Compute SHAP Values with Correlation Warnings — compute_shap_with

Computes SHAP values for model predictions and flags features that have high correlations with other features, which can make SHAP interpretations unreliable.

Usage

compute_shap_with_warnings(
  model,
  data,
  n_samples = 100,
  correlation_threshold = 0.7,
  method = "permutation",
  verbose = TRUE
)

Arguments

model: A trained model (mlr3 Learner, caret model, or predict function)
data: Data.frame of features used for SHAP computation
n_samples: Number of samples for SHAP approximation (default: 100)
correlation_threshold: Correlation threshold for warnings (default: 0.7)
method: SHAP method: "permutation" or "kernel" (default: "permutation")
verbose: Print progress messages

Value

A list with: - shap_values: Matrix of SHAP values (samples x features) - feature_importance: Mean absolute SHAP per feature - correlations: Correlation matrix of features - warnings: Data.frame of correlated feature pairs with warnings - reliable_features: Features with no high correlations (safe to interpret)

Examples

if (FALSE) { # \dontrun{
# Compute SHAP with correlation warnings
result <- compute_shap_with_warnings(model, test_data)

# Check which features have reliable SHAP interpretations
print(result$reliable_features)

# View correlation warnings
print(result$warnings)
} # }