[VoxBo] Breaking regressions into parts?

[jhyyoon at ucdavis.edu]

Sent via BlackBerry from T-Mobile  

-----Original Message-----
From: Daniel Y Kimberg <kimberg at mail.med.upenn.edu>
Date: Sat, 24 Mar 2007 16:36:09 
To:VoxBo Discussion List <voxbo-general at voxbo.org>
Subject: Re: [VoxBo] Breaking regressions into parts?

Thad Polk wrote:
> Not an hour after asking why the glm widget doesn't split
> regressions into parts, I see a sequence on my cluster, submitted by
> the glm widget, doing just that!  So let me rephrase the questions:
> Under what circumstances does the glm break the regression up? Into
> how many parts? Is this something the user can control?

By default, the glm scripting uses a pretty dumb heuristic.  You can
force the number of pieces in the .glm script (for use with vbmakeglm,
see "vbmakeglm -h"), but not when you use the graphical interface.
There's no particular reason it's not in the graphical interface, I
just happened to add it to the script parser without asking Dongbo to
add it.  Release 1.8.5 is pretty imminent, I can probably get him to
add this flag to the revamped glm widget.

dan

ps  in case you're wondering, the heuristic is to divide the glm into
ceil((ndata^2)/600000), where ndata is the number of data points.