SVM light

SVM light’s input file format

The thing below is basic format for input file of svm light you learn model about

<line> .=. <target> <feature>:<value> <feature>:<value> ... <feature>:<value> # <info>
<target> .=. +1 | -1 | 0 | <float> 
<feature> .=. <integer> | "qid"
<value> .=. <float>
<info> .=. <string>

According to the format above, if I make simple example like this :

1 1:0.43 3:0.12 9284:0.2 # abcdef

  • 1 means target class (total +1 | -1 | 0)
  • a pair of 1:0.42 means 1 is feature number and 0.24 is value of feature.

If a pair of 1:0.42 is vector, 1 is index of a vector, and 0.42 is value of the index(in here 1) of a vector.

If want to rank your data, you can use qid in your training file and test file

How To Learn SVM light and Classify

If you want to learn the model with your input file, type in terminal as follow :

svm_learn [options] example_file model_file

  • example_file : input file, usually training file when you learn
  • model_file : model file you will learn with your input file

If you want to see what the options do. type as follows :

svm_learn -h

Finally If you want to classify test file with your model

svm_classify [options] example_file model_file output_file

  • example_file : input file, usually test file when you classify
  • model_file : model file you learn with training file.
  • output_file : the result that svm light classify test file

SVM Multiclass

SVM Multiclass’s input file format

the basic form is below when you use SVM multiclass.

<line> .=. <target> <feature>:<value> <feature>:<value> ... <feature>:<value> # <info>
<target> .=. <integer>
<feature> .=. <integer>
<value> .=. <float>
<info> .=. <string>

Let’s say you have input file as follows :

3 1:0.43 3:0.12 9284:0.2 # abcdef

  • 3 means target class, in SVM multicalss, you can make more target classes.
  • 1:0.43 means this pair is totally the same from SVM light, i.e. 1 is index of a vector, 0.43 is the value of the index

How To Learn SVM multiclass and Classify

When you learn training file with SVM multiclass, one thing is different.

That is you have to use a option, -c float as follows :

svm_multiclass_learn -c 1.0 example_file model_file

Option c means trade-off between training error and margin (default 0.01)

svm_multiclass_classify [options] test_example_file model_file output_file

And the basic way to utilize SVM multiclass is usually the same from SVM light except for a option, -c float.

SVM light and SVM Multiclass Practice from hyunyoung Lee

Reference