| Author |
 Topic  |
|
|
mcarter
New Member
22 Posts |
 Posted - Jun 08 2007 : 3:08:57 PM
|
I am trying to use the support vector machine package from:
http://www.csie.ntu.edu.tw/~cjlin/libsvm/
Has anyone gotten in running here? The easiest solution seems to be finding/installing matlab 7.1 but if others have experience with it advice would be appreciated.
McKell
|
|
|
mcarter
New Member
22 Posts |
Posted - Jun 11 2007 : 09:12:09 AM
|
Going to bump this and just ask if there is a computer with matlab 7.1 available in BIAC somewhere.
|
 |
|
|
vinod
Average Member
USA
68 Posts |
Posted - Jun 11 2007 : 12:48:54 PM
|
| Maybe the original C-code can be compiled on golgi for general use? Is that possible? Then we can do away with the Matlab interface. I am interested in having access to this program too. |
|
|
|
syam.gadde
BIAC Staff
USA
421 Posts |
Posted - Jun 11 2007 : 2:22:03 PM
|
| I have compiled it and put the binaries in /usr/local/packages/libsvm-2.84 on golgi for you to try out. |
|
|
|
clithero
Junior Member
37 Posts |
Posted - Jun 13 2007 : 2:57:29 PM
|
Hey Syam,
Thanks for compiling those files. I have been playing around with it and am getting a seg fault pretty consistently.
Below is my command line:
[clithero@golgi: /usr/local/packages/libsvm-2.84/bin] $ ./svm-train /mnt/huxley/
data/Class.02/Students/RLclass/SVM/libsvm-2.84 heart_scale
Segmentation fault
If you have any advice we would all appreciate it.
Thanks,
John |
|
|
|
syam.gadde
BIAC Staff
USA
421 Posts |
Posted - Jul 02 2007 : 3:56:18 PM
|
I'm sorry I seem to have missed this last message. Did you find out what went wrong? svm-train takes only one argument, the model file (which presumably your heart_model file is), so
./svm_train heart_model
or, in general:
/usr/local/packages/libsvm-2.84/bin/svm-train /path/to/heart_model
should work, but let me know if you are still getting segfaults. |
|
|
|
clithero
Junior Member
37 Posts |
Posted - Jul 24 2007 : 4:45:27 PM
|
Thanks for that, Syam, those two seem to be working well.
Was there anything else to compile in there? Right now I see 'svm-train' and 'svm-predict', but do not see 'svm-scale', which I believe should be part of the package.
Thanks. |
|
|
|
syam.gadde
BIAC Staff
USA
421 Posts |
Posted - Jul 24 2007 : 4:48:39 PM
|
| Sorry. It is now there. |
|
|
|
clithero
Junior Member
37 Posts |
Posted - Jul 25 2007 : 5:59:10 PM
|
Thanks again, Syam.
We've been trying to run easy.py on golgi, which should output a figure. However, plotting the figure seems to be (perhaps) a problem when running python through golgi.
[mcarter@golgi:
/mnt/huxley/data/Class.02/Students/RLclass/SVM/RLclass2/voxels]$
../../libsvm-2.84/tools/easy.py RLclass2_svm_vox36.train
RLclass2_svm_vox36.train
Scaling training data...
Cross validation...
sem_trywait: Permission denied
sem_wait: Permission denied
sem_post: Permission denied
Warning: empty z range [58.3333:58.3333], adjusting to [57.75:58.9166]
Notice: cannot contour non grid data!
Warning: empty z range [58.3333:58.3333], adjusting to [57.75:58.9166]
Notice: cannot contour non grid data!
Warning: empty z range [58.3333:58.3333], adjusting to [57.75:58.9166]
Warning: empty z range [58.3333:58.3333], adjusting to [57.75:58.9166]
gnuplot is installed. We are most concerned with the "permission denied." Any suggestions? Thanks again. |
|
|
|
syam.gadde
BIAC Staff
USA
421 Posts |
Posted - Jul 26 2007 : 08:10:09 AM
|
I guess you found gnuplot in /usr/local/bin/gnuplot.
I ran this, and despite the error messages, it did produce a plot (RLclass2_svm_vox36.train.scale.png). However, I'm not sure what it is supposed to look like. The permission denied stuff is probably harmless, but not sure. Do you know what the output is supposed to look like and whether what you are getting is different than what you should be getting? |
|
|
|
vinod
Average Member
USA
68 Posts |
Posted - Jul 26 2007 : 09:26:21 AM
|
Looks to be some issue with creating semaphores, not sure if it is related to security settings in some way (apparently the problem does not result when executed from root). It also seems to be a well acknowledged problem with python, as revealed by a quick google search. Here is a possible useful thread, perhaps Syam can make more sense of it than me :)
http://mail.python.org/pipermail/python-bugs-list/2005-June/029290.html |
|
|
| |
Topic |
|
libsvm / matlab 7.1
