[FieldTrip] Problem with FT's nanmean

[Schoffelen, J.M. (Jan Mathijs)]

Hi Jo,

The size of the variable that you are using as an input is quite large, probably it takes up 20GB of RAM or so.
Perhaps that the low-level c-code, on which the nanmean mex-files are based, has problems dealing with such large variables? Although I would have expected an error to be thrown or a segmentation fault in that case. Does mr. Google provide any directions here (e.g. when searching for the maximum allowed variable size in mex files)?
As a side note, I am surprised that the code defaults to using the mex-file, while you seem to have the legacy nanmean function available.

Best wishes,
Jan-Mathijs



On 6 May 2020, at 12:27, joachim.lange at med.uni-duesseldorf.de<mailto:joachim.lange at med.uni-duesseldorf.de> wrote:

Dear all,
I have a dataset with 94 trials, 44800 (virtual) channels, and 600 timepoints. I applied ft_timelockanalysis on this dataset and the output was surprisingly "0" for all channel-time pairs, even for those channels that contained only NaNs (because they were outside the brain). If I tried a smaller number of trials, everything was fine, though (i.e., no zeros, but plausible values).
I was able to boil down the problem to the function "nanmean". If I extract a 94x44800x600 matrix from the dataset and use FT's nanmean (apparently fieldtrip-20200423/external/stats/nanmean.mexa64), the result is zeros, again. If I use matlab's nanmean, everythig seems fine.
Any help is appreciated.
Best,
Jo
_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://doi.org/10.1371/journal.pcbi.1002202

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20200506/4789ebcb/attachment.htm>