Paper:可解释性之SHAP之《A Unified Approach to Interpreting Model Predictions—解释模型预测的统一方法》论文解读与翻译
导读:2017年11月25日,来自华盛顿大学的Scott M. Lundberg和Su-In Lee在《解释模型预测的统一方法》论文中,提出了SHAP值作为特征重要性的统一度量。SHAP可以为每个特征分配一个特定预测的重要性值。它的意义在于解释现代机器学习中大多数的黑盒模型,为效果好的ML模型量化各个特征的贡献度。
目录
《A Unified Approach to Interpreting Model Predictions》论文解读与翻译
2 Additive Feature Attribution Methods
2.3 Layer-Wise Relevance Propagation
2.4 Classic Shapley Value Estimation
3 Simple Properties Uniquely Determine Additive Feature Attributions
4 SHAP (SHapley Additive exPlanation) Values
4.1 Model-Agnostic Approximations
Kernel SHAP (Linear LIME + Shapley values)
4.2 Model-Specific Approximations
5 Computational and User Study Experiments
5.2 Consistency with Human Intuition
5.3 Explaining Class Differences
相关文章
Paper:《A Unified Approach to Interpreting Model Predictions—解释模型预测的统一方法》论文解读与翻译