Back to the main page.
Bug 931 - ft_appendfreq fails when the data does not exactly overlap
| Status | CLOSED WONTFIX |
| Reported | 2011-09-05 09:15:00 +0200 |
| Modified | 2011-09-14 14:33:31 +0200 |
| 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: |
Robert Oostenveld - 2011-09-05 09:15:23 +0200
This was reported by Yoni at http://bugzilla.fcdonders.nl/show_bug.cgi?id=798#c19 and it can be reproduced by the following piece of code (which I'll save to the test directory) freq1.label = {'1'}; freq1.time = [1 2]; freq1.freq = [1 2 3]; freq1.dimord = 'chan_freq_time'; freq1.powspctrm = randn(1,3,2); freq2.label = {'1'}; freq2.time = [1 2]; freq2.freq = [1 2 3] + 1; % shifted by one freq2.dimord = 'chan_freq_time'; freq2.powspctrm = randn(1,3,2); cfg = []; cfg.parameter = 'powspctrm'; ft_appendfreq(cfg, freq1, freq2); the input is freq data with 1 channels, 3 frequencybins and 2 timebins the input is freq data with 1 channels, 3 frequencybins and 2 timebins the input is freq data with 1 channels, 3 frequencybins and 2 timebins the input is freq data with 1 channels, 3 frequencybins and 2 timebins ??? Error using ==> ft_appendfreq at 203 the input data structures have non-unique frequency bins, concatenation across frequency is not possible Error in ==> ft_appendfreq at 130 freq = ft_appendfreq(tmpcfg, varargin{:});
Robert Oostenveld - 2011-09-05 09:19:24 +0200
manzana> svn commit test_bug931.m Adding test_bug931.m Transmitting file data . Committed revision 4098.
Jan-Mathijs Schoffelen - 2011-09-05 09:29:24 +0200
fixed in ft_appendfreq with cfg.tolerance.
Robert Oostenveld - 2011-09-05 09:38:10 +0200
The error is > the input data structures have non-unique frequency bins, concatenation across > frequency is not possible which is actually consistent with the data: freq1 has [1 2 3], freq2 has [2 3 4] and these cannot be concatenated. Since the problem started in ft_freqstatistics, one might rephrase it as cfg = []; cfg.parameter = 'powspctrm'; cfg.appenddim = 'rpt'; % be explicit ft_appendfreq(cfg, freq1, freq2); failing. The question is whether ft_appendfreq should ensure the overlap of the freq axis (i.e. cut of the first sample in freq1 and the last sample in freq2)? The old ft_freqstatistics was able to deal with this input, so as such the bug remains present, but might not be in ft_appendfreq persee.
Yoni Levy - 2011-09-05 09:52:19 +0200
(In reply to comment #3) > The error is > > > the input data structures have non-unique frequency bins, concatenation across > > frequency is not possible > > which is actually consistent with the data: freq1 has [1 2 3], freq2 has [2 3 > 4] and these cannot be concatenated. > As far as I can see the freq1.freq and freq2.freq in my dataset are identical; but perhaps as you wrote it is inherently shifted by 1 by using the function.
Robert Oostenveld - 2011-09-05 11:17:29 +0200
I discussed this in detail with Jan-Mathijs. For the example script the error-message that is produced by ft_appendfreq is actually correct and desired. The problem in the data of Yoni is due to a small numerical difference in the time axis between the (otherwise consistent) two datasets. This particular problem does not have to be fixed, but ft_selectdata should use nearest (which it did once upon a time, but which it does not do any more).