Shapviz: SHAP Visualization Objects for Model Explanations

Michael Mayer created a visualization library for SHAP (SHapley Additive exPlanations) values. It provides plots like waterfall, force, importance, dependence, and interaction plots, acting on a 'shapviz' object created from a matrix of SHAP values and a corresponding feature dataset. The package includes wrappers for SHAP value computation from R packages like 'xgboost', 'lightgbm', 'fastshap', 'shapr', 'h2o', 'treeshap', 'DALEX', and 'kernelshap'.

Use Cases

• Generate waterfall plots to explain individual model predictions based on SHAP values.
• Create force plots to visualize the contribution of each feature to a specific prediction.
• Produce feature importance plots to rank global model features based on SHAP values.
• Build dependence plots to examine the relationship between a feature's value and its SHAP effect.
• Visualize interaction effects between features using interaction plots.

Strengths

• Separates visualization from computation, allowing display of factor variables even if the underlying model requires numeric inputs.
• Provides wrappers for SHAP value computation from at least eight popular R machine learning packages.
• Visualizations are inspired by the widely-used Python 'shap' package but have no dependency on it.

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Michael Mayer

Collection Method

Software library for creating visualizations from SHAP value matrices.