package org.fnlp.demo.nlp.tc;
import gnu.trove.iterator.TIterator;
import java.io.File;
import java.sql.Time;
import org.fnlp.data.reader.FileReader;
import org.fnlp.data.reader.Reader;
import org.fnlp.ml.classifier.Predict;
import org.fnlp.ml.classifier.LabelParser.Type;
import org.fnlp.ml.classifier.bayes.BayesClassifier;
import org.fnlp.ml.classifier.bayes.BayesTrainer;
import org.fnlp.ml.classifier.bayes.ItemFrequency;
import org.fnlp.ml.classifier.knn.KNN;
import org.fnlp.ml.classifier.knn.KNNClassifier;
import org.fnlp.ml.classifier.linear.Linear;
import org.fnlp.ml.classifier.linear.OnlineTrainer;
import org.fnlp.ml.eval.Evaluation;
import org.fnlp.ml.feature.FeatureSelect;
import org.fnlp.ml.types.Instance;
import org.fnlp.ml.types.InstanceSet;
import org.fnlp.ml.types.alphabet.AlphabetFactory;
import org.fnlp.ml.types.alphabet.IFeatureAlphabet;
import org.fnlp.ml.types.alphabet.StringFeatureAlphabet;
import org.fnlp.nlp.cn.tag.CWSTagger;
import org.fnlp.nlp.pipe.NGram;
import org.fnlp.nlp.pipe.Pipe;
import org.fnlp.nlp.pipe.SeriesPipes;
import org.fnlp.nlp.pipe.StringArray2IndexArray;
import org.fnlp.nlp.pipe.StringArray2SV;
import org.fnlp.nlp.pipe.Target2Label;
import org.fnlp.nlp.pipe.nlp.CNPipe;
import org.fnlp.nlp.similarity.SparseVectorSimilarity;
public class TextClassificationBasedOnKNN {
/**
* 训练数据路径
*/
// private static String dataPath="C:/dataset/SogouC/";
// private static String trainDataPath = dataPath+"ClassFile/";
// private static String dataPath="C:/dataset/SogouC.mini/";
// private static String trainDataPath = dataPath+"Sample/";
//private static String dataPath="D:/Documents/dataset/SogouC.reduced/";
//private static String trainDataPath = dataPath+"Reduced/";
private static String dataPath="D:/Documents/dataset/SogouC.mini/";
private static String trainDataPath = dataPath+"Sample/";
/**
* 模型文件
*/
private static String knnModelFile = dataPath+"modelKnn.gz";
public static void main(String[] args) throws Exception {
//分词
Pipe removepp=new RemoveWords();
CWSTagger tag = new CWSTagger("../models/seg.m");
Pipe segpp=new CNPipe(tag);
Pipe s2spp=new Strings2StringArray();
//建立字典管理器
AlphabetFactory af = AlphabetFactory.buildFactory();
//使用n元特征
Pipe ngrampp = new NGram(new int[] {2,3});
//将字符特征转换成字典索引;
Pipe sparsepp=new StringArray2SV(af);
//将目标值对应的索引号作为类别
Pipe targetpp = new Target2Label(af.DefaultLabelAlphabet());
//建立pipe组合
SeriesPipes pp = new SeriesPipes(new Pipe[]{removepp,segpp,s2spp,targetpp,sparsepp});
/**
* Knn
*/
System.out.print("\nKnn\n");
System.out.print("\nReading data......\n");
long time_mark=System.currentTimeMillis();
InstanceSet instset = new InstanceSet(pp,af);
Reader reader = new MyDocumentReader(trainDataPath,"gbk");
instset.loadThruStagePipes(reader);
System.out.print("..Reading data complete "+(System.currentTimeMillis()-time_mark)+"(ms)\n");
//将数据集分为训练是和测试集
System.out.print("Sspliting....");
float percent = 0.9f;
InstanceSet[] splitsets = instset.split(percent);
InstanceSet trainset = splitsets[0];
InstanceSet testset = splitsets[1];
System.out.print("..Spliting complete!\n");
System.out.print("Training Knn...\n");
time_mark=System.currentTimeMillis();
SparseVectorSimilarity sim=new SparseVectorSimilarity();
pp.removeTargetPipe();
KNNClassifier knn=new KNNClassifier(trainset, pp, sim, af, 9);
af.setStopIncrement(true);
ItemFrequency tf=new ItemFrequency(trainset);
FeatureSelect fs=new FeatureSelect(tf.getFeatureSize());
long time_train=System.currentTimeMillis()-time_mark;
System.out.print("..Training compelte!\n");
System.out.print("Saving model...\n");
knn.saveTo(knnModelFile);
knn = null;
System.out.print("..Saving model compelte!\n");
System.out.print("Loading model...\n");
knn =KNNClassifier.loadFrom(knnModelFile);
System.out.print("..Loading model compelte!\n");
System.out.println("Testing Knn...\n");
int count=0;
fs.fS_CS(tf, 0.1f);
knn.setFs(fs);
for(int i=0;i<testset.size();i++){
Instance data = testset.getInstance(i);
Integer gold = (Integer) data.getTarget();
Predict<String> pres=(Predict<String>) knn.classify(data, Type.STRING, 3);
String pred_label=pres.getLabel();
String gold_label = knn.getLabel(gold);
if(pred_label.equals(gold_label)){
//System.out.println(pred_label+" : "+testsetknn.getInstance(i).getTempData());
count++;
}
else{
// System.err.println(gold_label+"->"+pred_label+" : "+testset.getInstance(i).getTempData());
// for(int j=0;j<3;j++)
// System.out.println(pres.getLabel(j)+":"+pres.getScore(j));
}
}
int knnCount=count;
System.out.println("..Testing Knn Complete");
System.out.println("Knn Precision:"+((float)knnCount/testset.size())+"("+knnCount+"/"+testset.size()+")");
knn.noFeatureSelection();
int flag=0;
long time_sum=0,time_times=0;
float[] percents_cs=new float[]{1.0f,0.9f,0.8f,0.7f,0.5f,0.3f,0.2f,0.1f};
int[] counts_cs=new int[10];
for(int test=0;test<percents_cs.length;test++){
long time_st=System.currentTimeMillis();
System.out.println("Testing Bayes"+percents_cs[test]+"...");
if(test!=0){
fs.fS_CS(tf, percents_cs[test]);
knn.setFs(fs);
}
count=0;
for(int i=0;i<testset.size();i++){
Instance data = testset.getInstance(i);
Integer gold = (Integer) data.getTarget();
Predict<String> pres=(Predict<String>)knn.classify(data, Type.STRING, 3);
String pred_label=pres.getLabel();
String gold_label = knn.getLabel(gold);
if(pred_label.equals(gold_label)){
count++;
}
else{
}
}
counts_cs[test]=count;
long time_ed=System.currentTimeMillis();
time_sum+=time_ed-time_st;
time_times++;
System.out.println("Knn Precision("+percents_cs[test]+"):"
+((float)count/testset.size())+"("+count+"/"+testset.size()+")"+" "+(time_ed-time_st)+"ms");
}
knn.noFeatureSelection();
float[] percents_csmax=new float[]{1.0f,0.9f,0.8f,0.7f,0.5f,0.3f,0.2f,0.1f};
int[] counts_csmax=new int[10];
for(int test=0;test<percents_csmax.length;test++){
long time_st=System.currentTimeMillis();
System.out.println("Testing Bayes"+percents_csmax[test]+"...");
if(test!=0){
fs.fS_CS_Max(tf, percents_cs[test]);
knn.setFs(fs);
}
count=0;
for(int i=0;i<testset.size();i++){
Instance data = testset.getInstance(i);
Integer gold = (Integer) data.getTarget();
Predict<String> pres=(Predict<String>)knn.classify(data, Type.STRING, 3);
String pred_label=pres.getLabel();
String gold_label = knn.getLabel(gold);
if(pred_label.equals(gold_label)){
count++;
}
else{
}
}
counts_csmax[test]=count;
long time_ed=System.currentTimeMillis();
time_sum+=time_ed-time_st;
time_times++;
System.out.println("Knn Precision("+percents_csmax[test]+"):"
+((float)count/testset.size())+"("+count+"/"+testset.size()+")"+" "+(time_ed-time_st)+"ms");
}
knn.noFeatureSelection();
float[] percents_ig=new float[]{1.0f,0.9f,0.8f,0.7f,0.5f,0.3f,0.2f,0.1f};
int[] counts_ig=new int[10];
for(int test=0;test<percents_ig.length;test++){
long time_st=System.currentTimeMillis();
System.out.println("Testing Bayes"+percents_ig[test]+"...");
if(test!=0){
fs.fS_IG(tf, percents_cs[test]);
knn.setFs(fs);
}
count=0;
for(int i=0;i<testset.size();i++){
Instance data = testset.getInstance(i);
Integer gold = (Integer) data.getTarget();
Predict<String> pres=(Predict<String>)knn.classify(data, Type.STRING, 3);
String pred_label=pres.getLabel();
String gold_label = knn.getLabel(gold);
if(pred_label.equals(gold_label)){
count++;
}
else{
}
}
counts_ig[test]=count;
long time_ed=System.currentTimeMillis();
time_sum+=time_ed-time_st;
time_times++;
System.out.println("Knn Precision("+percents_ig[test]+"):"
+((float)count/testset.size())+"("+count+"/"+testset.size()+")"+" "+(time_ed-time_st)+"ms");
}
System.out.println("..Testing Bayes complete!");
for(int i=0;i<percents_cs.length;i++)
System.out.println("Knn Precision CS("+percents_cs[i]+"):"
+((float)counts_cs[i]/testset.size())+"("+counts_cs[i]+"/"+testset.size()+")");
for(int i=0;i<percents_csmax.length;i++)
System.out.println("Knn Precision CS_Max("+percents_csmax[i]+"):"
+((float)counts_csmax[i]/testset.size())+"("+counts_csmax[i]+"/"+testset.size()+")");
for(int i=0;i<percents_ig.length;i++)
System.out.println("Knn Precision IG("+percents_ig[i]+"):"
+((float)counts_ig[i]/testset.size())+"("+counts_ig[i]+"/"+testset.size()+")");
System.out.println("\nTrain time: "+time_train+"(ms) for "
+trainset.size()+" train instances\n");
if(time_times>0)
System.out.println("Ave Test time: "+time_sum/time_times+"(ms) for "
+testset.size()+" test instances\n");
}
}