Snowflake Model Explainability — Shapley-Value Feature Attribution (2026)

Official Snowflake documentation for Model Explainability in the Model Registry.

Key facts surfaced (2026-05-20):

• Implements explainability using SHAP library to compute Shapley values measuring average marginal contribution of each feature to model predictions.
• Supported model types: XGBoost, CatBoost, LightGBM, Scikit-learn, and Snowpark ML modeling classes.
• Enabled by default for supported models logged with Snowpark ML 1.6.2+; background data (up to 1,000 rows) supplied via sample_input_data parameter.
• Usage: explanations = mv.run(input_data, function_name="explain").
• Currently in preview release status.
• Curriculum reference: replaces the non-existent ml-functions/attribution URL for Module 5 Shapley-value MTA model (TC-87).