[FieldTrip] ft_sourcestatistics: cfg.design and memory issues

[Anne Urai]

Hi Paul,

in my experience, the cfg field can get rather big as it accumulates
information about previous steps in the pipeline. source = rmfield(source,
'cfg') might help a bit.

Cheers,
Anne

On 3 November 2015 at 00:34, Paul Metzak <pmetzak at gmail.com> wrote:

> Hello,
>
> I am new to FieldTrip (and MEG) and I have been working through the
> walkthroughs and tutorials in an attempt to use beamforming to source
> localize my MEG data. In this experiment, there are 3 conditions and 2
> groups. I have been following the Example: Source Statistics (
> http://www.fieldtriptoolbox.org/example/source_statistics) tutorial but I
> have encountered a couple of difficulties.
>
> 1) The first issue is in the choice of design matrix for the analyses I
> would like to run. The tutorial offers a very clear example of how to
> assess differences between conditions within a single group, but it is less
> straightforward (to me at least) how to run an analysis of the differences
> between two groups on a single condition. In my case, I have 17
> participants from group 1 and 14 participants from group 2. I have created
> 17x1 and 14x1 cell array containing the output from each subject's beamform
> (at the individual subject level, I contrasted a period of interest from
> condition 1 with an equivalent length of time from the prestim baseline).
> My call to ft_sourcestatistics is below. Is this the proper way to set up
> the design matrix to run this analysis? I know that participants are not
> true 'independent variables' but it seems to me that treating them as such
> will randomise group assignment to produce the monte carlo distribution
> that I am looking for.
>
> cfg=[];
> cfg.dim=cond1_h{1}.dim;
> cfg.method      = 'montecarlo';
> cfg.statistic   = 'ft_statfun_indepsamplesT';
> cfg.parameter   = 'pow';
> cfg.correctm    = 'cluster';
> cfg.numrandomization = 1000;
> cfg.alpha       = 0.05; % note that this only implies single-sided testing
> cfg.tail        = 0;
>
> nsubj=17;
> cfg2.design(1,:) = [ones(1,nsubj) ones(1,nsubj)*2];  %create over-large
> desmtx
> cfg.design=cfg.design(:,1:31); %trim desmtx to correct specs
> cfg.ivar        = 1; % row of design matrix that contains independent
> variable (the participants)
>
> stat = ft_sourcestatistics(cfg, cond1_h{:}, cond1_s{:});
>
> 2) The second issue is that it appears I have been inefficient in my
> memory use as matlab crashes when I attempt to run this analysis ( FYI I
> have 16GB of RAM and 16GB of swap space). I have employed the -nojvm flag
> when starting matlab, and I have used struct2single to try and reduce the
> memory load but matlab still crashes out (without message) when I attempt
> to run this analysis. This code will run on my computer with a maximum of 4
> subjects from each group and 50 randomizations, which is obviously not
> ideal for me. Please find the contents of a single cell from one of the
> groups below. Is there something that is obviously problematic in what I
> have so far?
>
>        inside: [256x256x256 logical]
>           pow: [256x256x256 single]
>           dim: [256 256 256]
>     transform: [4x4 single]
>       anatomy: [256x256x256 single]
>      coordsys: 'ctf'
>          unit: 'mm'
>           cfg: [1x1 struct]
>
>
> Thank you very much for any help you can offer me with this! If any
> further information or clarifications would be useful to help diagnose or
> overcome this issue, please let me know.
>
> Paul
>
>
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20151103/5e88203c/attachment.html>