The library can be used to attack large-scale classification problems. It is really fast!
- Pegasos SVM
- Linear Perceptron
- Passive-Agressive Perceptron
- Averaged Perceptron
Usage CLI (Experimental)
Simple-ML similar in usage to LibSVM and SVM-Light. The library is written in Java what requires JRE 7 installed on your OS. The convinient way to use
simple-ml-*.*-with-deps.jar distribution, which requires minimum efforts in installation.
Simple-ML consists of a training module and a classification module. Classification module is used to apply the learned model to test examples.
The training module is called with the following parameters:
java -jar "simple-ml-0.1-with-deps.jar" train [options] <training_set_file> <model_file>
The classification module is called with the following parameters:
java -jar "simple-ml-0.1-with-deps.jar" classify [options] <model_file> <test_file> <output_file>
The only available option now is:
-t classifier_type : 0 - linear perceptron (default) 1 - averaged linear perceptron 2 - passive-aggressive perceptron 3 - Pegasos SVM
LibSVM Input Data Format
Simple-ML compatible with LibSVM data format:
<label> <index1>:<value1> <index2>:<value2> ... . . .
Each line contains one instance. For classification,
<label> is an integer indicating the class label (CLI supports only binary classification what restricts
<label> to be
+1, but multilabel classification is supported in API). The pair
<index>:<value> gives a feature value:
<index> is an integer starting from 1 and
<value> is a real number.
Each line of output file contains
<label> for the corresponding instance of the test file.
If you are keen to use Simple-ML for non-commertial/research projects, please spread the project link https://code.google.com/p/simple-ml/ wherever you want and can. Also, drop me an e-mail if you want to use Simple-ML for commertial purposes.
Now you know everything to be able to use Simple-ML in your data minig tasks. The library is still under our lazy development, so if you find a bug or you have a suggestion we would be glad to know. Please use the issue tracker or e-mail. If you want to become a contributor, fell free to drop me an e-mail.
Since the training and classification algorithms implemented in Simple-ML are so fast, you will definitely have more time for
beer your research!