Back to the main page.

Bug 903 - cfg.numcomponent fails most methods in ft_componentanalysis

Reported 2011-08-29 21:28:00 +0200
Modified 2014-04-10 11:52:57 +0200
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Mac OS
Importance: P1 normal
Assigned to: Eelke Spaak
Depends on:
See also:

Robert Oostenveld - 2011-08-29 21:28:23 +0200

the second output should have one output channel less than the first, see also the test_ft_componentanalysis.m script which replicates the bug cfg = []; cfg.method = 'pca'; comp1f = ft_componentanalysis(cfg, data); cfg.numcomponent = nchan-1; comp1s = ft_componentanalysis(cfg, data);

Jan-Mathijs Schoffelen - 2011-08-30 09:07:08 +0200

Isn't it the other way around? Only dss and pca give n-1 components when requested, the other methods all give n components back.

Jan-Mathijs Schoffelen - 2011-08-30 09:07:57 +0200

I guess the assertion should be done on nchan-1 for the compXs structures, i.e. those which have been computed with numcomponent = nchan -1 ;

Robert Oostenveld - 2011-08-30 09:26:20 +0200

sorry, you are right. There is a copy-and-paste error in the test script and it should be the other way around

Robert Oostenveld - 2011-08-30 09:27:21 +0200

*** Bug 904 has been marked as a duplicate of this bug. ***

Jan-Mathijs Schoffelen - 2011-08-30 09:36:56 +0200

only pca and dss seem to work

Eelke Spaak - 2011-11-10 17:35:39 +0100

For the decomposition methods that support limiting the number of components estimated, I used this option. For the others, ft_componentanalysis now just deletes any extra components above cfg.numcomponent.

Robert Oostenveld - 2011-11-11 11:19:43 +0100

I closed all bugs that were recently resolved.

Roemer van der Meij - 2014-04-10 11:52:57 +0200

I "reopened with immediate closing" this bug in case of runica, where cfg.numcomponent should be translated into cfg.runica.pca. Fixed in rev 9389