Back to the main page.

Bug 2715 - ft_math: suggestions for improvement

Reported 2014-09-30 12:32:00 +0200
Modified 2015-01-12 09:20:02 +0100
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Mac OS
Importance: P5 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on:
See also:

Jan-Mathijs Schoffelen - 2014-09-30 12:32:37 +0200

I was looking into the possibility to use cfg.operation in combination with cfg.scalar, where cfg.scalar is a matrix ;-). Purpose: doing a bias correction on a TFR power spectrum according to e.g. Bokil et al J Neurosci Methods 20067: m = repmat(shiftdim(freq.dof./2,-1),[numel(freq.label) 1 1]); powspctrm_biascorrected = log(freq.powspctrm) - psi(m) + log(m); The idea here would be that m has the same size as freq.powspctrm, but is not scalar (because it may vary over time). I can get this to work with: cfg.operation = 'log(x1)-psi(s)+log(s)'; cfg.scalar = m; in combination with a small hack around line 237: if isscalar(s) y = arrayfun(... elseif size(s)==size(arginval{1}) y = feval(operation, arginval{:}); end This is of course not uber-nice because strictly speaking cfg.scalar is not scalar anymore. Shall we build in support for non-scalar cfg.scalar (and call it cfg.matrix)? I think it should be do-able if we explicitly require this field to have the same size as the data.(cfg.parameter). Along the way I noticed that cfg.scalar is not touched and lost to posterity in the output.cfg. Although ugly, I suggest to touch it (if user-defined) around line 177

Robert Oostenveld - 2014-09-30 20:43:38 +0200

agreed (cfg.matrix) and agreed (touching them for posterity). I suggest to keep cfg.matrix outside of the regular documentation.

Jan-Mathijs Schoffelen - 2014-10-01 09:54:43 +0200

bash-4.1$ svn commit -m "enhancement - added undocumented option cfg.matrix for element wise operations, and touch cfg.scalar/matrix for output cfg" ft_math.m Sending ft_math.m Transmitting file data . Committed revision 9868.