package edu.fudan.nlp.pipe;
import java.io.Serializable;
import java.util.Iterator;
import edu.fudan.ml.types.Instance;
import edu.fudan.ml.types.InstanceSet;
import edu.fudan.ml.types.sv.HashSparseVector;
import gnu.trove.iterator.TIntFloatIterator;
/**
* 由TF计算IDF
*
* @author xpqiu
*
*/
public class TF2IDF extends Pipe implements Serializable {
private static final long serialVersionUID = 5563900451276233502L;
public int[] idf;
public TF2IDF(InstanceSet train, InstanceSet test) {
int numFeatures = 0;
// 得到最大的特征维数
for (int i = 0; i < train.size(); i++) {
int len = ((HashSparseVector) train.getInstance(i).getData()).size();
if (len > numFeatures)
numFeatures = len;
}
for (int i = 0; i < test.size(); i++) {
int len = ((HashSparseVector) test.getInstance(i).getData()).size();
if (len > numFeatures)
numFeatures = len;
}
idf = new int[numFeatures + 1];
}
@Override
public void addThruPipe(Instance inst) {
HashSparseVector data = (HashSparseVector) inst.getData();
TIntFloatIterator it = data.data.iterator();
while (it.hasNext()) {
it.advance();
idf[it.key()]++;
}
}
}