/*
* Copyright (C) 2018 Edward Raff
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see .
*/
package jsat.outlier;
import java.util.Collections;
import java.util.List;
import java.util.Random;
import jsat.DataSet;
import jsat.classifiers.DataPoint;
import jsat.linear.DenseVector;
import jsat.linear.ScaledVector;
import jsat.linear.Vec;
import jsat.math.optimization.stochastic.AdaGrad;
import jsat.math.optimization.stochastic.GradientUpdater;
import jsat.utils.random.RandomUtil;
/**
* This class implements the One-Class SVM (OC-SVM) algorithm for outlier
* detection. This implementation works only in the primal or "linear" space. As
* such it works best when the data is sparse and high dimensional. If your data
* is dense and low dimensional, you may get better results by first applying a
* non-linear transformation to the data.
*
*
* See:
*
* - Schölkopf, B., Williamson, R., Smola, A., Shawe-Taylor, J., & Platt, J.
* (1999). Support Vector Method for Novelty Detection. In Advances in
* Neural Information Processing Systems 12 (pp. 582–588). Denver, CO.
* - Manevitz, L. M., & Yousef, M. (2001). One-class Svms for Document
* Classification. J. Mach. Learn. Res., 2, 139–154. Retrieved from
* http://dl.acm.org/citation.cfm?id=944790.944808
*
*
* @author Edward Raff
*/
public class LinearOCSVM implements Outlier
{
private Vec w;
private double p;
private int max_epochs = 100;
private double learningRate = 0.01;
private double v = 0.05;
public void setV(double v)
{
this.v = v;
}
public double getV()
{
return v;
}
@Override
public void fit(DataSet d, boolean parallel)
{
Random rand = RandomUtil.getRandom();
List X = d.getDataVectors();
int N = X.size();
w = new ScaledVector(new DenseVector(X.get(0).length()));
p = 0;
GradientUpdater gu = new AdaGrad();
gu.setup(w.length());
double cnt = 1/(v);
double prevLoss = Double.POSITIVE_INFINITY;
double curLoss = Double.POSITIVE_INFINITY;
for(int epoch = 0; epoch < max_epochs; epoch++)
{
Collections.shuffle(X, rand);
prevLoss = curLoss;
curLoss = 0;
for(int i = 0; i < X.size(); i++)
{
Vec x = X.get(i);
double loss = p - w.dot(x);
double p_delta = -1;
double x_mul = 0;
if(loss > 0)
{
p_delta += 1*cnt;
x_mul = -1*cnt;
}
curLoss += Math.max(0, loss);
p -= gu.update(w, new ScaledVector(x_mul, x), learningRate, p, p_delta);
w.mutableMultiply(1-learningRate);
}
// System.out.println("Epoch " + epoch + " " + curLoss + " " + (curLoss-prevLoss)/N);
if(Math.abs((curLoss-prevLoss)/N) < 1e-6*v)
break;//Convergence check
}
}
@Override
public double score(DataPoint x)
{
return w.dot(x.getNumericalValues())-p;
}
}
Ads help maintain this website.