go-pegasos

An implementation of the Pegasos algorithm [1] for solving Support Vector Machines in Go.

Build Instructions

Software Requirements

• Go

Get the code

$ go get github.com/tetsuok/go-pegasos

Installation of commands

$ go install github.com/tetsuok/go-pegasos/pegasos_learn
$ go install github.com/tetsuok/go-pegasos/pegasos_test

Testing

$ go test github.com/tetsuok/go-pegasos/pegasos

If you want to run testing including benchmarks, use check.sh

$ ./check.sh

Usage

Data format

go-pegasos accepts the same representation of training data as SVMlight uses. This format has potential to handle large sparse feature vectors.

Training

$ ./pegasos_learn -m model_file train_file

Please note "-m" is required to save the trained model.

Options

• -k INT: number of block size.
• -lambda FLOAT: Regularization parameter
• -m STRING: model file
• -r INT: seed
• -t INT: number of iterations
• -test STRING: If you set a test file, you can do training and testing at a time.

Testing with trained model

$ ./pegasos_test test_file model_file

Reference

[1] Shalev-Shwartz, Shai and Singer, Yoram and Srebro, Nathan. Pegasos: Primal Estimated sub-GrAdient SOlver for SVM. In Proceedings of the 24th international conference on Machine learning (ICML). 2007. pages 807-814.