Fitting a scikits.learn.hmm.GaussianHMM to variable length training sequences

Question

I'd like to fit a scikits.learn.hmm.GaussianHMM to training sequences of different length. The fit method, however, prevents using sequences of different length by doing

obs = np.asanyarray(obs)

which only works on a list of equally shaped arrays. Does anyone have a hint on how to proceed?

Marmot

Unfortunately, this is only one of the issues of the HMM in scikit-learn; it's also lacks numerical stability. There has talk on the mailing list of perhaps dropping it not too long ago. – Khoury

Answer 1

You may do re-sampling to "reshape" a given input to the desired length.

Ratiocinate

Not sure why this hasn't been upvoted. +1 for the easiest solution? – Cepeda