Hi Henk,
The -V makes no difference or whatsoever. It was an option that existed
in randomise to produce verbose outputs, and it eventually just ended up
staying in the script. You can safely ignore it.
The actual problem is that there is no outputs being produced. Three
possible causes are:
1) Perhaps some issue with your cluster settings
2) Perhaps editing wrong scripts;]
3) Perhaps some error when constructing the design matrix that you say
you did manually.
Regarding (1): It's not necessary much processing power or a full
cluster environment to compute just the t-stat and F-stats, as no
permutations will be performed. You can bypass the cluster altogether
and run randomise in a single machine and see if this solves the issue.
Just run:
randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d design.mat -t
design.con -f design.fts
Regarding (2): When editing the scripts, it's generally a better idea to
make the changes to fslvbm_3_proc, not to the fslvbm3b file. Although in
principle both should work, if you change the second you need to make
the submission manually, which is a source of potential errors.
Regarding (3): It is fine to make the design files by hand. However, if
you are using Windows, you need to take care of the new-line character,
which is different on that operating system. You can use "dos2unix"
command to fix this, or just use any text editor inside Linux (vi,
emacs, nano, kate, gedit, whatever you like), from scratch, to make the
files, and they will be correct.
About the /PPHeights: no, they are not necessary. The minimum fields the
design files have to contain for randomise are /NumWaves, /NumPoints and
/Matrix.
Hope this helps.
All the best,
Anderson
Am 06.02.14 08:59, schrieb H van Steenbergen:
> Dear FSL-ers,
>
> I tried to run randomise on VBM data with inference turned off, just to quickly check whether our statistical model makes sense.
>
> In the flsvbm/Userguide it is mentioned that the fslvbm_3_proc script runs randomise with inference (generation of p-value maps) turned off, so that it very quickly creates just the raw tstat maps. When looking into the fslvbm3b script created, it seems to call
>
> $FSLDIR/bin/randomise -i ${i}_s${j} -o ${i}_s${j} -m GM_mask -d design.mat -t design.con -V
>
> Because I also want to have raw F-tests I changed this line of code by adding the parameter -f design.fts, but running the command below seems not to work (it runs for hours on the cluster without producing output, also no errors reported).
>
> fsl_sub randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d design.mat -t design.con -f design.fts -V &
>
> Replacing -V with -1 (perform 1-sample group-mean test instead of generic permutation test) does work however. So I was wondering why the -V is here and whether it is necessary. Randomise documentation does not mention -V so what is the functionality?
>
> In addition, is it necessary to add PPheight information to the mat and con files? I have to create these files by hand because the Glm gui is very slow and it took hours to build up the graphics when I tried to model 420 inputs and 20 EVs on the visualization node of our cluster. I am using Windows, so I can't run Glm locally.
>
> Thanks in advance for your help.
>
> Best,
>
> Henk van Steenbergen
