1
$\begingroup$

I am trying to use LimeTabularExplainer class and explain_instance function to find explainations of my LightGbm (lgb) model. However, the lgb model uses complex feature set which are not interpretable.

I want to pass a subset of oringal features (which are interpretable) to the Lime explainer, so that my resultant explainations are also interpretable.

In sections 3.1 and 3.3 of original paper, the authors talk about this https://arxiv.org/abs/1602.04938

rf = sklearn.ensemble.RandomForestClassifier(n_estimators=500)
rf.fit(train, labels_train)

explainer = lime.lime_tabular.LimeTabularExplainer(train, 
                                                   feature_names=feature_names, 
                                                   class_names=target_names, 
                                                   discretize_continuous=True)
exp = explainer.explain_instance(test[i], rf.predict_proba, num_features=2, top_labels=1)

Suppose I need to pass test[i][:2], only the first two features to the surrogate model. Is there a way for this in LIME or even in SHAP?

$\endgroup$
1
  • $\begingroup$ I am not sure to understand what you want to do exactly. You have a model that is built on complex features (complex like what ? transformation ? aggregation of multiples features). And you want explanability in term of simpler features ? Why not building the moel on the simpler features ? $\endgroup$ Commented Oct 18, 2020 at 14:30

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.