
  type = "partial"
)

single_variable(explainer, variable, type = "pdp", ...)

Arguments

explainer	a model to be explained, preprocessed by the explain function
variables	character - names of variables to be explained
N	number of observations used for calculation of aggregated profiles. By default 100. Use NULL to use all observations.
...	other parameters that will be passed to ingredients::aggregate_profiles
groups	a variable name that will be used for grouping. By default NULL which means that no groups shall be calculated
k	number of clusters for the hclust function (for clustered profiles)
center	shall profiles be centered before clustering
type	the type of variable profile. Either partial, conditional or accumulated.
variable	deprecated, use variables instead

Details

Underneath this function calls the partial_dependence or accumulated_dependence functions from the ingredients package.

Value

An object of the class model_profile. It's a data frame with calculated average model responses.

References

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

Examples

titanic_glm_model <- glm(survived~., data = titanic_imputed, family = "binomial")
explainer_glm <- explain(titanic_glm_model, data = titanic_imputed)
model_profile_glm_fare <- model_profile(explainer_glm, "fare")
plot(model_profile_glm_fare)

 
library("ranger")
titanic_ranger_model <- ranger(survived~., data = titanic_imputed, num.trees = 50,
                               probability = TRUE)
explainer_ranger  <- explain(titanic_ranger_model, data = titanic_imputed)
model_profile_ranger <- model_profile(explainer_ranger)
plot(model_profile_ranger, geom = "profiles")

model_profile_ranger_1 <- model_profile(explainer_ranger, type = "partial",
                                        variables = c("age", "fare"))
plot(model_profile_ranger_1 , variables = c("age", "fare"), geom = "points")

model_profile_ranger_2  <- model_profile(explainer_ranger, type = "partial", k = 3)
plot(model_profile_ranger_2 , geom = "profiles")

model_profile_ranger_3  <- model_profile(explainer_ranger, type = "partial", groups = "gender")
plot(model_profile_ranger_3 , geom = "profiles")

model_profile_ranger_4  <- model_profile(explainer_ranger, type = "accumulated")
plot(model_profile_ranger_4 , geom = "profiles")

# Multiple profiles
model_profile_ranger_fare <- model_profile(explainer_ranger, "fare")
plot(model_profile_ranger_fare, model_profile_glm_fare)
 

DALEX documentation built on Jan. 16, 2023, 1:06 a.m.