last updated : Nov 3, 2009

This file contains code for encoding and learning piecewise
linear models trained using stochastic gradient descent described
in the paper:

Max-Margin Additive Classifiers for Detection, 
Subhransu Maji and Alexander C. Berg, ICCV 2009, Kyoto.

run the fake_example.m 
The code generates a simple training set where pos examples are points
inside a circle and negative are outside. Additive classifiers are trained
using PWLSGD by encoding the data into 20 dimensions to achieve near perfect
seperation.

you may also train phi2 encoded linear models 
by first downloading LIBLINEAR
http://www.csie.ntu.edu.tw/~cjlin/liblinear/

and adding the path to train/predict functions
(set the useLIBLINEAR flag to 1 in demo.m)

encode_data.m : contains code to do the encoding.
--
Subhransu Maji 
www.cs.berkeley.edu/~smaji

Alexander C. Berg, 
www.cs.berkeley.edu/~aberg


