Back to the main page.
Bug 480 - label, labelcmb, channel, channelcmb confusion/problems?
| Status | CLOSED DUPLICATE |
| Reported | 2011-02-09 02:24:00 +0100 |
| Modified | 2011-03-31 08:38:27 +0200 |
| Product: | FieldTrip |
| Component: | specest |
| Version: | unspecified |
| Hardware: | Macintosh |
| Operating System: | Mac OS |
| Importance: | P1 normal |
| Assigned to: | Jan-Mathijs Schoffelen |
| URL: | |
| Tags: | |
| Depends on: | |
| Blocks: | |
| See also: |
Matt Mollison - 2011-02-09 02:24:03 +0100
I'm using fieldtrip-20110204 on a Mac Pro, OS X 10.6.6. This may be better suited for the email list, but it seems like it could be bugs and not just my coherence analysis ineptitude. I will happily re-post there if that's the appropriate forum. (note: I've been searching the archives for a week or so now about coherence/plv analyses and haven't figured out how people accomplish what I'm trying to do) I'm doing a within-subjects experiment where I want to compare the coherence (or phase-locking values) between conditions that every subject ran in. I have 32 subjects and potentially more than 3 conditions depending on how I break things down, but I'm keeping things simple for this demonstration, so I'm using 3 subjects and 2 conditions for the following steps, and only 4 out of the 129 EEG electrodes. 1. calculate power and csd I ran data_freq = ft_freqanalysis(cfg_freq,data) on each condition for each subject with these settings: % pre- and post-stimulus times, in seconds exper.prepost = [-0.8 3.0]; exper.sampleRate = 250; cfg_freq = []; cfg_freq.keeptrials = 'no'; % subjects are my UO, don't need the trials cfg_freq.output = 'powandcsd'; cfg_freq.channel = {'e3','e60','e11','e62'}; cfg_freq.channelcmb = {'e3','e60';'e3','e62'}; % multi-taper method cfg_freq.method = 'mtmconvol'; cfg_freq.taper = 'hanning'; freqstep = exper.sampleRate/(sum(abs(exper.prepost))*exper.sampleRate)*2; % to space freqs evenly cfg_freq.foi = 3:freqstep:9; cfg_freq.t_ftimwin = 5./cfg_freq.foi; cfg_freq.tapsmofrq = 0.4*cfg_freq.foi; cfg_freq.toi = -0.5:0.04:1.5; It comes out as expected, with label and labelcmb fields, dimord = 'chan_freq_time', and powspctrm (4x12x51) and crsspctrm (2x12x51) fields. (Note: I tried 'fourier' but when I ran ft_connectivityanalysis it created dimord = 'chan_chan_freq_time', which it didn't seem like any functions knew how to deal with.) 2. calculate coherence (or plv or whatever) I ran data_conn = ft_connectivityanalysis(cfg,data_freq) on each condition for each subject with method = 'coh', and for each I got a data struct with labelcmb (2x2) and cohspctrm (2x12x51) fields. This makes sense based on my input. 3. statistics In order to run ft_freqstatistics on these data, I need to get them in a grand average struct, with keepindividual = 'yes'; however, I can't put data_conn into ft_freqgrandaverage (or ft_freqdescriptives, but I don't think that's the right function to use) because data_conn doesn't have a label field, only labelcmb, and labelcmb is a 2x2 cell that contains the channel comparisons. Am I doing something wrong with the label and labelcmb fields, or is there actually a bug and the label field should actually exist? Or maybe I should be using fourierspctrm instead? I tried a lot of testing with fourierspctrm and that didn't seem to solve any problems. Sorry for the long report and thanks for any help. Matt
Jan-Mathijs Schoffelen - 2011-02-25 10:28:39 +0100
Hi Matt, Thanks for posting this and as usual thanks for your detailed report. We really appreciate it. @ your note: (Note: I tried 'fourier' but when I ran ft_connectivityanalysis it created dimord = 'chan_chan_freq_time', which it didn't seem like any functions knew how to deal with.) Yes, well I understand that indeed the XXXstatistics functions may have a problem dealing with this, but I am working on the plotting routines. @ your point 3: 3. statistics In order to run ft_freqstatistics on these data, I need to get them in a grand average struct, with keepindividual = 'yes'; however, I can't put data_conn into ft_freqgrandaverage (or ft_freqdescriptives, but I don't think that's the right function to use) because data_conn doesn't have a label field, only labelcmb, and labelcmb is a 2x2 cell that contains the channel comparisons. Am I doing something wrong with the label and labelcmb fields, or is there actually a bug and the label field should actually exist? Or maybe I should be using fourierspctrm instead? I tried a lot of testing with fourierspctrm and that didn't seem to solve any problems. I can understand the need for doing statistics the way you intend them to be done. As of yet, ft_grandaverage does not average across data-fields which pertain to bivariate metrics, and indeed probably does not work at all, if no label is present in the input data. However, I don't see a fundamental objection against it. As a side note, did you try to call ft_freqstatistics with multiple input arguments? I would predict that this works (or may crash at a different point in the code ;o) ). cfg.parameter = 'cohspctrm'; cfg.xxx = ... etc stat = ft_freqstatistics(cfg, subj1, subj2, subj3, ..., subjN) I think this should theoretically work
Matt Mollison - 2011-02-25 17:55:00 +0100
Hi JM, Thanks for your reply. Thanks to an email you sent a week or two ago on the mailing list I only recently realized that I could use ft_freqstatistics with multiple-subject input and that I don't have to give it output from ft_grandaverage (I've since revised my scripts to do this). Note that none of the tutorials are set up this way so I think this is why I never realized it—the tutorials only use output from ft_grandaverage as input to timelock/freq stats functions. (I'm sure it is already on your list to revise the tutorials.) So I can make sure I'm on the right track for coherence/plv analyses for a within-subjects exp and doing comparisons between conditions, should I be using 'powandcsd' or 'fourier' as output in ft_freqanalysis (and subsequently run ft_connectivityanalysis)? Or is either a valid option? I don't really know the difference, there's not much info on this in the tutorials, and I have been unable to find a clear answer about this on the mailing list archives. Matt (In reply to comment #1) > Hi Matt, > Thanks for posting this and as usual thanks for your detailed report. We really > appreciate it. > > > @ your note: (Note: I tried 'fourier' but when I ran ft_connectivityanalysis it > created > dimord = 'chan_chan_freq_time', which it didn't seem like any functions knew > how to deal with.) > > Yes, well I understand that indeed the XXXstatistics functions may have a > problem dealing with this, but I am working on the plotting routines. > > > @ your point 3: > 3. statistics > In order to run ft_freqstatistics on these data, I need to get them in a grand > average struct, with keepindividual = 'yes'; however, I can't put data_conn > into ft_freqgrandaverage (or ft_freqdescriptives, but I don't think that's the > right function to use) because data_conn doesn't have a label field, only > labelcmb, and labelcmb is a 2x2 cell that contains the channel comparisons. > > Am I doing something wrong with the label and labelcmb fields, or is there > actually a bug and the label field should actually exist? Or maybe I should be > using fourierspctrm instead? I tried a lot of testing with fourierspctrm and > that didn't seem to solve any problems. > > I can understand the need for doing statistics the way you intend them to be > done. As of yet, ft_grandaverage does not average across data-fields which > pertain to bivariate metrics, and indeed probably does not work at all, if no > label is present in the input data. However, I don't see a fundamental > objection against it. > As a side note, did you try to call ft_freqstatistics with multiple input > arguments? I would predict that this works (or may crash at a different point > in the code ;o) ). > > cfg.parameter = 'cohspctrm'; > cfg.xxx = ... etc > stat = ft_freqstatistics(cfg, subj1, subj2, subj3, ..., subjN) > > I think this should theoretically work