News

This page is dedicated to the EasySVM package. It provides a set of tools based on the Shogun toolbox allowing to train and test SVMs in a simple way. This toolbox is integrated in our Galaxy server.

The updated release (easysvm-0.3.3.tar.bz2) can be downloaded here. Other releases can be found here.

Dependencies

• numpy (>=1.0.1)
• pylab (>=0.87.7) [optional]
• shogun (>=0.6.0)
• arff [optional]

A simple example

In the following very simple example we generate a two-dimensional data set with two Gaussian-distributed classes (60% positive examples, width of distribution 1.3):

python scripts/datagen.py cloud 1000 2 0.6 1.3 cloud.arff

python scripts/easysvm.py modelsel 5 0.1,1,10 gauss 0.1,1,10 arff cloud.arff modelsel-cloud.txt

python scripts/easysvm.py cv 5 10 gauss 1 arff cloud.arff cv-cloud.txt

Tutorial examples

Many examples of using easysvm are discussed in a tutorial paper. The results of the paper can by reproduced by a script in tutorial-example.py. Execute it in the data directory:

cd data

python ../splicesites/tutorial_example.py

The output of this script can be downloaded here: tutorial-example.out

Galaxy interface

The following command line arguments are what is behind the galaxy interface, which is available as a web service from http://galaxy.raetschlab.org/

There are three types of data creation methods:

datagen.py motif arff gattaca 10 50 10-15 0.1 tttt 100 50 15 0.1 testmotif1.arff

datagen.py cloud 100 3 0.6 1.3 testcloud1.arff

datagen.py motif arff gattaca 100 50 10-15 0.1 tttt 1000 50 15 0.1 testmotif2.arff

datagen.py cloud 1000 3 0.6 1.3 testcloud2.arff

datagen.py motif fasta gattaca 10 50 10-15 0.1 testmotifpos.fasta

datagen.py motif fasta tttt 100 50 15 0.1 testmotifneg.fasta

datagen.py motif fasta gattaca 100 50 10-15 0.1 tm1.fasta

datagen.py motif fasta tttt 1000 50 15 0.1 tm2.fasta

cat tm1.fasta tm2.fasta > testmotiftest.fasta

rm tm1.fasta tm2.fasta

Cross validation and evaluation on a independent validation set:

easysvm.py cv 5 10 gauss 0.6 arff testcloud1.arff cv_cloud.txt

easysvm.py eval cv_cloud.txt arff testcloud1.arff cv_cloud_eval.txt roc roc_cloud_cv.png

easysvm.py cv 5 10 wd 10 2 arff testmotif1.arff cv_motif.txt dna R

easysvm.py eval cv_motif.txt arff testmotif1.arff cv_motif_eval.txt roc roc_motif_cv.png

Predict on a test set:

easysvm.py pred 10 gauss 0.6 arff testcloud1.arff testcloud2.arff pred_cloud.txt

easysvm.py pred 10 linear arff testcloud1.arff testcloud2.arff pred_cloud.txt

easysvm.py pred 10 poly 3 true true arff testcloud1.arff testcloud2.arff pred_cloud.txt

easysvm.py pred 10 wd 10 2 arff testmotif1.arff testmotif2.arff pred_motif.txt dna R

easysvm.py pred 10 localalign arff testmotif1.arff testmotif2.arff pred_motif.txt dna R

easysvm.py pred 10 localimprove 10 1 1 arff testmotif1.arff testmotif2.arff pred_motif.txt dna R

For some kernels, investigate the importance of different motives:

easysvm.py poim 10 6 wd 10 2 arff testmotif1.arff poims.png dna R

We also support the fasta format:

easysvm.py cv 5 10 wd 10 2 fasta testmotifpos.fasta testmotifneg.fasta cv_motif.txt dna R

easysvm.py eval cv_motif.txt fasta testmotifpos.fasta testmotifneg.fasta cv_motif_eval.txt roc roc_motif_cv.png

easysvm.py pred 10 wd 10 2 fasta testmotifpos.fasta testmotifneg.fasta testmotiftest.fasta pred_motif.txt dna R

easysvm.py poim 10 6 wd 10 2 fasta testmotifpos.fasta testmotifneg.fasta poims.png dna R