Back to the main page.

Bug 204 - ft_freqanalysis crashes when specest implementation is used

Reported 2010-11-08 15:52:00 +0100
Modified 2010-11-09 16:51:07 +0100
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Mac OS
Importance: P1 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on:
See also:

Jan-Mathijs Schoffelen - 2010-11-08 15:52:24 +0100

this happens when a trial returns an empty spectrum in the case of mtmconvol, which happens when the data is shorter than the timewindow requested. JM will fix this.

Roemer van der Meij - 2010-11-09 09:29:28 +0100

Hmmm, doesn't this mean that all frequencies/timepoints will only have NaNs (because of equal padding across trials)? Leading to a faulty output in all cases? If I'm right, it might be easiest to throw an error if all(~isnan)==1, instead of building a fix around it. Or was this what you were thinking about?

Jan-Mathijs Schoffelen - 2010-11-09 16:50:59 +0100

the spectrum returned in such case actually contains only nans, which originally resulted in problems when the matrix containing non-nan numeric data was reshaped, i.e. the reshape operation failed. this has now been fixed by explicitly allocating the variables related to the new size in a different way. moreover, along the way it turned out that the nan-allocation for the output data (mtmconvol when no estimate can be made for a particular voxel) was incorrect in the case of keeptrials and keeptapers. this has also been resolved, I think.