Andrew,
The logs have no meaningful difference. The logs are identical through where they load confoundevs.txt, and then the bad one says:
"Warning: at least one EV is (close to) a linear combination of the others. You probably need to alter your design."
(And yes, confoundevs.txt looks fine.)
On closer inspection, the problem is a bit different than I thought. I saw the "linear combination" error and panicked a bit too hard - the absent confounds were due to stopping my test-analyses too early. HOWEVER, there's still a problem where spurious blank EVs get created and added to my model. (This won't affect the outcome of my analyses, but I'd still like to troubleshoot it.)
These spurious EVs appear for everything I've been running the last day or two, which use the same .fsf design across different inputs and confound-EVs. Near-identical analyses from last week don't have the problem.
Examples:
https://www.dropbox.com/s/o5oj7bg2f00ds7h/spuriousEVs.zipOn Fri, 30 May 2014 17:15:05 +0100, Matthew Webster <[log in to unmask]> wrote:
>Hi Benjamin,
> Are there any differences in the FEAT log between the runs where the confounds been successfully included in the model and the ones where they haven't?
>Kind Regards
>Matthew
>> I have just encountered a strange little problem in FEAT. I enter confound EVs (via the GUI), choosing a file created by fsl_motion_outliers. When I look at the design from inside the FEAT GUI, I see the confound EVs just fine. However, when I actually run the analysis (via GUI or command line), the confound EVs are no longer in the model: sometimes they are absent, sometimes they seem to have been zeroed out.
>>
>> I have previously (i.e. this week) run analyses on the same data where the confound EVs worked fine, I have no idea what's different this time. I've tried all standard troubleshooting steps (reloading/re-saving in GUI, restarting computer and praying, etc), and I'm at a loss for how to track down the source of this problem.
>>
>> Link containing some examples of .png designs, .fsf file, confound EV file:
>> https://www.dropbox.com/s/cf3k461uh3qt7d5/FEATvanishingConfound.zip
>>
>> Thanks,
>>
>> -----
>> Benjamin Philip, Ph.D.
>> Postdoctoral Fellow - Freylab
>> University of Missouri
>> 206 Melvin H. Marx Bldg
>> 1416 Carrie Francke Drive
>> Columbia, MO 65211
>> [log in to unmask]
|