Coverage remained the same when pulling b98b71b on rmcgibbo:hmmfix into a2580d6 on scikit-learn:master.

+1 for merge

Looks good as well to me but I am not familiar with the HMM code. Maybe @kmike or @SnippyHolloW can have a look.

If I understand correctly, observations with a sequence length of 1 are just ignored when updating the stats to estimate the transition probabilities?

The bypass seems fine to me. I previously did the check on my side (before calling fit()).

Maybe add a comment on the line of the "if > 1" to say there are no transitions to update with with 1 observation sequences? +1 for merge otherwise.

+1 for adding an explicit inline comment.

I was using end-of-sentence markers and haven't had this problem because of that. The PR looks good.

comment applied.

Coverage remained the same when pulling fc3fcb8 on rmcgibbo:hmmfix into a2580d6 on scikit-learn:master.

Thanks! merging.