Jacob Schreiber

Yes, initially it was just using pip but couldn't find blas. I couldn't pip or conda install openblas because of permission errors on the mac build so I went down...

Without knowing the contents of "observations" or the keys in the model it's difficult for me to debug. You might try passing in `numpy.nan` instead of None if the data...

Isn't this only going to produce builds on aarch64?

I'm not sure that this workflow is working. There only seems to be one set of jobs spawned, not two (one for aarch64 and one for x86). Can you clarify?

Oh, sorry, this is editing the other workflow. Is it possible to modify the "build" workflow first so I can see it in action?

Sorry for not responding to this sooner. I don't really know what a "featurized" HMM is. If what you're saying is that you want to pass in multi-dimensional inputs, you...

Unfortunately sparse matrix support is not yet added to pomegranate. What you can do instead is load up small chunks of data that do fit in memory, call the `summarize`...

It would be good to see your training scheme as well. There might be an issue with semi-supervised training, but I wouldn't be able to pinpoint it until I have...

Can you try making all keys strings? I haven't extensively tested mixed type tables.

Sorry, there has not been an update on this yet. I believe that sampling is implemented for ConditionalProbabilityTables, so you could start by sampling from one table, then plug those...