hmm has been removed from sklearn

[ThomasWalter]

When installing cell cognition from the sources, there is a problem with the most recent version of scikit-learn: they removed the hmm package.

For this reason, CellCognition cannot be started, and one needs to downgrade scikit-learn.

I have found the hmmlearn package that should in principle do the same thing as the former sklearn.hmm package. There is also the seqlearn package.

However, they are not exactly the same. For instance hmm.decoder_algorithms does not exist.

Should we go for this solution?

[sommerc]

is using virtualenv a solution for you?

[ThomasWalter]

Hello.

Thanks for your answer.

Yes, absolutely. Actually, I have downgraded sklearn, so it is now running for me (but virtualenv would have been the better solution, clearly).

So I am not blocked personally (and consequently, it is not urgent for me). But I guess that on the long run, we might want to do something about this, as potential users cannot run the current version of CellCognition from source code.

Thomas.

[ThomasWalter]

Actually, I can take care of this, but I was thinking that Rudi might prefer to deal with this, as he has written the corresponding function. Any comments?

[rhoef]

Virtualenv is a suitable solution, as sklearn.hmm in not the only depreciated dependency (libsvm, ontospy).

Actually it can be savely removed if we omit "Qings" TC3 code. There is still the "Smoothing Model" as HMM learning algorithm, which does not depend on sklearn.