SVMPy

By Andrew Tulloch (http://tullo.ch)

Introduction

This is a basic implementation of a soft-margin kernel SVM solver in Python using numpy and cvxopt.

See http://tullo.ch/articles/svm-py/ for a description of the algorithm used and the general theory behind SVMs.

Demonstration

Run bin/svm-py-demo --help.

∴ bin/svm-py-demo --help
usage: svm-py-demo [-h] [--num-samples NUM_SAMPLES]
                   [--num-features NUM_FEATURES] [-g GRID_SIZE] [-f
                   FILENAME]

optional arguments:
  -h, --help            show this help message and exit
  --num-samples NUM_SAMPLES
  --num-features NUM_FEATURES
  -g GRID_SIZE, --grid-size GRID_SIZE
  -f FILENAME, --filename FILENAME

For example,

bin/svm-py-demo --num-samples=100 --num-features=2 --grid-size=500 --filename=svm500.pdf

yields the image