Add support for log_likelihood to semisupervised_classification.iterative_labeling.AutonBox primitive

[roquelopez]

iterative_labeling is a powerful and useful primitive. We used it on text semi-supervised problems. I think would be very nice to have log_likelihood implemented in iterative_labeling primitive. I think the implementation should be simple since it just has to use the log_likelihood method of the classifiers. Do you think it is doable? Thanks!

[saswatiray]

We are looking into this!

[saswatiray]

Fixed. This will need some change in log_likelihoods() functions of the estimators to be able to handle data with unlabeled classes.