Spatio-temporal clusters permutation test

✉️ Mailing List Archive (read-only)
1

Hello,

I'm trying to run the non-parametric permutation test using the spatio_temporal_cluster_1samp_test function. I'm first setting the connectivity with:
connectivity = spatio_temporal_tris_connectivity(grade_to_tris(5), n_times)
The call is to the stats function is:
T_obs, clusters, cluster_pv, H0 = mem.cache(spatio_temporal_cluster_1samp_test,
                                                ignore=['n_jobs'])(X1 - X2,
                                                threshold=t,
                                                n_permutations=n_permutations,
                                                tail=0,
                                                stat_fun=stat_fun,
                                                connectivity=connectivity,
                                                n_jobs=n_jobs, seed=0)
(Where stat_fun is the scipy t-test: scipy.ttest_1samp.)

When this line is encountered, it appears to stall forever. It's been stuck on this line for several days. When I run the analogous script for spatial clusters only (averaging over a time window), it works and executes in about 15 minutes. Any ideas?

Thank you,
Ben

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140915/65034941/attachment.html

2

Have you tried doing just two time points? This should only take about
twice as long as the spatial case. Then try three, four, etc. -- it could
be that you have far too many times points (e.g. if n_times=1000, then 15
minutes * 1000 time points is more than 10 days). If that's the case, you
could benefit from choosing a stricter initial t-value threshold, smaller
source space, or temporally downsampling.

Also, you should probably use the function `ttest_1samp_no_p` -- it is
quite a bit faster than the `scipy` equivalent because our version does not
calculate the corresponding p values (which is very slow).

Eric

3

Ah yes, it appears that was the problem.
Thanks for the help,
Ben