Get SHAP scores from a trained XGBoost or LightGBM model

shap.values returns a list of three objects from XGBoost or LightGBM model: 1. a dataset (data.table) of SHAP scores. It has the same dimension as the X_train); 2. the ranked variable vector by each variable's mean absolute SHAP value, it ranks the predictors by their importance in the model; and 3. The baseline value (intercept), which is stored in the last column of the SHAP contribution matrix (named "BIAS" in older xgboost versions or "(Intercept)" in newer versions). The rowsum of SHAP values including the baseline would equal to the predicted value (y_hat) generally speaking.

Usage

shap.values(xgb_model, X_train)

Arguments

xgb_model

an XGBoost or LightGBM model object

X_train

the data supplied to the predict function to get the prediction. It should be a matrix. Notice that coercing the matrix to a dense matrix by using as.matrix might lead to wrong behaviors in some cases. See discussion in issues on this topic.

Value

a list of three elements:

shap_score

A data.table of SHAP values (without the baseline column)

mean_shap_score

Ranked features by mean absolute SHAP value

BIAS0

The baseline/intercept value (from the '(Intercept)' column in xgboost 3.x)

Examples

# Example: Basic workflow for SHAP summary plot
# Note: For xgboost 3.x, use xgb.DMatrix + xgb.train, and convert factor labels to numeric

data("iris")
X1 = as.matrix(iris[,1:4])
y1 = as.numeric(iris[[5]]) - 1  # Convert factor to numeric
dtrain = xgboost::xgb.DMatrix(data = X1, label = y1)
params = list(learning_rate = 1, min_split_loss = 0, reg_lambda = 0,
              objective = 'reg:squarederror', nthread = 1)
mod1 = xgboost::xgb.train(params = params, data = dtrain,
                          nrounds = 1, verbose = 0)

# Get SHAP values and feature importance
shap_values <- shap.values(xgb_model = mod1, X_train = X1)
shap_values$mean_shap_score  # Ranked features by mean|SHAP|
#> Petal.Length  Petal.Width Sepal.Length  Sepal.Width 
#>    0.6307042    0.2135736    0.0300757    0.0000000 
shap_values_iris <- shap_values$shap_score

# Prepare long-format data for plotting
shap_long_iris <- shap.prep(xgb_model = mod1, X_train = X1)
# Alternative: use pre-computed SHAP values
shap_long_iris <- shap.prep(shap_contrib = shap_values_iris, X_train = X1)

# SHAP summary plot
shap.plot.summary(shap_long_iris, scientific = TRUE)

shap.plot.summary(shap_long_iris, x_bound  = 1.5, dilute = 10)


# Alternative options:
# Option 1: directly from xgboost model
shap.plot.summary.wrap1(mod1, X = as.matrix(iris[,1:4]), top_n = 3)


# Option 2: from pre-computed SHAP values (useful for cross-validation)
shap.plot.summary.wrap2(shap_score = shap_values_iris, X = X1, top_n = 3)