slundberg/shap
Shap
SHAP is a machine learning explainer that uses a game-theoretic framework to estimate the contribution of each feature to a model prediction. It provides a set of tools for quantifying how individual input features push a specific output away from a baseline value.
The project includes specialized explainers for different architectures, including high-speed implementations for decision trees and ensemble models, linearization algorithms for deep learning networks, and covariance integration for linear models. It also features a model-agnostic interpretability tool that uses a kernel method to provide explanations for any model architecture.
The system covers both local and global interpretability, allowing for the ranking of the most influential drivers across a dataset and the calculation of pairwise interaction effects between features. These contributions and their distributions are represented through feature impact visualizations and summary plots.
Features
- Local Interpretability Tools - SHAP calculates the contribution of each feature to a prediction by quantifying how much each input pushes the output from the base value.
- Model Interpretability - Provides a game-theoretic framework to quantify and explain the decision logic and outputs of complex machine learning models.
- Shapley Value Calculators - Provides a game-theoretic framework for calculating Shapley values to interpret machine learning model predictions.
- Feature Importance Attribution - Implements additive feature attribution to quantify the relative contribution of individual input variables to model predictions.
- Game-Theoretic Explainers - Implements a game theoretic framework for estimating the contribution of each feature to a model prediction.
- Model Agnostic Interpretability - Estimates the contribution of input features for any model architecture using a kernel method.
- Model Interpretability Tools - Ships a system for explaining any model architecture using a kernel method to quantify feature importance.
- SHAP Value Computations - Implements a model-agnostic kernel method to estimate Shapley values for any model architecture.
- Applied Machine Learning - Provides a comprehensive system for explaining the output of any machine learning model through feature attribution.
- Local Attribution Methods - Calculates the contribution of each feature to a specific prediction using game theoretic Shapley values.
- Tree-Based Explainers - Provides a high-speed implementation for calculating exact Shapley values for decision trees and ensemble models.
- Tree-Path Attribution - Computes exact feature contributions by traversing decision tree paths and weighting average changes in expected value.
- Deep Learning Attribution Tools - Provides a linearization algorithm that approximates feature importance for complex neural networks.
- Sample Reduction Techniques - Provides coalitional sample reduction to minimize model evaluations for large input spaces.
- Feature Interaction Analyzers - Implements methods to compute pairwise interaction effects and non-linear relationships between features.
- Interaction Effect Calculation - Computes pairwise interaction effects between features for tree-based models to reveal higher-order dependencies.
- Language Model Interpretability - Uses coalitional rules to explain the outputs of large natural language models with reduced function evaluations.
- Language Model Interpretability - Explains the outputs of large natural language models using coalitional rules to reduce function evaluations.
- Covariance-Aware Attribution - Calculates exact contributions for linear models by incorporating the feature covariance matrix.
- Linear Model Interpretability - Calculates exact Shapley values for linear models to account for correlated features and covariance in predictions.
- Component Linearization - Speeds up deep learning explanations by replacing complex non-linear layers with linear approximations.
- Tree Model Diagnostics - Uncovers complex risk factors and feature interactions in decision trees using exact Shapley value calculations.
- Interpretability - Calculates feature importance for neural networks using high speed approximation algorithms.
- Pairwise Interaction Analysis - Identifies higher-order dependencies by calculating the difference between the joint effect of two features and their individual effects.
- Tree Model Interpretability - Computes exact Shapley values for trees and ensembles to uncover complex risk factors and feature contributions.
- Feature Interaction Visualizations - Plots a single feature against its contribution to the model output to identify non-linear effects.
- Deep Learning Interpretability - Uses high-speed approximation algorithms to linearize complex components and calculate feature importance for deep learning models.
- Explainable AI Libraries - Game-theoretic approach to explain the output of machine learning models.
- Model Interpretability - Unified approach to interpreting model predictions.
- Model Interpretation - Unified approach for explaining predictions of any model.
- Model Visualization - Game-theoretic approach for explaining model output predictions.
- Explainability - Listed in the “Explainability” section of the The Incredible Pytorch awesome list.
- Guardrails and AI Safety - Listed in the “Guardrails and AI Safety” section of the The Incredible Pytorch awesome list.
