Calculate Permutation Feature Importance.

When n_repeats == 1, the result is similar to the one from h2o.varimp(), i.e., it contains the following columns "Relative Importance", "Scaled Importance", and "Percentage".

h2o.permutation_importance(
  object,
  newdata,
  metric = c("AUTO", "AUC", "MAE", "MSE", "RMSE", "logloss", "mean_per_class_error",
    "PR_AUC"),
  n_samples = 10000,
  n_repeats = 1,
  features = NULL,
  seed = -1
)

Arguments

object	A trained supervised H2O model.
newdata	Training frame of the model which is going to be permuted
metric	Metric to be used. One of "AUTO", "AUC", "MAE", "MSE", "RMSE", "logloss", "mean_per_class_error", "PR_AUC". Defaults to "AUTO".
n_samples	Number of samples to be evaluated. Use -1 to use the whole dataset. Defaults to 10 000.
n_repeats	Number of repeated evaluations. Defaults to 1.
features	Character vector of features to include in the permutation importance. Use NULL to include all.
seed	Seed for the random generator. Use -1 to pick a random seed. Defaults to -1.

Value

H2OTable with variable importance.

Details

When n_repeats > 1, the individual columns correspond to the permutation variable importance values from individual runs which corresponds to the "Relative Importance" and also to the distance between the original prediction error and prediction error using a frame with a given feature permuted.

Examples

# NOT RUN {
library(h2o)
h2o.init()
prostate_path <- system.file("extdata", "prostate.csv", package = "h2o")
prostate <- h2o.importFile(prostate_path)
prostate[, 2] <- as.factor(prostate[, 2])
model <- h2o.gbm(x = 3:9, y = 2, training_frame = prostate, distribution = "bernoulli")
h2o.permutation_importance(model, prostate)
# }