practice libsvm

practice the guide.pdf saved in C:\Temp\learn_libsvm\test1_research_the_GUIDE

cd /media/sf_Temp/learn_libsvm/test3_test_libsvm/libsvm-3.23 && make

===========================================================================
this test is for demo purpose(use easy.py in one step):

use the data of project Astroparticle under https://www.csie.ntu.edu.tw/~cjlin/papers/guide/data/ ,
it has training dat as train.1 and test dat as test.1 :
save train.1 into svmguide1
save test.1 into  svmguide1.t

//feature|attr value range [-1, 1], 
./svm-scale -l -1 -u 1 -s range1 svmguide1 > svmguide1.scale
/*
1 1:2.617300e+01 2:5.886700e+01 3:-1.894697e-01 4:1.251225e+02
...
into
1 1:-0.823781 2:-0.783405 3:-0.233795 4:0.361305 
...
0 1:-0.885045 2:-0.949015 3:0.182549 4:-0.519316 

has 2 classes,  class label Infected or Clean
*/

./svm-scale -r range1 svmguide1.t > svmguide1.t.scale

./svm-train svmguide1.scale
/*
*
optimization finished, #iter = 496
nu = 0.202599
obj = -507.307046, rho = 2.627039
nSV = 630, nBSV = 621
Total nSV = 630
*/

./svm-predict svmguide1.t.scale svmguide1.scale.model svmguide1.t.predict
Accuracy = 96.15% (3846/4000) (classification)


+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

to achieve better Accuracy, ./svm-train requires cross-validation and specify following args:
-c -g 

the RBF kernel function argument C and galma

//the grid tools is a time consumer
./tools/grid.py svmguide1.scale
2.0 2.0 96.9893

./svm-train -c 2 -g 2 svmguide1.scale
Accuracy = 96.875%
===========================================================================
use easy.py in one step:
apt-get install -y gnuplot
cd tools/ && ./easy.py ../svmguide1 ../svmguide1.t

Scaling training data...
Cross validation...
Best c=2.0, g=2.0 CV rate=96.9893
Training...
Output model: svmguide1.model
Scaling testing data...
Testing...
Accuracy = 96.875% (3875/4000) (classification)
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++