Jacob Schreiber

Is it possible that you installed a version of PyTorch made to work with GPUs but you don't have a GPU?

Sorry that you encountered issues with this, but since it looks like it's an issue with PyTorch I can't provide more help. Please re-open if you're able to track down...

There were a few issues involved here that have now been fixed in pomegranate v0.15.0. Try that out and let me know if it works. Some of these issues included...

`MultivariateGaussianDistribution` and `HiddenMarkovModel` are objects in pomegranate before the recent rewrite. Please see the README for reasoning. What you're looking for is `Normal` (https://github.com/jmschrei/pomegranate/blob/master/pomegranate/distributions/normal.py#L22) and either `SparseHMM` or `DenseHMM` (probably...

I don't think this is a problem. If you initialize a distribution, you're basically saying that you've allocated memory for all the parameters but do not have values for those...

Can you post your code? You should be able to use literally the same distribution objects in both the GMM and the HMM.

This should be fixed in v1.1.0. Let me know if you still encounter issues.

You can also try adding `covariance_type='diag'`. It should probably default to that if the means/covars are only 1D but it currently assumes a full covariance matrix.

Good idea. I can add it in the next batch of fixes.

I haven't added in callbacks yet. It would be a nice thing to have, though.