前些日子研究了一下字符串匹配算法,突发奇想自己设计了一种新的字符串匹配算法,因为是基于BM的思想,所以暂且叫他BMY算法吧。传统的BM算法是基于坏字符规则和好后缀规则,从后向前的匹配字符串,每次发现失配时,会比较坏字符表和好后缀表,哪个对应的跳跃值大就用哪个跳跃值,这样的跳跃幅度比KMP算法要大很多。而我设计的BMY算法在失配后(失配字符称为尾1),会再比较一下失配字符前面那个字符(称为尾2):如果尾1和尾2都不在模式串中,直接跳跃模式串长度+2;如果尾1在模式串尾,则跳跃1;如果尾1和尾2的组合在模式串中,则跳跃模式串长度-尾1尾2在模式串中的位置;如果尾2在模式串头,则跳跃模式串长度+1;其他情况一律跳跃模式串长度+2。流程框图如下:
BMY算法设计维护了两张哈希表,其中哈希表m1用于验证匹配到的字符是否在模式串,哈希表m2存储【尾1+尾2】的双字符在模式串中的位置,具体的代码如下:
void BMY(string &pstr, const size_t &plen, string &sstr, const size_t &slen, int (&m1)[ALPHABET_SIZE], int (&m2)[ZuHe_SIZE], vector<int> &res)
{
int i,j; //i:模式串游标 j:文本串游标
i = j = plen-1;
int cnt = -1;
while(j < slen)
{
num_bmy++;
//模式串和文本串从后向前匹配
while( (i!=0) && pstr[i] == sstr[j])
{
--i;
--j;
bidui_bmy++;
}
bidui_bmy++;
//发现一个匹配的模式串
if(i==0 && pstr[i] == sstr[j])
{
res.push_back(j);
match_bmy++;
}
//匹配成功及出现失配的情况
j += (plen-i-1);
if(m1[sstr[j+1]] == 0 && m1[sstr[j+2]] == 0) //尾1和尾2都不在模式串中
j += (plen + 2);
else if(sstr[j+1] == pstr[plen-1]) //尾1在模式串尾
j += 1;
else if( (cnt = m2[sstr[j+1]*256+sstr[j+2]]) != -1) //尾1和尾2的组合在模式串中
j += (plen - cnt);
else if(sstr[j+2] == pstr[0]) //尾2在模式串头
j += (plen + 1);
else
j += (plen + 2);
cnt = -1;
i = plen - 1;
}
}
最后,比较KMP、BM、BMHS、BMY四种算法的匹配时间、匹配次数、比对次数,得到的结果如下:
从结果可以看出,自己设计的BMY算法和BM、BMHS相比时间性能上略差,但是已经比KMP强很多了。此外,BMY的匹配次数和比对次数明显少于其他算法,但在运行时间上却没有对应的优势,我猜想是BMY在选择跳跃幅度上做了太多选择比较耗费了时间。
下面贴出全部测试代码:
#include <iostream>
#include <cstring>
#include <sstream>
#include <vector>
#include <algorithm>
#include <unordered_map>
#include <map>
#include <ctime>
#include <iterator>
#include <fstream>
using namespace std;
#define ALPHABET_SIZE 32767
#define ZuHe_SIZE 32767
int match_kmp = 0;
int match_bmy = 0;
int match_bm = 0;
int match_bmhs = 0;
int num_kmp = 0;
int num_bmy = 0;
int num_bm = 0;
int num_bmhs = 0;
int bidui_kmp = 0;
int bidui_bm = 0;
int bidui_bmy = 0;
int bidui_bmhs = 0;
void InitBMHS(int *alphabet, string &des, size_t plen)
{
for (int i=0; i<ALPHABET_SIZE; i++)
alphabet[i] = plen;
for (int i=0; i<plen; i++)
alphabet[des[i]] = plen - i - 1;
}
void sunday(string &src, size_t len_s, string &des, size_t len_d, int *alphabet, vector<int> &res)
{
int i, pos = 0;
for (pos = 1; pos <= len_s - len_d + 1;)
{
num_bmhs++;
for (i=pos+len_d-2; i>=pos-1 ; i--)
{
if (src[i] == des[i-pos+1])
{
bidui_bmhs++;
}
else if (src[i] != des[i-pos+1])
{
bidui_bmhs++;
break;
}
}
if ((i-pos+2+len_d) == len_d)
{
match_bmhs++;
res.push_back(pos-1);
pos += len_d;
}
else
{
pos += alphabet[src[pos+len_d-1]] + 1;
}
}
}
void initP(string &pstr, int (&m1)[ALPHABET_SIZE], int (&m2)[ZuHe_SIZE])
{
int len = pstr.size();
for(int i=0; i<len; ++i)
{
m1[pstr[i]] = 1;
}
for(int i=0; i<len-1; ++i)
{
m2[pstr[i]*256+pstr[i+1]] = i;
}
}
void BMY(string &pstr, const size_t &plen, string &sstr, const size_t &slen, int (&m1)[ALPHABET_SIZE], int (&m2)[ZuHe_SIZE], vector<int> &res)
{
int i,j; //i:模式串游标 j:文本串游标
i = j = plen-1;
int cnt = -1;
while(j < slen)
{
num_bmy++;
//模式串和文本串从后向前匹配
while( (i!=0) && pstr[i] == sstr[j])
{
--i;
--j;
bidui_bmy++;
}
bidui_bmy++;
//发现一个匹配的模式串
if(i==0 && pstr[i] == sstr[j])
{
res.push_back(j);
match_bmy++;
} //匹配成功及出现失配的情况
j += (plen-i-1);
if(m1[sstr[j+1]] == 0 && m1[sstr[j+2]] == 0) //尾1和尾2都不在模式串中
j += (plen + 2);
else if(sstr[j+1] == pstr[plen-1]) //尾1在模式串尾
j += 1;
else if( (cnt = m2[sstr[j+1]*256+sstr[j+2]]) != -1) //尾1和尾2的组合在模式串中
j += (plen - cnt);
else if(sstr[j+2] == pstr[0]) //尾2在模式串头
j += (plen + 1);
else
j += (plen + 2);
cnt = -1;
i = plen - 1;
}
}
void GetNext(string p, int (&next)[ALPHABET_SIZE])
{
int pLen = p.size();
next[0] = -1;
int k = -1;
int j = 0;
while (j < pLen-1 )
{
if(k == -1 || p[j] == p[k]) //p[k]表示前缀,p[j]表示后缀
{
++k;
++j;
next[j] = k; // k == next[j-1] + 1
}
else
{
k = next[k]; // next[k] == next[next[j]](递归的思想)
}
}
}
void KMP_Search(string &s, size_t slen, string &p, size_t plen, int (&next)[ALPHABET_SIZE], vector<int> &res)
{
int i = 0;
int j = 0;
while (j < slen) // i:模式串游标 j:文本串游标
{
num_kmp++;
//模式串和文本串从前向后匹配
while(i == -1 || s[j] == p[i])
{
i++;
j++;
bidui_kmp++;
if(i == plen)
break;
}
bidui_kmp++;
//找到匹配的模式串
if(i == plen)
{
res.push_back(j-i);
j += 1;
i = 0;
match_kmp++;
}
//出现失配的情况
else
{
//如果i != -1,且当前字符匹配失败(即S[i] != P[j]),则令j不变,i = next[j],使模式串右移
i = next[i];
}
}
}
void BuildBadC(string &pattern, size_t pattern_length, unsigned int *badc, size_t alphabet_size)
{
unsigned int i;
for(i = 0; i < alphabet_size; ++i)
{
badc[i] = pattern_length;
}
for(i = 0; i < pattern_length; ++i)
{
badc[pattern[i] ] = pattern_length - 1 - i;
}
}
void BuildGoodS(const char *pattern, size_t pattern_length, unsigned int* goods)
{
unsigned int i, j, c;
for(i = 0; i < pattern_length - 1; ++i)
{
goods[i] = pattern_length;
}
//初始化pattern最末元素的好后缀值
goods[pattern_length - 1] = 1;
//此循环找出pattern中各元素的pre值,这里goods数组先当作pre数组使用
for(i = pattern_length -1, c = 0; i != 0; --i)
{
for(j = 0; j < i; ++j)
{
if(memcmp(pattern + i, pattern + j, (pattern_length - i) * sizeof(char)) == 0)
{
if(j == 0)
{
c = pattern_length - i;
}
else
{
if(pattern[i - 1] != pattern[j - 1])
{
goods[i - 1] = j - 1;
}
}
}
}
}
//根据pattern中个元素的pre值,计算goods值
for(i = 0; i < pattern_length - 1; ++i)
{
if(goods[i] != pattern_length)
{
goods[i] = pattern_length - 1 - goods[i];
}
else
{
goods[i] = pattern_length - 1 - i + goods[i];
if(c != 0 && pattern_length - 1 - i >= c)
{
goods[i] -= c;
}
}
}
}
void BM(string &pattern, size_t pattern_length, string &text, size_t text_length, unsigned int *badc, unsigned int *goods, vector<int> &res)
{
unsigned int i, j;
i = j = pattern_length - 1;
while(j < text_length)
{
num_bm++;
//发现目标传与模式传从后向前第1个不匹配的位置
while((i != 0) && (pattern[i] == text[j]))
{
--i;
--j;
bidui_bm++;
}
bidui_bm++;
//找到一个匹配的情况
if(i == 0 && pattern[i] == text[j])
{
res.push_back(j);
j += goods[0];
match_bm++;
}
else
{
//坏字符表用字典构建比较合适
j += goods[i] > badc[text[j]] ? goods[i] : badc[text[j]];
}
i = pattern_length - 1;
}
}
int main()
{
string pstr = "want";
ifstream fin("2.txt");
string sstr( (istreambuf_iterator<char>(fin)), istreambuf_iterator<char>() );
for(int i=0; i<5; i++)
sstr.insert(sstr.end(),sstr.begin(),sstr.end());
const char* pstr_ = pstr.data();
const size_t plen = pstr.size();
const size_t slen = sstr.size();
/*BMHS*/
int alphabet[ALPHABET_SIZE] = { 0 };
InitBMHS(alphabet, pstr, plen);
vector<int> res4;
clock_t start4 = clock();
sunday(sstr, slen, pstr, plen, alphabet, res4);
cout<<"BMHS 时间: "<<(double)(clock()-start4)/CLOCKS_PER_SEC*1000<<"ms"<<endl;
cout<<"BMHS 识别模式串个数: "<<match_bmhs<<endl;
cout<<"BMHS 匹配次数: "<<num_bmhs<<endl;
cout<<"BMHS 比对次数:"<<bidui_bmhs<<endl<<endl;
// for(auto it : res4)
// cout<<it<<endl;
/*KMP*/
int next[ALPHABET_SIZE];
for(int i=0; i<ALPHABET_SIZE; i++)
next[i] = -1;
GetNext(pstr, next);
vector<int> res2;
clock_t start2 = clock();
KMP_Search(sstr, slen, pstr, plen, next, res2);
cout<<"KMP 时间: "<< (double)(clock()-start2)/CLOCKS_PER_SEC*1000<<"ms"<<endl;
cout<<"KMP 识别模式串个数: "<<match_kmp<<endl;
cout<<"KMP 匹配次数: "<<num_kmp<<endl;
cout<<"KMP 比对次数:"<<bidui_kmp<<endl<<endl;
// for(auto it : res2)
// cout<<it<<endl;
/*BMY*/
int m1[ALPHABET_SIZE] = {0};
int m2[ZuHe_SIZE];
for(int i=0; i<ZuHe_SIZE; i++)
m2[i] = -1;
initP(pstr, m1, m2);
vector<int> res1;
clock_t start = clock();
BMY(pstr, plen, sstr, slen, m1, m2, res1);
cout<<"BMY 时间: "<< (double)(clock()-start)/CLOCKS_PER_SEC*1000<<"ms"<<endl;
cout<<"BMY 识别模式串个数: "<<match_bmy<<endl;
cout<<"BMY 匹配次数: "<<num_bmy<<endl;
cout<<"BMY 比对次数:"<<bidui_bmy<<endl<<endl;
// for(auto it : res1)
// cout<<it<<endl;
/*BM*/
unsigned int badc[256];
unsigned int goods[plen];
BuildBadC(pstr, plen, badc, 256);
BuildGoodS(pstr_, plen, goods);
vector<int> res3;
clock_t start3 = clock();
BM(pstr, plen, sstr, slen, badc, goods, res3);
cout<<"BM 时间: "<< (double)(clock()-start3)/CLOCKS_PER_SEC*1000<<"ms"<<endl;
cout<<"BM 识别模式串个数: "<<match_bm<<endl;
cout<<"BM 匹配次数: "<<num_bm<<endl;
cout<<"BM 比对次数:"<<bidui_bm<<endl<<endl;
// for(auto it : res3)
// cout<<it<<endl;
return 0;
}