This repo contains investigation into measures of local importance for tree based models. In particular Shapley Values via the treeSHAP algorithm introduced in Consistent Individualized Feature Attribution for TreeEnsembles are compared to the 'prediction decomposition method' introduced in the Random forest interpretation with scikit-learn blog post.
richardangell/tree-local-importance
Repo containing investigation into measure of local importance for tree based models
Jupyter NotebookMIT