Calculates importance of hierarchically grouped aspects

This function creates a tree that shows order of feature grouping and calculates importance of every newly created aspect.

hierarchical_importance(
  x,
  data,
  y = NULL,
  predict_function = predict,
  type = "predict",
  new_observation = NULL,
  N = 1000,
  loss_function = DALEX::loss_root_mean_square,
  B = 10,
  fi_type = c("raw", "ratio", "difference"),
  clust_method = "complete",
  cor_method = "spearman",
  ...
)

# S3 method for hierarchical_importance
plot(
  x,
  absolute_value = FALSE,
  show_labels = TRUE,
  add_last_group = TRUE,
  axis_lab_size = 10,
  text_size = 3,
  ...
)

Arguments

x	a model to be explained.
data	dataset NOTE: Target variable shouldn't be present in the data
y	true labels for data
predict_function	predict function
type	if predict then aspect_importance is used, if model than feature_importance is calculated
new_observation	selected observation with columns that corresponds to variables used in the model, should be without target variable
N	number of rows to be sampled from data NOTE: Small N may cause unstable results.
loss_function	a function that will be used to assess variable importance, if type = model
B	integer, number of permutation rounds to perform on each variable in feature importance calculation, if type = model
fi_type	character, type of transformation that should be applied for dropout loss, if type = model. "raw" results raw drop losses, "ratio" returns drop_loss/drop_loss_full_model.
clust_method	the agglomeration method to be used, see hclust methods
cor_method	the correlation method to be used see cor methods
...	other parameters
absolute_value	if TRUE, aspects importance values will be drawn as absolute values
show_labels	if TRUE, plot will have annotated axis Y
add_last_group	if TRUE, plot will draw connecting line between last two groups
axis_lab_size	size of labels on axis Y, if applicable
text_size	size of labels annotating values of aspects importance

Value

ggplot

Examples

library(DALEX)
apartments_num <- apartments[,unlist(lapply(apartments, is.numeric))]
apartments_num_lm_model <- lm(m2.price ~ ., data = apartments_num)
hi <- hierarchical_importance(x = apartments_num_lm_model,
 data = apartments_num[,-1],
 y = apartments_num[,1],
 type = "model")
plot(hi, add_last_group = TRUE, absolute_value = TRUE)