/**
* This file is part of FNLP (formerly FudanNLP).
*
* FNLP is free software: you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* FNLP 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 Lesser General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with FudanNLP. If not, see <http://www.gnu.org/licenses/>.
*
* Copyright 2009-2014 www.fnlp.org. All rights reserved.
*/
package org.fnlp.ml.classifier.linear.update;
import org.fnlp.ml.loss.Loss;
import org.fnlp.ml.types.Instance;
import org.fnlp.ml.types.sv.HashSparseVector;
/**
* 抽象参数更新类，采用PA算法
* \mathbf{w_{t+1}} = \w_t + {\alpha^*(\Phi(x,y)- \Phi(x,\hat{y}))}.
* \alpha =\frac{1- \mathbf{w_t}^T \left(\Phi(x,y) - \Phi(x,\hat{y})\right)}{||\Phi(x,y) - \Phi(x,\hat{y})||^2}.
*
*/
public abstract class AbstractPAUpdate implements Update {
/**
* \mathbf{w_t}^T \left(\Phi(x,y) - \Phi(x,\hat{y})\right)
*/
protected float diffw;
/**
* \Phi(x,y)- \Phi(x,\hat{y})
*/
protected HashSparseVector diffv;
protected Loss loss;
/**
* 是否使用样本的权重进行加权
*/
public boolean useInstWeight;
public AbstractPAUpdate(Loss loss) {
diffw = 0;
diffv = new HashSparseVector();
this.loss = loss;
}
@Override
public float update(Instance inst, float[] weights, int k, float[] extraweight, Object predict, float c) {
return update(inst, weights, k, extraweight, inst.getTarget(), predict, c);
}
@Override
public float update(Instance inst, float[] weights, int k, float[] extraweight, Object target,
Object predict, float c) {
int lost = diff(inst, weights, target, predict);
if(lost==0)
return 0f;
float lamda = diffv.l2Norm2();
if (diffw <= lost) {
float alpha = (lost - diffw) / lamda;
if(useInstWeight)
alpha = alpha*inst.getWeight();
if(alpha>c){
alpha = c;
}
int[] idx = diffv.indices();
for (int i = 0; i < idx.length; i++) {
float t = diffv.get(idx[i]) * alpha;
weights[idx[i]] += t;
extraweight[idx[i]] += t *k;
}
for (int i = 0; i < idx.length; i++) {
}
}
diffv.clear();
diffw = 0;
return loss.calc(target, predict);
}
/**
* 计算预测类别和对照类别之间的距离
* @param inst 样本实例
* @param weights 权重
* @param target 对照类别
* @param predict 预测类别
* @return 预测类别和对照类别之间的距离
*/
protected abstract int diff(Instance inst, float[] weights, Object target,
Object predict);
protected void adjust(float[] weights, int ts, int ps) {
assert (ts != -1 && ps != -1);
diffv.put(ts, 1.0f);
diffv.put(ps, -1.0f);
diffw += weights[ts] - weights[ps];
}
}