search

Centre-IRM-INT / GT-MVPA-nilearn

GT MVPA nilearn from Marseille
3 stars 3 forks source link

Help with Warning Liblinear failed to converge? #21

Open JeanneCaronGuyon opened 2 years ago

JeanneCaronGuyon commented 2 years ago

Hi everyone,

Has anyone ever encountered this warning _"warnings.warn("Liblinear failed to converge, increase " /Users/caron-guyon/opt/anaconda3/lib/python3.9/site-packages/sklearn/svm/base.py:985: ConvergenceWarning: Liblinear failed to converge, increase the number of iterations." with a searchlight procedure on nilearn ?

I had never seen it before, I ran a script I used to use (that worked fine) and I can't find where the problem comes from... If someone has an idea to help sort this out I'm interested :)

Thanks ! Jeanne

SylvainTakerkart commented 2 years ago

Hi Jeanne,

This is not a bug, just a diagnostic message telling you that the combination of 'data' + 'algorithm' + 'default paramaters' of your algorithm' has not allowed the algorithm to converge on this data... Do you get this message tons of times (which would mean that this happens in tons of locations) or just once (which would mean this happens only in one sphere/location)? If just once, this means that your accuracy will be badly estimated in only one voxel, which you can probably ignore...

Sylvain

JeanneCaronGuyon commented 2 years ago

Hi Sylvain,

Oh I see, that makes sense. I'll check how many times it happens then !!

Thank you so much Jeanne