ceteris_paribus: Ceteris Paribus Profiles aka Individual Variable Profiles

In ingredients: Effects and Importances of Model Ingredients

Ceteris Paribus Profiles aka Individual Variable Profiles

Description

This explainer works for individual observations. For each observation it calculates Ceteris Paribus Profiles for selected variables. Such profiles can be used to hypothesize about model results if selected variable is changed. For this reason it is also called 'What-If Profiles'.

Usage

ceteris_paribus(x, ...)

## S3 method for class 'explainer'
ceteris_paribus(
  x,
  new_observation,
  y = NULL,
  variables = NULL,
  variable_splits = NULL,
  grid_points = 101,
  variable_splits_type = "quantiles",
  ...
)

## Default S3 method:
ceteris_paribus(
  x,
  data,
  predict_function = predict,
  new_observation,
  y = NULL,
  variables = NULL,
  variable_splits = NULL,
  grid_points = 101,
  variable_splits_type = "quantiles",
  variable_splits_with_obs = FALSE,
  label = class(x)[1],
  ...
)

Arguments

x	an explainer created with the DALEX::explain() function, or a model to be explained.
...	other parameters
new_observation	a new observation with columns that corresponds to variables used in the model
y	true labels for new_observation. If specified then will be added to ceteris paribus plots. NOTE: It is best when target variable is not present in the new_observation
variables	names of variables for which profiles shall be calculated. Will be passed to calculate_variable_split. If NULL then all variables from the validation data will be used.
variable_splits	named list of splits for variables, in most cases created with calculate_variable_split. If NULL then it will be calculated based on validation data available in the explainer.
grid_points	maximum number of points for profile calculations. Note that the finaln number of points may be lower than grid_points, eg. if there is not enough unique values for a given variable. Will be passed to calculate_variable_split.
variable_splits_type	how variable grids shall be calculated? Use "quantiles" (default) for percentiles or "uniform" to get uniform grid of points
data	validation dataset. It will be extracted from x if it's an explainer NOTE: It is best when target variable is not present in the data
predict_function	predict function. It will be extracted from x if it's an explainer
variable_splits_with_obs	if TRUE then all values in new_observation will be included in variable_splits
label	name of the model. By default it's extracted from the class attribute of the model

Details

Find more details in Ceteris Paribus Chapter.

Value

an object of the class ceteris_paribus_explainer.

References

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

Examples

library("DALEX")
library("ingredients")
titanic_small <- select_sample(titanic_imputed, n = 500, seed = 1313)

# build a model
model_titanic_glm <- glm(survived ~ gender + age + fare,
                         data = titanic_small,
                         family = "binomial")

explain_titanic_glm <- explain(model_titanic_glm,
                               data = titanic_small[,-8],
                               y = titanic_small[,8])

cp_rf <- ceteris_paribus(explain_titanic_glm, titanic_small[1,])
cp_rf

plot(cp_rf, variables = "age")


library("ranger")
model_titanic_rf <- ranger(survived ~., data = titanic_imputed, probability = TRUE)


explain_titanic_rf <- explain(model_titanic_rf,
                              data = titanic_imputed[,-8],
                              y = titanic_imputed[,8],
                              label = "ranger forest",
                              verbose = FALSE)

# select few passangers
selected_passangers <- select_sample(titanic_imputed, n = 20)
cp_rf <- ceteris_paribus(explain_titanic_rf, selected_passangers)
cp_rf

plot(cp_rf, variables = "age") +
  show_observations(cp_rf, variables = "age") +
  show_rugs(cp_rf, variables = "age", color = "red")

ingredients documentation built on Jan. 15, 2023, 5:09 p.m.