tfanalysis - NaN

Virginie van Wassenhove vvw at CALTECH.EDU
Wed Jun 27 04:15:00 CEST 2007


Hi Jan-M and Christian

thanks very much for all your advice.
- there was indeed a time discrepancy in that the imported files were
in milliseconds (is this a usual thing when importing from BESA? If
so, it may be helpful to have this piece of information on the Besa
section in the website). So, converting everything in seconds did
help together with...
- longer epochs which of course provide more sensible results.

No NaN on the horizon, clusteranalysis on the way, I might inquire
some more soon on that part...

Thanks again for your help!
-vv

>Sorry to hop in into this discussion, but did you (V) explicitly check the
>time-axes present in your data? In other words: there is a field in the data
>structure (data.time) which specifies the time axis for each trial.
>According to this axis, windows (cfg.t_ftimwins) are placed around the
>(cfg.tois) you specify in your configuration. If there is a discrepancy
>between cfg.toi, and data.time, NaNny things could happen. Because you
>imported your files from BESA, something might be going on here.
>
>Yours,
>
>Jan-M
>
>-----Original Message-----
>From: FieldTrip discussion list [mailto:FIELDTRIP at NIC.SURFNET.NL] On Behalf
>Of Virginie van Wassenhove
>Sent: Tuesday, June 26, 2007 6:25 AM
>To: FIELDTRIP at NIC.SURFNET.NL
>Subject: Re: [FIELDTRIP] tfanalysis - NaN
>
>Hi Christian,
>
> >In the configuration options you have specified cfg.taper =
> >'hanning' which will apply only one (i.e., the hanning) taper to
> >your data. If you want to use multitapers you need to set cfg.taper
> >= 'dpss'  or just leave the field it out of the cfg (since
> >multitapers are the default).
>
>Ah yes, sorry, I have been playing around with most parameters so
>both 'hanning' and 'dpss' lead to the same result (and other windows
>and wavelet analysis).
>
> >I'm not really sure what you mean by normalized (the csd is
> >obviously a density) but if there were any divide by zero issues
> >then matlab would be chucking warnings. What happens when you set
> >cfg.output = 'pow'; ?
>
>Same thing whether I use 'pow' or 'powandcsd'. So I meant 'pow'
>normalization (I assume it's baseline corrected on the pre-stimulus
>interval?).
>
>Since I imported raw data from besa (as .dat files) there might be an
>error in the previous specifications / preprocessing. I don't see
>otherwise see any major confound in the tf analysis... I will look
>some more into that part.
>
>Thanks for your help!
>-vv
>
>
>
>
> >>>>>>>Hi Virginie,
> >>>>>>>
> >>>>>>>the function freqanalysis will return NaNs when a time
> >>>>>>>frequency tile extends beyond a corresponding data segment.
> >>>>>>>One would usually expect a "U" shaped padding of NaNs at the
> >>>>>>>boundaries of any TFR which has longer time windows for lower
> >>>>>>>frequencies. From the cfg options you specify below I can see
> >>>>>>>that you have a data segment of 1.1 seconds length and the
> >>>>>>>moving windows cfg.t_tfimwin are quite long at the low
> >>>>>>>ffrequencies, e.g., 1 Hz = 4 s, 6 Hz  = 0.6666 s. Hence the
> >>>>>>>NaNs seem perfectly "normal" in this case. The only way to
> >>>>>>>avoid NaN in the time window of interest (using this type of
> >>>>>>>time-frequency tiling) is to make your epochs longer.
> >>>>>>>
> >>>>>>>Hope this helps,
> >>>>>>>Christian
> >>>>>>>
> >>>>>>>
> >>>>>>>On 21 Jun 2007, at 20:44, Virginie van Wassenhove wrote:
> >>>>>>>
> >>>>>>>>Hello again.
> >>>>>>>>
> >>>>>>>>I am finding NaN whatever method I use (multitaper and
> >>>>>>>>wavelet analysis), yet using a different analysis tool with
> >>>>>>>>similar parameters I obtain clean results.
> >>>>>>>>epoch length    = 1.1 s
> >>>>>>>>baseline length = 0.3 s
> >>>>>>>>sampling rate   = 500Hz
> >>>>>>>>
> >>>>>>>>I tried several suggestions listed in the previous message
> >>>>>>>>list for a similar problem, but none seem to provide
> >>>>>>>>reasonable results (aka always NaN).
> >>>>>>>>Also tried with different subjects, no particular error
> >>>>>>>>otherwise. I am sending a sample script in case someone has a
> >>>>>>>>suggestion on this...
> >>>>>>>>Thanks for your help once more!
> >>>>>>>>
> >>>>>>>>Virginie
> >>>>>>>>
> >>>>>>>>%% ================================================
> >>>>>>>>% TF analysis - MULTITAPER
> >>>>>>>>cfg             = [];
> >>>>>>>>cfg.method      = 'mtmconvol';
> >>>>>>>>cfg.taper       = 'hanning';
> >>>>>>>>cfg.output      = 'powandcsd';
> >>>>>>>>%cfg.channel    = 'all';
> >>>>>>>>cfg.channel     = 'E75';
> >>>>>>>>
> >>>>>>>>%win_length     = 0.128;
> >>>>>>>>%win_n  = floor(2.*(1.1./win_length)-1);
> >>>>>>>>
> >>>>>>>>cfg.foi = 1:5:45;
> >>>>>>>>%cfg.t_ftimwin = ones(1,win_n).*win_length;
> >>>>>>>>cfg.t_ftimwin   = 4./cfg.foi;
> >>>>>>>>%cfg.toi        = -0.3:win_length/2:0.8;
> >>>>>>>>cfg.toi         = -0.3:0.05:0.8;
> >>>>>>>>
> >>>>>>>>% cfg.pad       = 'maxperlen';
> >>>>>>>>% cfg.keeptrials = 'no';
> >>>>>>>>% cfg.keeptapers ='no';
> >>>>>>>>s08_ill_mult = freqanalysis(cfg,s08_ill);
> >>>>>>>>
> >>>>>>>>%% ========================
> >>>>>>>>% plot freq data
> >>>>>>>>cfg             = [];
> >>>>>>>>cfg.baseline    = [-0.3 0];
> >>>>>>>>cfg.layout      = egilay;
> >>>>>>>>cfg.zlim        = [-0.00001 0.000001];          % a comparer
> >>>>>>>>entre participants
> >>>>>>>>% warning off                 % NaN
> >>>>>>>>clf
> >>>>>>>>%multiplotTFR(cfg,s08_ill_mult);
> >>>>>>>>singleplotTFR(cfg,s08_ill_mult);
> >>>>>>>>
> >>>>>>>>%% ================================================
> >>>>>>>>% TF analysis - wavelet
> >>>>>>>>cfg             = [];
> >>>>>>>>cfg.method      = 'wltconvol';
> >>>>>>>>cfg.width               = 4;
> >>>>>>>>cfg.output       = 'pow';
> >>>>>>>>cfg.foi                 = 1:2:30;
> >>>>>>>>cfg.toi                 = -0.3:0.05:0.8;
> >>>>>>>>s14_ill_wvlt    = freqanalysis(cfg, s14_ill);
> >>>>>>>>%% ========================
> >>>>>>>>cfg = [];
> >>>>>>>>cfg.baseline    = [-0.3 0];
> >>>>>>>>cfg.zlim        = [-1 1];
> >>>>>>>>cfg.showlabels = 'yes';
> >>>>>>>>cfg.layout      = egilay;
> >>>>>>>>clf
> >>>>>>>>multiplotTFR(cfg, s14_ill_wvlt)
> >>>>>>>>
> >>>>>>>>%% ================================================

----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/fcdonders/fieldtrip.



More information about the fieldtrip mailing list