Back to the main page.

Bug 2647 - No clusters using actvsblT in ft_freqstatistics

Reported 2014-07-10 16:42:00 +0200
Modified 2015-01-12 09:21:46 +0100
Product: FieldTrip
Component: core
Version: unspecified
Hardware: All
Operating System: All
Importance: P5 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on:
See also:

Golan Karvat - 2014-07-10 16:42:27 +0200

Created attachment 652 Data and baseline (reduced foi, trials and chan) and the cfg Running ft_freqstatistics with montecarlo correction and using cfg.statistic = 'actvsblT' yields no clusters, and the result structure doesn't have the "interesting" fields (such as posclusterslabelmat and clusterslabelmat). Running the same analysis on the same dataset with cfg.statistic = 'depsamplesT' works fine (but is not correct methodologically because I want to compare the trial time to average baseline power, per trial). It cannot be due to real lack of clusters because even if increase the power of the signal by 1000 (data.powspctrm=data.powspctrm.*1000) or set all base-line levels to -5 there are no clusters (when one would expect the whole data to be one big cluster).

Jan-Mathijs Schoffelen - 2014-09-10 15:30:45 +0200

bash-4.1$ svn add test_bug2647.m A test_bug2647.m bash-4.1$ svn commit -m "enhancement - created test function" test_bug2647.m Adding test_bug2647.m Transmitting file data . Committed revision 9786. The data from the first rpt contains only NaNs, causing the statistics function to return all NaNs in the stat.stat. Apparently, ft_statfun_depsamplesT is robust against this, ft_statfun_actvsblT isn't.

Jan-Mathijs Schoffelen - 2014-09-10 15:31:53 +0200

bash-4.1$ svn commit -m "enhancement - made ft_statfun_actvsblT robust for NaNs in the data" statfun/ft_statfun_actvsblT.m test/test_bug2647.m Sending statfun/ft_statfun_actvsblT.m Sending test/test_bug2647.m Transmitting file data .. Committed revision 9787.