kernelshap: Kernel SHAP Implementation for Model Interpretability

Michael Mayer's R package provides an efficient implementation of Kernel SHAP, permutation SHAP, and additive SHAP algorithms for explaining model predictions. The package supports multi-output models, case weights, parallel computations, and integrates with meta-learning frameworks like 'tidymodels', 'caret', and 'mlr3'. Visualizations can be created using the companion 'shapviz' package.

Use Cases

• Explain predictions from black-box models based on the implemented Kernel SHAP algorithm.
• Perform scalable feature importance analysis based on support for parallel computations and iterative algorithms for high-dimensional data.
• Integrate model explanations into machine learning workflows based on compatibility with meta-learning packages like 'tidymodels'.
• Visualize SHAP explanations for model debugging based on integration with the 'shapviz' package.

Strengths

• Implements multiple established SHAP algorithms (Kernel, permutation, additive) from the 2017 Lundberg and Lee paper.
• Designed for efficiency with iterative algorithms for high-dimensional data and support for parallel computations.
• Offers integration with popular R machine learning meta-packages like 'tidymodels', 'caret', and 'mlr3'.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Michael Mayer

Collection Method

Software package implementing published SHAP algorithms.