Back to the main page.
Bug 535 - can't use ft_singleplotTFR on wavelet/mtmconvol powandcsd data
| Status | CLOSED FIXED |
| Reported | 2011-03-17 00:27:00 +0100 |
| Modified | 2011-03-31 08:38:54 +0200 |
| Product: | FieldTrip |
| Component: | plotting |
| 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 - 2011-03-17 00:27:50 +0100
Created attachment 32 connectivity data and cfg for the problem described I'm using fieldtrip-20110315 on a Mac Pro running MATLAB 7.11 and OS X 10.6.6. I attempted to use ft_singleplotTFR on connectivity data that was calculated first with ft_freqanalysis (cfg.method='wavelet', cfg.output='powandcsd') and second with ft_connectivityanalysis using cfg.method='plv'. my input to ft_singleplotTFR looks like this (labelcmb and plvspctrm have dims with lengths 8256 because there are 129 channels and powandcsd/plv was calculated between all pairs, i.e., nchoosek(129,2)=8256): data_conn = labelcmb: {8256x2 cell} dimord: 'chan_freq_time' plvspctrm: [8256x9x33 double] freq: [2.9630 3.7037 4.4444 5.1852 5.9259 6.6667 7.4074 8.1481 8.8889] time: [1x33 double] cfg: [1x1 struct] cfg_ft = xparam: 'time' yparam: 'freq' zparam: 'plvspctrm' cohrefchannel: 'E3' zlim: [0 1] channel: 'E60' I get this error: selected 1408 channels for plvspctrm ??? Error using ==> ft_channelselection at 72 data with non-unique channel names is not supported Error in ==> ft_singleplotTFR at 290 selchannel = ft_channelselection(cfg.channel, data.label); I've attached data_conn and cfg_ft in a mat file. ==================================================== I should mentioned that I get a similar ft_channelselection error in ft_topoplotTFR (only setting a cohrefchannel, not cfg_ft.channel as in the cfg above; and I do set cfg_ft.elecfile='GSN-HydroCel-129.sfp'; cfg_ft.layout = ft_prepare_layout(cfg_ft);): creating layout from data.elec creating layout for egi128 system selected 1408 channels for plvspctrm Warning: Duplicate x-y data points detected: using average of the z values. > In griddata at 108 In ft_plot_topo at 146 In ft_topoplotER at 732 In ft_topoplotTFR at 116 ??? Error using ==> ft_channelselection at 72 data with non-unique channel names is not supported Error in ==> ft_topoplotER at 778 [dum labelindex] = match_str(ft_channelselection(setdiff(1:length(data.label),highlightchansel), data.label),lay.label); Error in ==> ft_topoplotTFR at 116 cfg=ft_topoplotER(cfg, varargin{:});
Matt Mollison - 2011-03-17 02:29:11 +0100
I specifically wrote about and attached wavelet data, but this also applies to ft_freqanalysis data with cfg.method='mtmconvol'. After running mtmconvol data through ft_connectivityanalysis, data_conn has the exact same structure as I posted about for the wavelet data. Also, for both wavelet and mtmconvol, I set cfg.keeptrials='no' and cfg.keeptapers='no' because subjects are my units of observation.
Jan-Mathijs Schoffelen - 2011-03-17 13:26:35 +0100
(In reply to comment #1) > I specifically wrote about and attached wavelet data, but this also applies to > ft_freqanalysis data with cfg.method='mtmconvol'. After running mtmconvol data > through ft_connectivityanalysis, data_conn has the exact same structure as I > posted about for the wavelet data. > > Also, for both wavelet and mtmconvol, I set cfg.keeptrials='no' and > cfg.keeptapers='no' because subjects are my units of observation. To this comment (#1): you need to specify cfg.keeptrials = 'yes', because the connectivity metric you want can only be computed when the single replicates per subject are still there. Once the plv is computed the trials become irrelevant
Jan-Mathijs Schoffelen - 2011-03-17 13:37:27 +0100
found it! the strmatch-operations (e.g. line 196 ft_singleplotTFR) need an 'exact' as third input. Otherwise also partial matches are selected, which caused the problem. I changed this and also in the relevant other plotting functions. Thanks for posting the data, very helpful, as usual. Could you check whether it works?
Matt Mollison - 2011-03-17 15:39:14 +0100
I will definitely check on it when today's version is on the FTP server (I thought I could immediately get the changes with a public SVN checkout <http://code.google.com/p/fieldtrip/source/checkout>, but I don't see the addition of the 'exact' argument yet.) Also, Matlab 7.11 tells me that strmatch will be removed in future versions. Might be a good idea to use strcmp instead. I'm confused about your comment that I need to specify cfg.keeptrials='yes' with powandcsd data. Dr. Maris wrote in this thread <http://mailman.science.ru.nl/pipermail/fieldtrip/2010-September/003086.html> that one could use cfg.keeptrials='no' for a within-subjects analysis (and using less RAM/disk would be ideal). This seems to make sense because the crsspctrm is already computed in ft_freqanalysis for all combinations of channels (8256 in my case). Am I missing something here? Related to this, it seems that with 'fourier' memory requirements will only be slightly increased in ft_freqanalysis because keeptrials='yes' (one spectrum per electrode per trial), but even more memory will be required when using ft_connectivityanalysis because it computes the cross-spectra between sensors for every trial and every frequency (there's no keeptrials='no' option here). So, if I want to keep memory usage down, I was thinking I should use powandcsd, while fourier will always require lots of RAM. Does that reasoning make sense?
Matt Mollison - 2011-03-17 16:08:03 +0100
Quick followup. I just tried keeptrials='yes' and keeptapers='yes' an I get an error in ft_frequanalysis (line 319) that I can only set both to 'yes' if I'm doing fourier output. I then tried keeptrials='yes' and keeptapers='no' and it has brought my computer (one-year-old Mac Pro with 14GB of RAM) to its knees. If keeptrials='yes' is required to do connectivity analyses, is that to be expected? I can't imagine lots of people have access to computers with that much RAM. I was planning on submitting this to a computer cluster that has 2GB of ram per node (running 1 subject per node), but that does not seem like a viable option when 14GB was maxed out for a few minutes after processing the first trial.
Matt Mollison - 2011-03-18 00:07:15 +0100
I've verified that the bug is fixed for ft_singleplotTFR and ft_topoplotTFR with the addition of the 'exact' match in strfind. I didn't check the ft_*plotER functions. However, I don't think I ever tried ft_multiplotTFR before, but trying it now get a weird plot. I only get 2 channels (labeled E1 and E1), not plv/TF plots of all channels using the cohrefchan as the reference. I used the same data_conn that I attached and this cfg_ft: cfg_ft = []; cfg_ft.xparam = 'time'; cfg_ft.yparam = 'freq'; cfg_ft.zparam = 'plvspctrm'; cfg_ft.cohrefchannel = {'E3'}; cfg_ft.channel = 'all'; cfg_ft.xlim = [0.5 0.8]; % time cfg_ft.ylim = [3 8]; % freq cfg_ft.zlim = [0 1]; % spctrm cfg_ft.elecfile='GSN-HydroCel-129.sfp'; cfg_ft.layout = ft_prepare_layout(cfg_ft); cfg_ft.showlabels = 'yes'; cfg_ft.colorbar = 'yes'; cfg_ft.showoutline = 'yes'; figure ft_multiplotTFR(cfg_ft,data_conn); I'm still curious about the other questions I posted. I'd be happy to take that discussion elsewhere (I can reply to the email you sent on the list) if this bug report isn't the right place. Let me know. Thanks, Matt