Question 11

Explanation

Correct Answer:
mlflow.shap

Explanation:

1. Why This Is Correct?

   • mlflow.shap is MLflow’s built-in integration with SHAP (SHapley Additive exPlanations), a leading library for model interpretability. It:

     • Generates feature importance scores (global interpretability).

     • Produces local explanations (per-prediction reasoning).

     • Logs visualizations (e.g., force plots, summary plots) as MLflow artifacts.

   • Example:

     
     

     import mlflow.shap
     import shap
      
     # Train model
     model = train_model(X_train, y_train)
      
     # Log SHAP summary
     explainer = shap.Explainer(model)
     shap_values = explainer(X_test)
     mlflow.shap.log_explanation(explainer, X_test)

2. Key Features:

   • Unified Tracking: SHAP outputs appear in the MLflow UI under the run’s artifacts.

   • Model-Agnostic: Works with sklearn, PyTorch, TensorFlow, etc.

Why Other Options Are Incorrect?

• mlflow.sklearn/mlflow.pytorch/mlflow.tensorflow:

  • These log models and metrics, but not explanations.

Key Takeaway:

For model interpretability in MLflow:

1. Use mlflow.shap for SHAP-based explanations.

2. For non-SHAP methods (e.g., LIME), log custom plots via mlflow.log_artifact().

Pro Tip: Compare explanations across runs using MLflow’s artifact diff view.