This exercise comes from “A Practical Guide to Support Vector Classification” paper by Chih-WEi Hsu, Chih-Chung Chang and Chih-Jen Lin. It can be downloaded from http://www.csie.ntu.edu.tw/~cjlin/libsvm/index.html.
The goal for writing this page is to record step-by-step instructions of an example in the paper above using Java version in Debian Etch. So, this record will be used as a reference for my future review on LIBSVM.
1) Download libsvm-2.84.tar.gz from the web address above.
2) Unzip the file
3) Change directory to the unzipped directory, libsvm-2.84
> cd libsvm-2.84
> cd java
4) Compile all java files in the libsvm-2.84/java directory
5) Download a dataset “Astroparticle Physics” from http://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/. On the page, click on “svmguide1”. Then, download “svmguide1” and “svmguide1.t”. “svmguide1” is a training dataset and “svmguide1.t” is a testing dataset.
6) Train the dataset
>java svm-train svmguide1
>java svm-predict svmguide1.t svmguide1.model svmguide1.predict
The output from step 7, is Accuracy = 66.925%, which is the same as described in the paper.