v2.1 改进分类器，更新模型

Author: xpqiu

.classpath

@@ -1,6 +1,5 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <classpath>
-	<classpathentry kind="src" path="src/main/java"/>
 	<classpathentry kind="con" path="org.eclipse.m2e.MAVEN2_CLASSPATH_CONTAINER">
 		<attributes>
 			<attribute name="maven.pomderived" value="true"/>

fnlp-core/src/main/java/org/fnlp/data/reader/SequenceReader.java

@@ -99,18 +99,19 @@ private Instance readSequence() {
 		cur = null;
 		try {
 			ArrayList<ArrayList<String>> seq = new ArrayList<ArrayList<String>>();
-			ArrayList<String> first = new ArrayList(); //至少有一列元素
-			seq.add(first);
+			ArrayList<String> firstColumnList = new ArrayList(); //至少有一列元素
+			seq.add(firstColumnList);
 			ArrayList<String> labels = null;
 			if(hasTarget){
 				labels = new ArrayList<String>();
 			}
 			String content = null;
+			
 			while ((content = reader.readLine()) != null) {
 				lineNo++;
-//				content = content.trim();
+				content = content.trim();
 				if (content.matches("^$")){
-					if(first.size()>0) //第一列个数>0
+					if(firstColumnList.size()>0) //第一列个数>0
 						break;
 					else
 						continue;
@@ -140,10 +141,17 @@ private Instance readSequence() {
 				}else{
 					ensure(colsnum,seq);
 					seq.get(colsnum).add(content.substring(start));
-				}					
+				}	
+				//debug
+//				if(colsnum>2){
+//					System.out.println(content);
+//				}
 			}
-			if (first.size() > 0){
+			
+			if (firstColumnList.size() > 0){
 				cur = new Instance(seq, labels);
+				//debug
+//				cur.setSource(firstColumnList.toString());
 			}
 			seq = null;
 			labels = null;

fnlp-core/src/main/java/org/fnlp/ml/classifier/linear/OnlineTrainer.java

@@ -53,8 +53,6 @@ public class OnlineTrainer extends AbstractTrainer {
 	 */
 	public static float eps = 1e-10f;
 
-	public TrainMethod method = TrainMethod.FastAverage;
-
 	public boolean DEBUG = false;
 	public boolean shuffle = true;
 	public boolean finalOptimized = false;
@@ -75,9 +73,6 @@ public class OnlineTrainer extends AbstractTrainer {
 	public int iternum;
 	protected float[] weights;
 
-	public enum TrainMethod {
-		Perceptron, Average, FastAverage
-	}
 	public OnlineTrainer(AlphabetFactory af, int iternum) {
 		//默认特征生成器
 		Generator gen = new SFGenerator();
@@ -167,61 +162,60 @@ public Linear train(InstanceSet trainset, InstanceSet devset) {
 		long beginTimeIter, endTimeIter;
 		int iter = 0;
 		int frac = numSamples / 10;
-
-		float[] averageWeights = null;
-		if (method == TrainMethod.Average || method == TrainMethod.FastAverage) {
-			averageWeights = new float[weights.length];
-		}
-
+
+		//平均化感知器需要减去的权重
+		float[] extraweight = null;
+		extraweight = new float[weights.length];
+		
+			
+		
 		beginTime = System.currentTimeMillis();
 
-		if (shuffle)
-			trainset.shuffle(random);
+		
+		
+		//遍历的总样本数
+		int k=0;
 
 		while (iter++ < iternum) {
 			if (!simpleOutput) {
 				System.out.print("iter "+iter+":  ");
-			}
+			}
+			
 			float err = 0;
 			float errtot = 0;
 			int cnt = 0;
 			int cnttot = 0;
-			int progress = frac;
+			int progress = frac;			
 
+			if (shuffle)
+				trainset.shuffle(random);
+			
 			beginTimeIter = System.currentTimeMillis();
-
-			float[] innerWeights = null;
-			if (method == TrainMethod.Average) {
-				innerWeights = Arrays.copyOf(weights, weights.length);
-			}
-
-			for (int ii = 0; ii < numSamples; ii++) {
+			for (int ii = 0; ii < numSamples; ii++) {
+				
+				k++;
 				Instance inst = trainset.getInstance(ii);
 				Predict pred = (Predict) inferencer.getBest(inst,2);				
 
 				float l = loss.calc(pred.getLabel(0), inst.getTarget());
 				if (l > 0) {
 					err += l;
 					errtot++;
-					update.update(inst, weights, pred.getLabel(0), c);
+					update.update(inst, weights, k, extraweight, pred.getLabel(0), c);
 
 				}else{
 					if (pred.size() > 1)
-						update.update(inst, weights, pred.getLabel(1), c);
+						update.update(inst, weights, k, extraweight, pred.getLabel(1), c);
 				}
 				cnt += inst.length();
-				cnttot++;
-				if (method == TrainMethod.Average) {
-					for (int i = 0; i < weights.length; i++) {
-						innerWeights[i] += weights[i];
-					}
-				}
+				cnttot++;				
 
 				if (!simpleOutput && progress != 0 && ii % progress == 0) {
 					System.out.print('.');
 					progress += frac;
-				}
-			}
+				}				
+				
+			}//end for
 
 			float curErrRate = err / cnt;
 
@@ -253,17 +247,7 @@ public Linear train(InstanceSet trainset, InstanceSet devset) {
 			if (devset != null) {
 				evaluate(devset);
 			}
-			System.out.println();
-
-			if (method == TrainMethod.Average) {
-				for (int i = 0; i < innerWeights.length; i++) {
-					averageWeights[i] += innerWeights[i] / numSamples;
-				}
-			} else if (method == TrainMethod.FastAverage) {
-				for (int i = 0; i < weights.length; i++) {
-					averageWeights[i] += weights[i];
-				}
-			}
+			System.out.println();			
 
 			if (interim) {
 				Linear p = new Linear(inferencer, trainset.getAlphabetFactory());
@@ -277,18 +261,15 @@ public Linear train(InstanceSet trainset, InstanceSet devset) {
 			if(MyArrays.viarance(hisErrRate) < eps){
 				System.out.println("convergence!");
 				break;	
-			}
-		}
-
-		if (method == TrainMethod.Average || method == TrainMethod.FastAverage) {
-			for (int i = 0; i < averageWeights.length; i++) {
-				averageWeights[i] /= iternum;
-			}
-			weights = null;
-			weights = averageWeights;
-			inferencer.setWeights(weights);
-		}
-
+			}
+			
+		}// end while 外循环
+		
+		//平均化参数
+		for (int i = 0; i < weights.length; i++) {
+			weights[i] -= extraweight[i]/k;
+		}
+		
 		System.out.print("Non-Zero Weight Numbers: " + MyArrays.countNoneZero(weights));
 		if (finalOptimized) {
 			int[] idx = MyArrays.getTop(weights.clone(), threshold, false);

fnlp-core/src/main/java/org/fnlp/ml/classifier/linear/update/AbstractPAUpdate.java

@@ -27,7 +27,6 @@
  * 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}.
- * @author Feng Ji
  *
  */
 public abstract class AbstractPAUpdate implements Update {
@@ -52,28 +51,14 @@ public AbstractPAUpdate(Loss loss) {
 		this.loss = loss;
 	}
 
-	/**
-	 * 参数更新方法
-	 * @param inst 样本实例
-	 * @param weights 权重
-	 * @param predict 预测答案
-	 * @param c 步长阈值
-	 * @return 预测答案和标准答案之间的损失
-	 */
-	public float update(Instance inst, float[] weights, Object predict, float c) {
-		return update(inst, weights, inst.getTarget(), predict, c);
-	}
+	@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);
+	}
+	
 
-	/**
-	 * 参数更新方法
-	 * @param inst 样本实例
-	 * @param weights 权重
-	 * @param target 对照答案
-	 * @param predict 预测答案
-	 * @param c 步长阈值
-	 * @return 预测答案和对照答案之间的损失
-	 */
-	public float update(Instance inst, float[] weights, Object target,
+	@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);
@@ -87,14 +72,17 @@ public float update(Instance inst, float[] weights, Object target,
 				alpha = alpha*inst.getWeight();
 			if(alpha>c){
 				alpha = c;
-			}else{
-				alpha=alpha;
-			}
+			}
+			
 			int[] idx = diffv.indices();
 
-			for (int i = 0; i < idx.length; i++) {
-				
-				weights[idx[i]] += diffv.get(idx[i]) * alpha;
+			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++) {				
+				
 			}
 		}
 
@@ -105,12 +93,12 @@ public float update(Instance inst, float[] weights, Object target,
 	}
 
 	/**
-	 * 计算预测答案和对照答案之间的距离
+	 * 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 对照答案
-	 * @param predict 预测答案
-	 * @return 预测答案和对照答案之间的距离
+	 * @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);

fnlp-core/src/main/java/org/fnlp/ml/classifier/linear/update/LinearMaxPAUpdate.java

@@ -24,8 +24,6 @@
 
 /**
  * 线性分类的参数更新类，采用PA算法
- * @author Feng Ji
- *
  */
 public class LinearMaxPAUpdate extends AbstractPAUpdate {
 
@@ -58,5 +56,4 @@ protected int diff(Instance inst, float[] weights, Object target,
 		return 1;
 	}
 
-
 }

fnlp-core/src/main/java/org/fnlp/ml/classifier/linear/update/Update.java

@@ -22,11 +22,32 @@
 import org.fnlp.ml.types.Instance;
 
 public interface Update {
-	
-	public float update(Instance inst, float[] weights, Object predictLabel,
-			float c);
-
-	public float update(Instance inst, float[] weights, Object predictLabel,
+	
+	/**
+	 * 
+	 * @param inst 样本实例
+	 * @param weights 权重
+	 * @param k 目前遍历的样本数
+	 * @param extraweight 平均化感知器需要减去的权重
+	 * @param predictLabel 预测类别
+	 * @param c 步长阈值
+	 * @return 预测类别和真实类别之间的损失
+	 */
+	public float update(Instance inst, float[] weights, int k, float[] extraweight, Object predictLabel,
+			 float c);
+	
+	/**
+	 * 
+	 * @param inst 样本实例
+	 * @param weights 权重
+	 * @param k 目前遍历的样本数
+	 * @param extraweight 平均化感知器需要减去的权重
+	 * @param predictLabel 预测类别
+	 * @param goldenLabel 真实类别
+	 * @param c 步长阈值
+	 * @return 预测类别和真实类别之间的损失
+	 */
+	public float update(Instance inst, float[] weights, int k, float[] extraweight, Object predictLabel,
 			Object goldenLabel, float c);
 
 }