Back to the main page.
Bug 836 - ft_componentanalysis crashes with method='fastica' and MEG data in input
| Status | CLOSED WONTFIX |
| Reported | 2011-07-20 12:22:00 +0200 |
| Modified | 2011-11-11 11:19:55 +0100 |
| Product: | FieldTrip |
| Component: | core |
| Version: | unspecified |
| Hardware: | PC |
| Operating System: | Mac OS |
| Importance: | P1 normal |
| Assigned to: | Jan-Mathijs Schoffelen |
| URL: | |
| Tags: | |
| Depends on: | |
| Blocks: | |
| See also: |
Jan-Mathijs Schoffelen - 2011-07-20 12:22:05 +0200
Some methods seem to have some hard coded assumptions on the magnitude of the eigenvalues of the data covariance matrix. With fastica the small values lead to an explicit crash. With dss no sphering will be performed. There may be other instances causing problems. Solution: temporarily rescale the data within ft_componentanalysis (and 'unscale' before applying the weights).
Jan-Mathijs Schoffelen - 2011-11-10 14:26:51 +0100
This has been fixed by now. Ft_componentanalysis scales the data before the decomposition and unscales it afterwards...