Multi-beam 解析过程

package com.multitrigger;

import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.List;
import java.util.Map;

import com.triggerInfo.InputInfo;
import com.triggerInfo.OutputInfo;
import com.triggerInfo.Protein;
import com.triggerInfo.Segment;
import com.triggerfeature.*;

public class MultiTriggerParsing 
{	
	State state = new State();  	
	CorpusProcessing corpus = new CorpusProcessing();
	
	public void beamSearch(Model model, InputInfo sentInputInfo, OutputInfo sentOutputInfo) 
	{	
		List<List<State>> beamList = new ArrayList<List<State>>();      
		List<String> tokenList = sentInputInfo.sentToken;
		//"Anaphora","Entity","Ubiquitination","Protein_modification"
		List<String> tag = Arrays.asList("Gene_expression","Transcription","Protein_catabolism","Phosphorylation","Localization","Binding","Regulation","Positive_regulation","Negative_regulation","None");
			
	    for(int endTokenId = 0; endTokenId < tokenList.size(); endTokenId++)          
		{	
			List<State> endTokenStateList = new ArrayList<State>();        //其实就是保存在Beam Search中的候选值,将K个state添加到Agenda(可能的生成结果)			
						 
			outer:for(String triggerType : tag)
			{
				int maxSegLength = 5;
				 
				if(endTokenId + 1 > maxSegLength)  //此处把 >= 换成了 > ,因为等于时， List<State> candiList为空，仍然要初始化一个State实例
				{					 
					for(int segLength = 1; segLength <= maxSegLength; segLength++)
					{
						List<State> candiList = beamList.get(endTokenId - segLength);  //此处不为空
								 
						int startTokenId = endTokenId - segLength + 1;
						 
						//判断是否是蛋白质,如果为蛋白质则不需要判断是不是trigger
						//*******************************************************
						for(int proteinId = 0; proteinId < sentInputInfo.proteinList.size(); proteinId++)
						{
							Protein protein = sentInputInfo.proteinList.get(proteinId);
							int proStartIndex = protein.proteinStart;
							int proEndIndex = protein.proteinEnd;
							
							if(startTokenId == proStartIndex && endTokenId == proEndIndex)
							{
								//说明是蛋白质，此时执行一些程序之后跳出
														
								for(int stateId = 0; stateId < candiList.size(); stateId++)
							    {
									State oldState = candiList.get(stateId);
									 
									endTokenStateList.add(state.append(startTokenId, endTokenId, "None", oldState, model, sentInputInfo)); 
								}								
								break outer;
							}
						}
						//*********************************************************
						//以上代码用于判断是否是蛋白质，如果是蛋白质怎么样，如果不是怎么样？
						
						for(int stateId = 0; stateId < candiList.size(); stateId++)
					    {
							State oldState = candiList.get(stateId);
							 
							endTokenStateList.add(state.append(startTokenId, endTokenId, triggerType, oldState, model, sentInputInfo)); 
						}						
					}
					
				}else
				{
                    maxSegLength = endTokenId + 1;
					 
					for(int segLength = 1; segLength <= maxSegLength; segLength++)
					{
						int startTokenId = endTokenId - segLength + 1;
						
						//********************************************************
						for(int proteinId = 0; proteinId < sentInputInfo.proteinList.size(); proteinId++)
						{
							Protein protein = sentInputInfo.proteinList.get(proteinId);
							int proStartId = protein.proteinStart;
							int proEndId = protein.proteinEnd;
							
							if(startTokenId == proStartId && endTokenId == proEndId)
							{
								//说明是蛋白质，此时执行一些程序之后跳出
								if(segLength == maxSegLength) //此时不能从beamList中取candiList：segLength == endTokenId + 1
								{							 
									State oldState = new State();
										
									endTokenStateList.add(state.append(startTokenId, endTokenId, "None", oldState, model, sentInputInfo));

								 }else
								 {
									 List<State> candiList = beamList.get(endTokenId - segLength); //注意此处的 candiList
									 
									 for(int stateId = 0; stateId < candiList.size(); stateId++)
									 {
										 State oldState = candiList.get(stateId);
											 
										 endTokenStateList.add(state.append(startTokenId, endTokenId, "None", oldState, model, sentInputInfo)); 
									 }							
								 }	
								 break outer;
							}
						}	
						//********************************************************						
						
						if(segLength == maxSegLength) //此时不能从beamList中取candiList：segLength == endTokenId + 1
						{							 
							State oldState = new State();
								
							endTokenStateList.add(state.append(startTokenId, endTokenId, triggerType, oldState, model, sentInputInfo));

						 }else
						 {
							 List<State> candiList = beamList.get(endTokenId - segLength); //注意此处的 candiList
							 
							 for(int stateId = 0; stateId < candiList.size(); stateId++)
							 {
								 State oldState = candiList.get(stateId);
									 
								 endTokenStateList.add(state.append(startTokenId, endTokenId, triggerType, oldState, model, sentInputInfo)); 
							 }							
						 }						 
					 }
				 }							 
			 }
			 
		     Collections.sort(endTokenStateList,new Comparator<State>(){ public int compare(State sta1, State sta2) {  return new Double(sta2.score).compareTo(new Double(sta1.score));} });
		     
		     corpus.containK(endTokenStateList, 20);
		     
		     beamList.add(endTokenStateList);
		    
	         //判断endTokenId 是否是 goldSegment的结尾词，如果不是，不更新，否则更新
	 		 boolean mark = false;
	 		 
	 		 List<Segment> goldSegList = sentOutputInfo.segmentList;
	 		 
	 		 List<Segment> partGoldList = new ArrayList<Segment>();
	 		
	 		 for(int segId = 0; segId < goldSegList.size(); segId++)
	         {
	        	 Segment seg = goldSegList.get(segId);
	        	 
	        	 int segEnd = seg.segmentEnd;
	        	 
	        	 if(segEnd == endTokenId)//此时更新权重
		 		 {
		 			 mark = true;
		 			 
		 			 partGoldList = corpus.getPartGoldList(goldSegList, segId); 
		 			 		 			
		 			 break;
		 		 }
	         }//循环结束的有问题
	 		 //***********************************************************
		     if(!corpus.isContain(endTokenStateList, partGoldList) && mark == true)
		     {		    	 
		    	 List<Map<String, Double>> goldFeature = sentOutputInfo.sentGoldFeature;//一句话的所有token的goldFeature 都存储在List<Map>中
		    			    	 
		    	 Map<String, Double> endTokenGoldFeature = goldFeature.get(endTokenId); //endTokenId对不对？
		    	 
		    	 List<State> staList = beamList.get(beamList.size() - 1); // 检查此处有否合理
				 
				 State bestCandidate = staList.get(0);
				 
				 Map<String, Double> bestSegmentPreFeature = bestCandidate.segmentPreFeatures;
				 
				 //权重更新
				 Map<String, Double> result = corpus.cutWeight(endTokenGoldFeature, bestSegmentPreFeature);  
			    	
				 model.triggerWeight = corpus.addWeight(model.triggerWeight, result); 
				 
				 break;
			 }	     
		}
    }
}