Hey Folks-

I've made a script to generate large feat setup files. You can find it here:

      https://netfiles.umn.edu/xythoswfs/webui/_xy-20594449_1-t_BJWDfhGf

The script is set up to require some text files that list your inputs, X matrix, contrast vectors, and F-test designations. It also requires a basic .fsf file as a template that has all the other info about your analysis set to go. You could skip the input names part by using the Feat GUI to set up everything but the Stats model and then save that as 'template.fsf.' As noted in the script, I wrote this for use with Darwin for Mac OS X, so most users will need to switch out the parts that use 'jot' in favor of 'seq' or some other numerical list generator.

-Jim

___
Jim Porter, M.A.
Ph.D. Candidate
Clinical Science and Psychopathology Research
University of Minnesota








Print

Hi Michael,

Ah right, my tip only helps you with the .mat file.  The other files, though, are much less of a hassle to construct, but I agree, would be nice if it could be scripted up.

-Tom

On Fri, Dec 9, 2011 at 4:48 PM, Michael Harms <[log in to unmask]> wrote:
Hi Tom,
So once you have appropriate .mat, .con, .grp, and .fts files (including
their header information), you would have to run flameo with appropriate
inputs from the command line, right?  That is, there is no capability to
use a manually created design matrix in conjunction with the FEAT GUI,
is there?

thanks,
-MH

On Fri, 2011-12-09 at 15:56 +0000, Thomas Nichols wrote:

If you are comfortable with R, try building your model there and then

asking for the model.matrix, i.e.

       mod=lm(Y~FacA+FacB+Cov+Cov*FacA)  # whatever

       X=model.matrix(mod);

       write.table(X,'X.dat')

Note that R uses "treatment contrasts" to create the dummy variables

for categorical (factor) effects; this means that the first level of

the factor is the reference and coded with all zeros, which means the

intercept will then be interpreted as the fit of that first factor

level.  Alternatively, you can get 'sum to zero' contrasts (where the

first factor level is coded with -1's) by running this before you call

lm

       options(contrasts=c("contr.sum", "contr.helmert"))

And if you have a simple model and can afford to have K regressors for

a K-level factor variable, you can use simple 1/0 dummy variables.

Ask lm to do this by dropping the intercept with this the -1 syntax:

       mod=lm(Y~-1+FacA)

This only works for the first factor variable in the model, though.

-Tom

On Fri, Dec 9, 2011 at 2:39 PM, Stephen Smith <[log in to unmask]>

wrote:

       Indeed - though it could be scripted without the GUI.

       Cheers.

       On 9 Dec 2011, at 14:35, Michael Harms wrote:

Not to my knowledge, unfortunately.

-MH

On Fri, 2011-12-09 at 13:48 +0100, Diederick Stoffers wrote:

Hi,

Is there any easy way around using the FEAT GUI for

setting up huge

design matrices? It would be great if you could specify

text files as

an input in stead of copy/paste into Glm.

Cheers,

Diederick

Begin forwarded message:

From: Michael Harms <[log in to unmask]>

Subject: Re: [FSL] large quantity glm setup

Date: 7 juli 2011 15:40:03 GMT+02:00

To: [log in to unmask]

Reply-To: FSL - FMRIB's Software Library

<[log in to unmask]>

The FEAT GUI can be quite challenging to work with when

specifying

large

designs, as it becomes very slow in responding and

updating.  How

big of

a design are you trying to specify?  Are you having an

issue pasting

in

the values themselves, or is it "freezing" after you

Click OK within

the

Paste Window?

That said, with patience we've always been able to use

the Paste

feature

to accomplish what we need.

cheers,

-MH

On Wed, 2011-07-06 at 15:13 -0700, Amy Zhong Sheng Zheng

wrote:

Thank you for the response. Do I just copy the numbers

from excel

(or

whichever) and paste them after opening up the paste

button?

Because I

just tried that and every time I tried to paste the

numbers in the

paste window, feat would freeze on me. Have any ideas

why that

might

be the case?

On Wed, Jul 6, 2011 at 1:59 PM, Michael Harms

<[log in to unmask]>

wrote:

     Sure, create it in Excel, Matlab, text editor,

or whatever

is

     most

     convenient and just use the "Paste" button under

"Full

model

     setup".

     cheers,

     -MH

     On Wed, 2011-07-06 at 20:55 +0100, Amy wrote:

Hi all,

I'm doing an anova with multiple factors and levels

with

     many inputs, so setting up the 1's and -1's on

the glm has

     been very tedious. I was wondering if anyone

knows a

     faster/efficient way to fill in the values? Is

there a way

to

     export from excel and fill that in the

design.fsf?

Thanks,

Amy

       ---------------------------------------------------------------------------

       Stephen M. Smith, Professor of Biomedical Engineering

       Associate Director,  Oxford University FMRIB Centre

       FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK

       +44 (0) 1865 222726  (fax 222717)

       [log in to unmask]    http://www.fmrib.ox.ac.uk/~steve

       ---------------------------------------------------------------------------

--

__________________________________________________________

Thomas Nichols, PhD

Principal Research Fellow, Head of Neuroimaging Statistics

Department of Statistics & Warwick Manufacturing Group

University of Warwick, Coventry  CV4 7AL, United Kingdom

Web: http://go.warwick.ac.uk/tenichols

Email: [log in to unmask]

Phone, Stats: +44 24761 51086, WMG: +44 24761 50752

Fax:  +44 24 7652 4532

--
__________________________________________________________
Thomas Nichols, PhD
Principal Research Fellow, Head of Neuroimaging Statistics
Department of Statistics & Warwick Manufacturing Group
University of Warwick, Coventry  CV4 7AL, United Kingdom

Web: http://go.warwick.ac.uk/tenichols
Email: [log in to unmask]
Phone, Stats: +44 24761 51086, WMG: +44 24761 50752
Fax:  +44 24 7652 4532