ConvergenceWarning: Liblinear failed to converge, increase the number of iterations. "the number of iterations."

[ziodave]

When running tapioca train-classifier I get the following warning multiple times:

lib/python3.6/site-packages/sklearn/svm/_base.py:986: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.
  "the number of iterations.", ConvergenceWarning)

Is it okey? or should I change some parameters?

Cheers.

[wetneb]

I have also seen this message and remember having tried to increase the number of iterations without much success. The learning process still seems to work, but perhaps it would be worth investigating if we could make it better or faster by changing some parameters.

[ziodave]

Maybe we can ping someone from scikit-learn like @ogrisel (we worked together on Apache Stanbol) or @amueller, @thomasjpfan, @hermidalc...

I see that there's an open issue, scikit-learn/scikit-learn#11536

[amueller]

You can also try scaling your data. Currently the tolerance might be a bit too strict so the warning doesn't necessarily mean the results are not valid (but that the required tolerance was not met).