扫二维码与项目经理沟通
我们在微信上24小时期待你的声音
解答本文疑问/技术咨询/运营咨询/技术建议/互联网交流
基于UKKonen实现后缀树总结以及代码怎么写,很多新手对此不是很清楚,为了帮助大家解决这个难题,下面小编将为大家详细讲解,有这方面需求的人可以来学习下,希望你能有所收获。
网站建设哪家好,找成都创新互联公司!专注于网页设计、网站建设、微信开发、微信小程序定制开发、集团企业网站建设等服务项目。为回馈新老客户创新互联还提供了新乐免费建站欢迎大家使用!
几年前曾实现过一个菜鸟版的SuffixTree。最近要用到后缀树处理些问题,认真实现了一个,主要是基于UKKonen的On-Line算法。稍微总结下。
网上关于后缀树介绍的文章有几篇写的挺好的,我就不再费力去做重复工作了。这个只是我的个人总结帖,所以定位是给看了后缀树的简介,知道什么后缀树,然后看了UKKonen的加速文章,有点迷迷糊糊的同学的一个总结帖。
正宗的Paper应该是Ukkonen的下面这一篇paper。
(1) E. Ukkonen, On-Line Construction ofSuffix Trees, Algorithmica, 14 (1995), 249-260
但是我看了,对我来说真心有点难懂啊,然后Gusfield后来写了不知道书还是Paper的下面一个文章,讲的就通俗易懂多了,想学习后缀树OnLine算法的话,强烈推荐看下面的Guesfield的文章。
(2) Gusfield, Dan (1999) [1997].Algorithms on Strings, Trees and Sequences: Computer Science and ComputationalBiology. USA: Cambridge University Press.
上面两篇Paper都是英文,想看的同学Google下即可。
大部分的同学一看后缀树都明白是什么回事了,但是一看到UKKonen的算法,三个加速大段大段的描述后,就晕掉了。
我尝试忽略证明,简单并不严谨地总结下UKKonen算法中(没看过Guesfield或相关文章的同学,我只能表示对不住了),关键的三个加速:
(1) SuffixLink : 各种理论证明起来有点小复杂,但是道理用处说白了很简单。 因为当我们将s[i+1] 加到子串s[j-1…i]后,下一步我们就想将s[i+1]加到s[j….i]后面。正常来说我们就从根节点遍历s[j….i]呗,但是这个花时间啊,所以我们为什么不从s[j-1…..i]直接就跳到s[j….i]呢,而不要每次都从根节点遍历下来。所以所谓的SuffixLink,对s[j-1....i]来说,就是s[j…..i]的地址。
(2) 在Ukkonen算法中,叶节点总是叶节点(这个加速认真看下Gusfield的文章一看就懂,这里只是总结,就不深入去讲),所以每次遍历只需从最后一个叶节点开始。
(3) 但发现s[j…i+1]已经在后缀树中,那么s[j+1….i+1]这些后缀肯定也在后缀树中了,所以就不需要再遍历。
Ukkoen的后缀树我觉得最难得还是代码实现。网上代码比较少,特来分享下。我这个肯定不是最快的,不过应该是后缀树注释最多之一的一份代码了,而且代码结构和Guesfield那文章的整体描述比较接近。然后为了方便入门,这个只实现了加速1,慢慢一个个的来。大家有兴趣的,稍微修改下加速2和加速3就来了。然后有错误,也麻烦大家指正,我做了不少测试了,结果都正确,但是暂时不敢100%包票。
头文件:
#pragma once #include#include using namespace std; class SuffixNode { public: vector m_pSons; SuffixNode* m_pFarther; SuffixNode* m_pSuffixLink; int m_iPathPosition; int m_iEdgeStart; int m_iEdgeEnd; }; class SuffixTree { public: //int m_iE;//The virtual end of all leaves. SuffixNode* m_pRoot;//The root of the tree. string m_czTreeStr;//the string that the tree represent. }; //Means a sub string of the suffix tree (string[beging],string[end]). class TreePath { public: int m_iBegin; int m_iEnd; }; //Represent the char in a node's incoming edge. class TreePos { public: TreePos() { m_iEdgePos = 0; m_pNode = NULL; } TreePos(int edgePos,SuffixNode* pNode) { m_iEdgePos = edgePos; m_pNode = pNode; } int m_iEdgePos;//The ith char of the incoming edge. SuffixNode* m_pNode;//The node we are going to search. }; //=====================================Class Declarations============================== void SingleCharExtesion(SuffixTree* pTree,TreePos* pPos ,TreePath extendStrPath,int* firstExtensionFlag); /* Add s[0....i+1],s[1...i+1].... to the suffix tree Input: SuffixNode* pNode : When we only use trick 1,pNode is the pointer to the longest leaf,s[0........i]. phase : Equals i+1 in the paper. */ void SinglePhaseExtend(SuffixTree* pTree,TreePos pPos,int phase); SuffixNode* CreateTreeNode(SuffixNode* pFarther,int iedgeStart,int iedgeEnd); /* FollowSuffixLink : Follows the suffix link of the source node according to Ukkonen's rules(jump from s[j-1...i] to s[j....i]). Input : The tree, and node. The node is the last internal node we visited. Output: The destination node that represents the longest suffix of node's path. Example: if node represents the path "abcde" then it returns the node that represents "bcde". */ void FollowSuffixLink(SuffixTree* pTree,TreePos* pPos, TreePath strji); int GetNodeLabelLength(SuffixTree* pTree, SuffixNode* pNode); int GetNodeLabelEnd(SuffixTree* pTree,SuffixNode* pNode); /* Find the son node which starts with the char,ch. */ SuffixNode* Find_Son(SuffixTree* pTree,SuffixNode* pFarNode, char ch); bool IsTheLastCharInEdge(SuffixTree* pTree, SuffixNode* pNode, int edge_pos); SuffixNode* ApplyExtensionRule2(SuffixNode* pNode,int edgeLabelBeg,int edgeLabelEnd,int edgePos,bool newLeafFlag); /* Trace the sub string(TreePath str) from the node(SuffixNode* pNode). Input: int* edgePos : where the last char is found at that edge int* charsFound : how many chars of str have been found. bool skipFlag : Use skip trick or not. */ SuffixNode* TraceString(SuffixTree* pTree,SuffixNode* pNode,TreePath str,int* edgePos,int* charsFound,bool skipFlag); /* Trace the substring(TreePath strPath) in one single edge out of pNode. */ SuffixNode* TraceSingleEdge(SuffixTree* pTree,SuffixNode* pNode,TreePath strPath,int* charsFound,int* edgePos,bool* searchDone,bool skipFlag); SuffixNode* CreateFirstCharacter(SuffixTree* pTree);//Add the first character to the suffix tree. SuffixTree* CreateSuffixTree(string tStr); /* For Debug: See if the sub string (from root to pPos) equals pTree->string[subPath.m_iBegin,subPath.m_iEnd] */ bool TestPosSubStringEqualPath(SuffixTree* pTree,TreePos *pPos, TreePath subPath); bool FindSubString(SuffixTree* pTree,string subStr); //=====================================================================================
在看具体的实现前,先看看如何调用我这个后缀树的类吧,最简单的应用,查找某子串是否在母串中:
string str="MISSISSIPPI"; string subStr="PP"; SuffixTree* pTree = CreateSuffixTree(str); bool existFlag = FindSubString(pTree,subStr);
最后来看具体的实现:
#pragma once #include "SuffixTree.h" #includeusing namespace std; SuffixNode* pNodeNoSuffixLink=NULL; //=====================================Class Definitions============================== /* Trace the substring(TreePath strPath) in one single edge going out of pNode. Input: int* edgeCharsFound : how many characters we find matched in the outgoing edge of pNode. */ SuffixNode* TraceSingleEdge(SuffixTree* pTree,SuffixNode* pNode,TreePath strPath,int* edgeCharsFound,int* edgePos,bool* searchDone,bool skipFlag) { //Find outgoing edge of pNode with our first character. SuffixNode* nextNode = Find_Son(pTree,pNode,pTree->m_czTreeStr[strPath.m_iBegin]); *searchDone = true; if(nextNode == NULL) {//There is no match in pNode's sons,so we can only return to pNode. *edgePos = GetNodeLabelLength(pTree,pNode); *edgeCharsFound = 0; return pNode; } int edgeLen = GetNodeLabelLength(pTree,nextNode); int strLen = strPath.m_iEnd - strPath.m_iBegin + 1; if(skipFlag == true)//Use the trick1 : skip { if(edgeLen < strLen) { *searchDone = false; *edgeCharsFound = edgeLen; *edgePos = edgeLen - 1; } else if(edgeLen == strLen) { *edgeCharsFound = edgeLen; *edgePos = edgeLen - 1; } else { *edgeCharsFound = strLen; *edgePos = strLen - 1; } return nextNode; } else//No skip,match each char one after another { *edgePos = 0; *edgeCharsFound = 0; //Find out the min length if(strLen < edgeLen) edgeLen = strLen; for(*edgeCharsFound=1,*edgePos=1;(*edgePos) m_czTreeStr[ nextNode->m_iEdgeStart + *edgePos ] != pTree->m_czTreeStr[strPath.m_iBegin + *edgePos ]) { (*edgePos)--; return nextNode; } } } //When it comes here, (*edgePos) is one more; (*edgePos)--; if(*edgeCharsFound < strLen) { *searchDone = false; } return nextNode; } /* Trace the sub string(TreePath str) from the node(SuffixNode* pNode). Input: int* edgePos :For output , where the last char is found at that edge int* charsFound : How many chars of str have been found. bool skipFlag : Use skip trick or not. */ SuffixNode* TraceString(SuffixTree* pTree,SuffixNode* pNode,TreePath str,int* edgePos,int* charsFound,bool skipFlag) { bool searchDone=false; *charsFound = 0; *edgePos=0 ; int edgeCharsFound=0; while(searchDone==false) { edgeCharsFound = 0; *edgePos=0; pNode = TraceSingleEdge(pTree,pNode,str,&edgeCharsFound,edgePos,&searchDone,skipFlag); str.m_iBegin += edgeCharsFound; *charsFound += edgeCharsFound; } if(*charsFound == 0) return NULL; return pNode; } /* Input: (1) pNode : the node who is going to add a new son or whose edge is going to be split. (2) edgeLabelBeg : when newleafFlag==true,it's the edge begin label of the new leaf. when when newleafFlag==false, it's the edge begin label of the new new leaf( the leaf of s[i+1], not s[i]). (3) like above : just the end (4 )int edgePos : where split is done to pNode if newLeafFlag==false (the 0th position or 1th position or...) */ SuffixNode* ApplyExtensionRule2(SuffixNode* pNode,int edgeLabelBeg,int edgeLabelEnd,int edgePos,bool newLeafFlag) { if(newLeafFlag==true) { //Add an new leaf SuffixNode* newLeaf = CreateTreeNode(pNode,edgeLabelBeg,edgeLabelEnd); return newLeaf; } else { //Add an new internal node and an new leaf //First create the new internal node. SuffixNode* nInternalNode = CreateTreeNode(pNode->m_pFarther,pNode->m_iEdgeStart,pNode->m_iEdgeStart + edgePos); //Remove pNode from its farther's sons for(vector ::iterator pNodeIter=pNode->m_pFarther->m_pSons.begin(); pNodeIter!=pNode->m_pFarther->m_pSons.end();pNodeIter++) { if( pNode == *pNodeIter ) { pNode->m_pFarther->m_pSons.erase(pNodeIter); break; } } //Adjust pNode's information. pNode->m_iEdgeStart += (edgePos + 1); pNode->m_pFarther = nInternalNode; nInternalNode->m_pSons.push_back(pNode); //Create the new leaf for s[i+1] SuffixNode* nLeafNode = CreateTreeNode(nInternalNode,edgeLabelBeg,edgeLabelEnd); return nInternalNode; } } bool IsTheLastCharInEdge(SuffixTree* pTree, SuffixNode* pNode, int edge_pos) { if( edge_pos == GetNodeLabelLength(pTree,pNode) - 1 ) return true; return false; } int GetNodeLabelEnd(SuffixTree* pTree,SuffixNode* pNode) { //if(pNode->m_pSons.size() == NULL) //{ // return pTree->m_iE; //} return pNode->m_iEdgeEnd; } int GetNodeLabelLength(SuffixTree* pTree, SuffixNode* pNode) { int length = GetNodeLabelEnd(pTree,pNode) - pNode->m_iEdgeStart + 1; return length; } SuffixNode* CreateTreeNode(SuffixNode* pFarther,int iedgeStart,int iedgeEnd) { SuffixNode* pNode=new SuffixNode(); pNode->m_iEdgeStart = iedgeStart; pNode->m_iEdgeEnd = iedgeEnd; pNode->m_pFarther = pFarther; pNode->m_pSuffixLink = NULL; if(pFarther!=NULL) pFarther->m_pSons.push_back(pNode); return pNode; } //Find the son node which starts with the ch SuffixNode* Find_Son(SuffixTree* pTree,SuffixNode* pFarNode, char ch) { for(vector ::iterator nodeIter=pFarNode->m_pSons.begin(); nodeIter!=pFarNode->m_pSons.end();nodeIter++) { if(pTree->m_czTreeStr[(*nodeIter) -> m_iEdgeStart] == ch ) { return *nodeIter; } } return NULL; } /* FollowSuffixLink : Follows the suffix link of the source node according to Ukkonen's rules(jump from s[j-1...i] to s[j....i]). Input : The tree, and node. The node is the last internal node we visited. Output: The destination node that represents the longest suffix of node's path. Example: if node represents the path "abcde" then it returns the node that represents "bcde". */ void FollowSuffixLink(SuffixTree* pTree,TreePos * pPos,TreePath strji) { if(strji.m_iEnd < strji.m_iBegin)//Empty string,then we return to root. { pPos->m_iEdgePos=0; pPos->m_pNode = pTree->m_pRoot; return; } /*gama : the string(r in Gusfield's paper) between node and its father. If the node doesn't have suffix link , we need to go up to its farther*/ TreePath gama; if(pPos->m_pNode == pTree->m_pRoot) { int charsFound=0; pPos->m_pNode = TraceString(pTree,pTree->m_pRoot,strji,&pPos->m_iEdgePos,&charsFound,false); if(pPos->m_pNode == NULL) { pPos->m_iEdgePos = 0; pPos->m_pNode =pTree->m_pRoot; if(strji.m_iBegin != strji.m_iEnd) { cout<<"There is s[j-1..i](not empty) doesn't exist!"< m_pNode->m_pSuffixLink == NULL ) { if(pPos->m_pNode->m_pFarther == pTree->m_pRoot) {//its farther is the root pPos->m_pNode = pTree->m_pRoot; int charsFound=0; pPos->m_pNode = TraceString(pTree,pTree->m_pRoot,strji,&pPos->m_iEdgePos,&charsFound,false); if(pPos->m_pNode == NULL) { pPos->m_iEdgePos = 0; pPos->m_pNode =pTree->m_pRoot; if(strji.m_iBegin != strji.m_iEnd) { cout<<"There is s[j-1..i](not empty) doesn't exist!"< m_pNode->m_iEdgeStart; gama.m_iEnd = pPos->m_pNode->m_iEdgeStart + pPos->m_iEdgePos;// the end of s[j..i] //Follow farther's suffix link pPos->m_pNode = pPos->m_pNode->m_pFarther->m_pSuffixLink; //Down-walk gamma (until we found s[i],the character we add last extension) int charsFound=0; pPos->m_pNode = TraceString(pTree,pPos->m_pNode,gama,&pPos->m_iEdgePos,&charsFound,true); //////////////////////////////////////////////// } } else { //A suffix link exists - just follow it. pPos->m_pNode = pPos->m_pNode->m_pSuffixLink; pPos->m_iEdgePos = GetNodeLabelLength(pTree,pPos->m_pNode) - 1; //The last char of pPos's suffix link represents s[i] (the character we add last extension). } return; } /* For Debug: See if the sub string (from root to pPos) equals pTree->string[subPath.m_iBegin,subPath.m_iEnd] */ bool TestPosSubStringEqualPath(SuffixTree* pTree,TreePos *pPos, TreePath subPath) { if(pTree->m_pRoot == pPos->m_pNode && subPath.m_iBegin == subPath.m_iEnd) { return true; } int strRevIndex = subPath.m_iEnd; SuffixNode* tmpNode = pPos->m_pNode; int edgeRevIndex = tmpNode->m_iEdgeStart + pPos->m_iEdgePos; while(tmpNode != pTree->m_pRoot) { while( edgeRevIndex >= tmpNode->m_iEdgeStart && strRevIndex >= subPath.m_iBegin) { if( pTree->m_czTreeStr[edgeRevIndex] != pTree->m_czTreeStr[strRevIndex] ) { return false; } edgeRevIndex--; strRevIndex--; } tmpNode = tmpNode->m_pFarther; edgeRevIndex = tmpNode->m_iEdgeEnd; } if(strRevIndex != subPath.m_iBegin-1) return false; return true; } /* Input: (1) SuffixTree* pTree : The suffix tree (2) TreePos* pPos : The last internal node we visited , then we are going to jump to its suffix link in this extension. (3) TreePath extendStrPath : The suffix (s[j...i+1]) we are goint to add to the tree. */ void SingleCharExtesion(SuffixTree* pTree,TreePos* pPos ,TreePath extendStrPath,int* firstExtend) { TreePath sji; sji.m_iBegin = extendStrPath.m_iBegin; sji.m_iEnd = extendStrPath.m_iEnd - 1; if(*firstExtend != -1) { //Ready to jump from suffix link at or above s[j-1...i] that either has a suffix link (to s[j-1...i]) or is the root. FollowSuffixLink(pTree,pPos,sji); } *firstExtend = 1; //////////////////////////////////////////For Debug////////////////////////////////////////////////// if(sji.m_iEnd >= sji.m_iBegin) { if(TestPosSubStringEqualPath(pTree,pPos, sji) == false) { cout<<"FollowSuffixLink doesn't go to the right s[j..i]:"< m_pNode,pPos->m_iEdgePos)) { SuffixNode* pTmp = Find_Son(pTree,pPos->m_pNode,pTree->m_czTreeStr[extendStrPath.m_iEnd]); if(pTmp != 0) { //s[i+1] exits already. chars_found = 1; } } //Else see if can find extendStrPath.m_iEnd in the current edge else { if( pTree->m_czTreeStr[ pPos->m_pNode->m_iEdgeStart + pPos->m_iEdgePos + 1] == pTree->m_czTreeStr[extendStrPath.m_iEnd] )//Notice that " + 1 " means the next char of s[j...i] : yes, s[i+1] {//s[i+1] exits already. chars_found = 1; } } } //If s[i+1] was found - rule 3 applies if(chars_found == 1) { /* If there is an internal node that has no suffix link yet (only one may exist) - create a suffix link from it to the father - node of the current position in the tree*/ if(pNodeNoSuffixLink != NULL) { if(pPos->m_pNode->m_pSons.size() != 0) { pNodeNoSuffixLink->m_pSuffixLink = pPos->m_pNode; pNodeNoSuffixLink=NULL; } } //if(pPos->m_pNode->m_pSons.size()==0) // *ruleApplied = 1; //else // *ruleApplied = 3; return; } /*Since rule3 doesn't fit ( that s[j...i+1] is not in the tree), we are going to see rule2 and rule1. */ /* If last char s[j...i] found is the last char of an edge - create an new leaf ,apply rule2(add a new leaf) or rule1 */ if(IsTheLastCharInEdge(pTree,pPos->m_pNode,pPos->m_iEdgePos) || pPos->m_pNode==pTree->m_pRoot) { if(pPos->m_pNode->m_pSons.size() != 0) { //Internal node or root,apply rule2 that add a new leaf ApplyExtensionRule2(pPos->m_pNode, extendStrPath.m_iEnd, extendStrPath.m_iEnd, 0, true); //Suffix Link if(pNodeNoSuffixLink != NULL) { pNodeNoSuffixLink->m_pSuffixLink = pPos->m_pNode; pNodeNoSuffixLink = NULL; } /**ruleApplied = 2;*/ } //else it's a leaf, We do nothing. else { pPos->m_pNode->m_iEdgeEnd++; /**ruleApplied = 1;*/ } } //Else apply rule2 that adds an new intern node else { SuffixNode* nInternalNode = ApplyExtensionRule2(pPos->m_pNode,extendStrPath.m_iEnd,extendStrPath.m_iEnd,pPos->m_iEdgePos,false); if(pNodeNoSuffixLink != NULL) { pNodeNoSuffixLink->m_pSuffixLink = nInternalNode; pNodeNoSuffixLink = NULL; } //See the new internal node's suffix link. if( GetNodeLabelLength(pTree,nInternalNode)==1 && nInternalNode->m_pFarther == pTree->m_pRoot) { nInternalNode->m_pSuffixLink = pTree->m_pRoot; pNodeNoSuffixLink = NULL; } else { pNodeNoSuffixLink = nInternalNode; } //Adjust the node for the next extension pPos->m_pNode = nInternalNode; //*ruleApplied = 2; } } /* Add s[0....i+1],s[1...i+1].... to the suffix tree Input: SuffixNode* pNode: When we only use trick 1,pNode is the pointer to the longest leaf,s[0........i]. */ void SinglePhaseExtend(SuffixTree* pTree,TreePos pPos,int phase) { int iExtension=0; //pTree->m_iE = phase-1; int ruleApplied=-1; while(iExtension <= phase ) { TreePath extendPath; extendPath.m_iBegin=iExtension; extendPath.m_iEnd=phase; SingleCharExtesion(pTree,&pPos,extendPath,&ruleApplied); iExtension++; } return; } SuffixNode* CreateFirstCharacter(SuffixTree* pTree) { SuffixNode* firstLeaf = CreateTreeNode(pTree->m_pRoot,0,0); return firstLeaf; } SuffixTree* CreateSuffixTree(string tStr) { SuffixTree* psTree=new SuffixTree(); psTree->m_czTreeStr = tStr+"$"; psTree->m_pRoot = CreateTreeNode(NULL,0,0); //Add the first char into it. SuffixNode* firstLeaf = CreateFirstCharacter(psTree); TreePos* firstLeafPos = new TreePos(0,firstLeaf); for(int phase = 1 ; phase m_czTreeStr.length() ; phase++) { firstLeafPos->m_iEdgePos = firstLeafPos->m_pNode->m_iEdgeEnd - firstLeafPos->m_pNode->m_iEdgeStart; //start from s[j..i] SinglePhaseExtend(psTree,*firstLeafPos,phase); } return psTree; } bool FindSubString(SuffixTree* pTree,string subStr) { SuffixNode* node = Find_Son(pTree,pTree->m_pRoot,subStr[0]); if(node == NULL) { return false; } int startIndex = node->m_iEdgeStart; int strIndex=0; int edgeIndex; while(node != NULL) { edgeIndex = node->m_iEdgeStart; int edgeLabelEnd = node->m_iEdgeEnd;//GetNodeLabelEnd(pTree,node); while(strIndex < subStr.length() && edgeIndex <= edgeLabelEnd && pTree->m_czTreeStr[edgeIndex] == subStr[strIndex]) { strIndex++; edgeIndex++; } if(strIndex == subStr.length()) { //we found it return true; } else if(edgeIndex > node->m_iEdgeEnd) { node = Find_Son(pTree,node,subStr[strIndex]); } else { return false; } } return false; }
看完上述内容是否对您有帮助呢?如果还想对相关知识有进一步的了解或阅读更多相关文章,请关注创新互联行业资讯频道,感谢您对创新互联的支持。
我们在微信上24小时期待你的声音
解答本文疑问/技术咨询/运营咨询/技术建议/互联网交流