Open kordjamshidi opened 8 years ago
I see these lines here in the ILPInference are problematic for Saul joint training still, when we start with the classifiers that their lableLexicon is not complete e.g. because they have net seen any examples yet, so the ILP inference can not return any score but the default could be to return zero, not throwing error. Can anyone address this issue here? https://github.com/IllinoisCogComp/lbjava/blob/master/lbjava/src/main/java/edu/illinois/cs/cogcomp/lbjava/infer/ILPInference.java#L173
I see these lines here in the ILPInference are problematic for Saul joint training still, when we start with the classifiers that their lableLexicon is not complete e.g. because they have net seen any examples yet, so the ILP inference can not return any score but the default could be to return zero, not throwing error. Can anyone address this issue here? https://github.com/IllinoisCogComp/lbjava/blob/master/lbjava/src/main/java/edu/illinois/cs/cogcomp/lbjava/infer/ILPInference.java#L173