OneVsOneClassifier Support

[nareshshah139]

Looks like ELI5 does not work with Multi-class classifiers in sklearn though it works with the same in xgboost. Perhaps this can be a useful new feature.

Currently get the following error:

Error: estimator OneVsOneClassifier(estimator=RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini', max_depth=10, max_features='auto', max_leaf_nodes=None, min_impurity_split=1e-07, min_samples_leaf=20, min_samples_split=2, min_weight_fraction_leaf=0.0, n_estimators=1000, n_jobs=-1, oob_score=False, random_state=None, verbose=1, warm_start=False), n_jobs=1) is not supported

[kmike]

Yeah, currently we support sklearn multi-class classifiers which use OvR strategy, internally or with OneVsRestClassifier wrapper; OvO is not yet supported.

For example, if you use RandomForestClassifeir without OvO wrapper, explanations would work, as RandomForestClassifeir supports multi-class natively.

[nareshshah139]

The reason I ask this is the performance differs between multi-class RandomForest and OvO RandomForest/OvR Random Forest as well. It is puzzling though as to why there is a significant difference in the first place. That is something I was trying to debug.