Model Interpretability with SHAP and LIME
Chapter 1: Foundations of Model Interpretability
Why Explain Model Predictions?
Interpretability vs. Explainability
Taxonomy of Interpretability Methods
Scope of Explanations: Global vs. Local
Challenges in Model Interpretation
Quiz for Chapter 1
Chapter 2: Local Interpretable Model-agnostic Explanations (LIME)
Intuition Behind LIME
How LIME Works: Perturbation and Surrogate Models
Applying LIME to Tabular Data
Applying LIME to Text Data
Interpreting LIME Explanations
LIME Implementation with Python
Limitations and Considerations for LIME
Hands-on Practical: Generating LIME Explanations
Quiz for Chapter 2
Chapter 3: SHapley Additive exPlanations (SHAP)
Introduction to Shapley Values
SHAP Values: Connecting Shapley to Model Features
Properties of SHAP Values
KernelSHAP: A Model-Agnostic Approach
TreeSHAP: Optimized for Tree-Based Models
Interpreting SHAP Plots: Force Plots
Interpreting SHAP Plots: Summary and Dependence Plots
SHAP Implementation with Python
Hands-on Practical: Calculating SHAP Values
Quiz for Chapter 3
Chapter 4: Comparing and Applying Interpretability Techniques
LIME vs. SHAP: Differences
LIME vs. SHAP: Strengths and Weaknesses
Choosing Between LIME and SHAP
Interpreting Explanations for Regression Models
Interpreting Explanations for Classification Models
Integrating Interpretability into the ML Workflow
Tradeoff and Limitations
Practice: Comparing LIME and SHAP Outputs
Quiz for Chapter 4
Tradeoff and Limitations
Was this section helpful?
- “Why Should I Trust You?”: Explaining the Predictions of Any Classifier, Marco Tulio Ribeiro, Sameer Singh, Carlos Guestrin, 2016 Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ACM) DOI: 10.1145/2939672.2939778 - Introduces the LIME algorithm for local, interpretable explanations of black-box model predictions.
- A Unified Approach to Interpreting Model Predictions, Scott M. Lundberg, Su-In Lee, 2017 Advances in Neural Information Processing Systems 30 (NIPS 2017), Vol. 30 DOI: 10.48550/arXiv.1705.07874 - Presents SHAP (SHapley Additive exPlanations), a unified framework for interpreting model predictions based on Shapley values.
- On the (In)fidelity and Sensitivity of Explanations, Chih-Kuan Yeh, Cheng-Yu Hsieh, Arun Sai Suggala, David I. Inouye, Pradeep Ravikumar, 2019 Advances in Neural Information Processing Systems 32 (NeurIPS 2019) DOI: 10.48550/arXiv.1901.09392 - Investigates the reliability of local explanation methods, highlighting issues of fidelity (faithfulness) and sensitivity (stability).
- Explaining Individual Predictions When Features Are Dependent: A Step-by-Step Guide to the Conditional SHAP Approach, Kjersti Aas, Martin Jullum, Anders Løland, 2021 Machine Learning: ECML PKDD 2021 Workshops, Vol. 12975 (Springer, Cham) DOI: 10.1007/978-3-030-80279-0_8 - Describes methods for handling feature dependencies in SHAP explanations, which is a common challenge.