[FieldTrip] slight change in default behaviour ft_componentanalysis (method = 'runica'/'binica')

jan-mathijs schoffelen jan.schoffelen at donders.ru.nl
Fri Jul 8 21:06:10 CEST 2011


Dear all,

I implemented a slight change in the default behaviour of ft_componentanalysis when using runica or binica as method. 

In general componentanalysis unmixes the sensor level data, and provides a spatial mixing matrix (comp.topo), as well as the time course of the components (in comp.trial). By construction (if you did comp=ft_componentanalysis(cfg,data)) data.trial{x} = comp.topo*comp.trial{x};

The change in behaviour pertains to the relative scaling of the mixing matrix and the component time-courses. Before, default behaviour caused the magnitude of the values in the mixing matrix to be of the same order of magnitude as the values in the original time courses (data.trial). For reasons I will not go into, this could cause numerical problems at some later stage of some analysis pipeline. Also, when using other methods in ft_componentanalysis (such as pca), the magnitude of the values in the component time courses approximately matched the magnitude of the values in the original time courses. 
For consistency and to avoid potential problems at later stages of analysis, I changed the default behaviour of ft_componentanalysis (with method runica or binica) such that the scaling is contained in the component time courses, rather than in the mixing matrix. Note that this only changes the output of the function up to a scaling of the mixing matrix and the time courses of the components.
If you don't want this behaviour, or if you want to keep the behaviour consistent to the old times, you need to specify cfg.normalisesphere = 'no'.

Have a good weekend,

Jan-Mathijs

Dr. J.M. (Jan-Mathijs) Schoffelen 
Donders Institute for Brain, Cognition and Behaviour, 
Centre for Cognitive Neuroimaging,
Radboud University Nijmegen, The Netherlands
J.Schoffelen at donders.ru.nl
Telephone: 0031-24-3614793

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110708/2d688007/attachment.html>


More information about the fieldtrip mailing list