Back to the main page.

Bug 1478 - fast vs very-slow processing in ft_freqanalysis on similar data.

Reported 2012-05-16 10:29:00 +0200
Modified 2012-05-16 10:56:45 +0200
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Windows
Importance: P3 normal
Assigned to:
Depends on:
See also:

Johanna - 2012-05-16 10:29:11 +0200

On EEG datasets from Madelon, she noticed some subjects run very quickly through ft_freqanalysis (20s or so) and others very slowly (e.g. 14000s or so). The datasets seems fairly equivalent (around 350 trials, of similar length trials) etc. I obtained 1 slow and 1 fast example from her to further test, and can indeed replicate it (i.e. not dependent on someone else running on the same mentat at the same time). The call to the raw data is like this: cfg=[]; cfg.output = 'pow'; cfg.method = 'mtmconvol'; cfg.taper = 'hanning'; cfg.keeptrials = 'yes'; cfg.foi = 2:2:30; cfg.t_ftimwin = .5*ones(1,length(cfg.foi)); % 5 cycles per time window cfg.toi = -0.5:0.05:1.5; freq = ft_freqanalysis(cfg,rawdata); Weirdly, if I select only the first 50 trials for the 'slow' and 'fast' raw datasets, then they run in similar speeds (a few seconds). But only when using all trials does the 'slow' one really slow down. I will create a test script for this, once this bug has an assigned number, and put the data for now on /home/common/matlab/fieldtrip/data/test but it is largish (300MB total), so is there a better place to put it?

Roemer van der Meij - 2012-05-16 10:49:58 +0200

Hey Johanna, I looked at the raw data, and it turns out the 'slow dataset' has a trial of about 377 seconds long (about 188000 samples at 500hz), being trial 163. The 'fast dataset' only has trials that go up to about 1600 samples.

Johanna - 2012-05-16 10:56:45 +0200

Thanks Roemer! JM just told me to look at the same thing (cfg.pad).