0

I have a Hidden Markov Model that has been pickled with joblib using the sklearn.hmm module. Apparently, in version 0.17.x this module has been deprecated and moved to hmmlearn. I am unable to load the model and I get the following error:

ImportError: No module named 'sklearn.hmm'

I have tried to revert back to version 0.16.x but still cannot load the model. I get the following error:

ImportError: libopenblas.so.0: cannot open shared object file: No such file or directory

I do not have access to the source code to recreate the model and re-pickle it

I am running Python 3.5

Has anyone else experienced this problem and have you found a solution? Does anyone know if scikit-learn has any way to guarantee persistence since the deprecation?

Sergei Lebedev
  • 2,659
  • 20
  • 23
user6858731
  • 1
  • 2
  • I don't think there are guarantees. Loading pickled objects will only work with the same version of the lib. Going back to this version is the right approach here, but the error above tells me, that your installation was not successful ```ImportError: libopenblas.so.0:```. So try to install the version you need and use the test to reason about the success. No need to check loading the pickled object with a broken install. – sascha Sep 21 '16 at 14:16

1 Answers1

0

After reverting to scikit-learn 0.16.x, I just needed to install OpenBlas for Ubuntu. It appears that the problem was more a feature of the operating system rather than Python.

user6858731
  • 1
  • 2