Back to the main page.

Bug 2891 - test_ft_componentanalysis fails in dashboard

Reported 2015-05-07 14:04:00 +0200
Modified 2016-01-14 21:28:30 +0100
Product: FieldTrip
Component: test
Version: unspecified
Hardware: PC
Operating System: Windows
Importance: P5 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on: 2585
See also:

Jim Herring - 2015-05-07 14:04:37 +0200

test_ft_componentanalysis fails in the dashboard due to a mismatch with the generated data and the data on disk for a specific dataset. However, when I run it (matlab2014b, FT-version r10831), it runs fine. Robert suggested the tolerance could be too strict as there are some differences in convergence due to randomness from run to run. The tolerance is currently set to 1e-7, but the differences I get are in the order of magnitude of e-10. I've now run test_ft_componentanalysis roughly 20 times and it never fails for me so it seems something else is going wrong.

Jan-Mathijs Schoffelen - 2015-07-02 09:19:50 +0200

This one is possibly related to bug 2585, taking over... Note also, that currently the dashboard runs with matlab2012b, thus in general any numeric differences that arise from a different matlab2014b are candidates to cause dashboard failures when testing against da data stored on disk.

Jan-Mathijs Schoffelen - 2015-09-17 09:12:44 +0200

It appeared to be a polarity issue with regard to the principal components in itab153 data. I saved a fresh set of reference data for this datatype. Runs through fine now.