[Mne_analysis] ConvergenceWarning: FastICA did not converge.

[Hermine Berberyan]

        External Email - Use Caution        

Dear MNE community,
 
I have been applying independent component analysis (ICA) to the raw MEG data of mine. For this I used the following code:
 
ica = ICA(n_components=n_channels)
ica.fit(subject_data)
subject_data.save(ica_file)
 
While for most participants, it worked fine; for some it gave the following warning:
/opt/anaconda3/lib/python3.8/site-packages/sklearn/decomposition/_fastica.py:118: 
ConvergenceWarning: FastICA did not converge. Consider increasing tolerance or the maximum number
 of iterations. warnings.warn('FastICA did not converge. Consider increasing '
 
First question I have whether that means that my ICA analysis is not valid. I, unfortunately, can not find an answer to this question (sorry, in case it has been previously addressed).
 
And the second connected question is how to solve this warning. Based on whatI have found so far, i have tried 2 things: 1) increase the number of iterations. I tried all the different numbers till 3500 for now. ICA takes much longer then but the warning still appears. 2) Because of memory considerations, I also tried not preloading dataset and also deleting ica after every participant but this also did not change warning.
   
I'll be thankful for any feedback and help.
 
Kind regards

Hermine Berberyan
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20210506/d6a48701/attachment-0001.html