Fixes the Tree.value attribute for classifiers. Instead of taking the weighted average as we do with regressors we take the weighted class count for each node.
Along with removing the data param, this allows us to create shap explainers. Unfortunately the explainer still segfaults when creating shap_values. Still unsure why.
Fixes https://github.com/crflynn/skranger/issues/66#issuecomment-959028691
Fixes the Tree.value attribute for classifiers. Instead of taking the weighted average as we do with regressors we take the weighted class count for each node.
Along with removing the
data
param, this allows us to create shap explainers. Unfortunately the explainer still segfaults when creatingshap_values
. Still unsure why.