[Mne_analysis] Optimal algorithms for CSP
JR KING
jeanremi.king at gmail.com
Tue Jan 8 08:13:29 EST 2019
External Email - Use Caution
Hi
On Tue, 8 Jan 2019 at 13:00, A S <eng.emetsasa at gmail.com> wrote:
> External Email - Use Caution
>
> Hi all,
> I have eeg dataset for 10 subjects. The subjects do the same tasks. I
> applied the common spatial patterns from mne-python package. I used
> with CSP the following algorithms (logistic regression, svc with
> linear kernel,).
> I wanted to compare the algorithms. I found that, for some subjects
> the Logistic Regression performed better, for other subjects the SVC
> with linear Kernel performed better.
> I have following questions:
> for the same tasks is it normal that the optimal algorithm differs
> from subject to another?
>
linear svm with hinge square loss and logistic regression are likely to
provide very similar solutions. So it's likely that they'll randomly be
better for some subjects over other. Some work by Cichy et al suggest that
SVM may be slightly ore robust, but they havent tried it with CSP
Is it normal that the CSP patterns are totally different from subject
>
to another? Why? Is there a literature for this?
>
Subjects topographies, whether taken from CSP or other types of analyses
can vastly differ between subjects because of anatomical differences.
Kindest regards,
JR
>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190108/12cd2563/attachment.html
More information about the Mne_analysis
mailing list