Generate SHAP values for XGBoost model interpretation

Usage

tl_xgboost_shap(model, data = NULL, n_samples = 100, trees_idx = NULL)

Arguments

model: A tidylearn XGBoost model object
data: Data for SHAP value calculation (default: NULL, uses training data). The response column is not needed.
n_samples: Number of samples to use (default: 100, NULL for all)
trees_idx: Boosting rounds to include, as a run of consecutive rounds counted from 1 such as 1:20 (default: NULL, uses every round)

Value

A data frame with one column of SHAP values per feature (the columns of the model's design matrix), a BIAS column and a row_id column giving the row of the (sampled) data each row explains. For a multiclass model there is one block of rows per class, told apart by a class column. The columns of data whose names are not already taken – the response, and any factor predictor, whose SHAP columns are named after its levels – are appended for reference.

Examples

# \donttest{
if (requireNamespace("xgboost", quietly = TRUE)) {
  model <- tl_model(mtcars, mpg ~ ., method = "xgboost", nthread = 2)
  shap <- tl_xgboost_shap(model, n_samples = 20)
}
# }