surv_feature_importance function

Helper functions for model_parts.R

This function is used to calculate permutational feature importance using a time-dependent metric. The result is the change in loss function at each time point after permuting each selected variable.

surv_feature_importance(x, ...)

## S3 method for class 'surv_explainer'
surv_feature_importance(
  x,
  loss_function = NULL,
  ...,
  type = c("raw", "ratio", "difference"),
  B = 10,
  variables = NULL,
  variable_groups = NULL,
  N = NULL,
  label = NULL
)

Arguments

• x: an explainer object - model preprocessed by the explain() function
• ...: other parameters, currently ignored
• loss_function: a function that will be used to assess variable importance, by default loss_brier_score for survival models. The function can be supplied manually but has to have these named parameters (y_true, risk, surv, times), where y_true represents the survival::Surv object with observed times and statuses, risk is the risk score calculated by the model, and surv is the survival function for each observation evaluated at times.
• type: a character vector, if "raw" the results are losses after the permutation, if "ratio" the results are in the form loss/loss_full_model and if "difference" the results are of the form loss - loss_full_model`
• B: numeric, number of permutations to be calculated
• variables: a character vector, names of variables to be included in the calculation
• variable_groups: a list of character vectors of names of explanatory variables. For each vector, a single variable importance measure is computed for the joint effect of the variables which names are provided in the vector. By default, variable_groups = NULL, in which case variable importance measures are computed separately for all variables indicated in the variables argument.
• N: numeric, number of observations that are to be sampled from the dataset for the purpose of calculation
• label: label of the model, if provided overrides explainer$label

Returns

A data.frame containing results of the calculation.

Details

Note: This function can be run within progressr::with_progress() to display a progress bar, as the execution can take long, especially on large datasets.