use automation

This commit is contained in:
wyy 2014-11-12 18:55:17 +08:00
parent b9736ee132
commit 3ced451212
5 changed files with 162 additions and 74 deletions

View File

@ -23,24 +23,6 @@ namespace CppJieba
const size_t DICT_COLUMN_NUM = 3;
const char* const UNKNOWN_TAG = "x";
struct DictUnit
{
Unicode word;
double weight;
string tag;
};
inline ostream & operator << (ostream& os, const DictUnit& unit)
{
string s;
s << unit.word;
return os << string_format("%s %s %.3lf", s.c_str(), unit.tag.c_str(), unit.weight);
}
typedef std::vector<std::pair<size_t, const DictUnit*> > DagType;
class DictTrie
{
public:
@ -107,6 +89,14 @@ namespace CppJieba
{
return _trie->find(begin, end, dag, offset);
}
void find(
Unicode::const_iterator begin,
Unicode::const_iterator end,
vector<SegmentChar>& res
) const
{
_trie->find(begin, end, res);
}
private:

View File

@ -1,7 +1,3 @@
/************************************
* file enc : ASCII
* author : wuyanyi09@gmail.com
************************************/
#ifndef CPPJIEBA_MPSEGMENT_H
#define CPPJIEBA_MPSEGMENT_H
@ -17,17 +13,6 @@
namespace CppJieba
{
struct SegmentChar
{
uint16_t uniCh;
DagType dag;
const DictUnit * pInfo;
double weight;
size_t nextPos;
SegmentChar():uniCh(0), pInfo(NULL), weight(0.0), nextPos(0)
{}
};
class MPSegment: public SegmentBase
{
private:
@ -85,16 +70,9 @@ namespace CppJieba
{
return false;
}
vector<SegmentChar> segmentChars(end - begin);
vector<SegmentChar> segmentChars;
//calc DAG
for(size_t i = 0; i < segmentChars.size(); i ++)
{
segmentChars[i].uniCh = *(begin + i);
segmentChars[i].dag.clear();
segmentChars[i].dag.push_back(std::pair<size_t, const DictUnit*>(i, NULL));
_dictTrie.find(begin + i, end, segmentChars[i].dag, i);
}
_dictTrie.find(begin, end, segmentChars);
_calcDP(segmentChars);
@ -112,24 +90,25 @@ namespace CppJieba
}
private:
void _calcDP(vector<SegmentChar>& SegmentChars) const
void _calcDP(vector<SegmentChar>& segmentChars) const
{
size_t nextPos;
const DictUnit* p;
double val;
for(int i = SegmentChars.size() - 1; i >= 0; i--)
for(ssize_t i = segmentChars.size() - 1; i >= 0; i--)
{
SegmentChars[i].pInfo = NULL;
SegmentChars[i].weight = MIN_DOUBLE;
for(DagType::const_iterator it = SegmentChars[i].dag.begin(); it != SegmentChars[i].dag.end(); it++)
segmentChars[i].pInfo = NULL;
segmentChars[i].weight = MIN_DOUBLE;
assert(!segmentChars[i].dag.empty());
for(DagType::const_iterator it = segmentChars[i].dag.begin(); it != segmentChars[i].dag.end(); it++)
{
nextPos = it->first;
p = it->second;
val = 0.0;
if(nextPos + 1 < SegmentChars.size())
if(nextPos + 1 < segmentChars.size())
{
val += SegmentChars[nextPos + 1].weight;
val += segmentChars[nextPos + 1].weight;
}
if(p)
@ -140,10 +119,10 @@ namespace CppJieba
{
val += _dictTrie.getMinWeight();
}
if(val > SegmentChars[i].weight)
if(val > segmentChars[i].weight)
{
SegmentChars[i].pInfo = p;
SegmentChars[i].weight = val;
segmentChars[i].pInfo = p;
segmentChars[i].weight = val;
}
}
}

View File

@ -3,18 +3,51 @@
#include "Limonp/StdExtension.hpp"
#include <vector>
#include <queue>
namespace CppJieba
{
using namespace std;
struct DictUnit
{
Unicode word;
double weight;
string tag;
};
inline ostream & operator << (ostream& os, const DictUnit& unit)
{
string s;
s << unit.word;
return os << string_format("%s %s %.3lf", s.c_str(), unit.tag.c_str(), unit.weight);
}
typedef std::vector<std::pair<size_t, const DictUnit*> > DagType;
struct SegmentChar
{
uint16_t uniCh;
DagType dag;
const DictUnit * pInfo;
double weight;
size_t nextPos;
SegmentChar():uniCh(0), pInfo(NULL), weight(0.0), nextPos(0)
{}
};
template <class KeyType, class ValueType>
class TrieNode
{
public:
typedef unordered_map<KeyType, TrieNode<KeyType, ValueType>* > KeyMapType;
typedef unordered_map<KeyType, TrieNode<KeyType, ValueType>* > NextMap;
public:
KeyMapType * ptKeyMap;
TrieNode * fail;
NextMap * next;
const ValueType * ptValue;
public:
TrieNode(): fail(NULL), next(NULL), ptValue(NULL) {
}
};
template <class KeyType, class ValueType, class KeyContainerType = vector<KeyType>, class KeysContainerType = vector<KeyContainerType>, class ValueContainerType = vector<const ValueType* > >
@ -28,10 +61,8 @@ namespace CppJieba
Trie(const KeysContainerType& keys, const ValueContainerType& valuePointers)
{
_root = new TrieNodeType;
_root->ptKeyMap = NULL;
_root->ptValue = NULL;
_createTrie(keys, valuePointers);
_build();// build automation
}
~Trie()
{
@ -43,12 +74,12 @@ namespace CppJieba
public:
const ValueType* find(typename KeyContainerType::const_iterator begin, typename KeyContainerType::const_iterator end) const
{
typename TrieNodeType::KeyMapType::const_iterator citer;
typename TrieNodeType::NextMap::const_iterator citer;
const TrieNodeType* ptNode = _root;
for(typename KeyContainerType::const_iterator it = begin; it != end; it++)
{
{// build automation
assert(ptNode);
if(NULL == ptNode->ptKeyMap || ptNode->ptKeyMap->end() == (citer = ptNode->ptKeyMap->find(*it)))
if(NULL == ptNode->next || ptNode->next->end() == (citer = ptNode->next->find(*it)))
{
return NULL;
}
@ -56,6 +87,47 @@ namespace CppJieba
}
return ptNode->ptValue;
}
void find(
typename KeyContainerType::const_iterator begin,
typename KeyContainerType::const_iterator end,
vector<struct SegmentChar>& res
) const
{
res.resize(end - begin);
const TrieNodeType * now = _root;
typename TrieNodeType::NextMap::const_iterator iter;
for (size_t i = 0; i < end - begin; i++) {
bool flag = false;
res[i].uniCh = *(begin + i);
assert(res[i].dag.empty());
res[i].dag.reserve(4);//TODO
while( now != _root && (now->next == NULL || (iter = now->next->find(*(begin + i))) == now->next->end())) {
now = now->fail;
}
if(now->next == NULL || (iter = now->next->find(*(begin + i))) == now->next->end()) {
now = _root;
} else {
now = iter->second;
const TrieNodeType * temp = now;
while(temp != _root) {
if (temp->ptValue) {
string str;
TransCode::encode(temp->ptValue->word, str);
size_t pos = i - temp->ptValue->word.size() + 1;
res[pos].dag.push_back(pair<typename KeysContainerType::size_type, const ValueType* >(i, temp->ptValue));
if(temp->ptValue->word.size() == 1) {
flag = true;
}
}
temp = temp->fail;
assert(temp);
}
}
if(!flag) {
res[i].dag.push_back(pair<typename KeysContainerType::size_type, const ValueType* >(i, NULL));
}
}
}
bool find(
typename KeyContainerType::const_iterator begin,
typename KeyContainerType::const_iterator end,
@ -63,11 +135,11 @@ namespace CppJieba
size_t offset = 0) const
{
const TrieNodeType * ptNode = _root;
typename TrieNodeType::KeyMapType::const_iterator citer;
typename TrieNodeType::NextMap::const_iterator citer;
for(typename KeyContainerType::const_iterator itr = begin; itr != end ; itr++)
{
assert(ptNode);
if(NULL == ptNode->ptKeyMap || ptNode->ptKeyMap->end() == (citer = ptNode->ptKeyMap->find(*itr)))
if(NULL == ptNode->next || ptNode->next->end() == (citer = ptNode->next->find(*itr)))
{
break;
}
@ -86,6 +158,42 @@ namespace CppJieba
}
return !res.empty();
}
private:
void _build()
{
queue<TrieNodeType*> que;
assert(_root->ptValue == NULL);
assert(_root->next);
_root->fail = NULL;
for(typename TrieNodeType::NextMap::iterator iter = _root->next->begin(); iter != _root->next->end(); iter++) {
iter->second->fail = _root;
que.push(iter->second);
}
TrieNodeType* back = NULL;
typename TrieNodeType::NextMap::iterator backiter;
while(!que.empty()) {
TrieNodeType * now = que.front();
que.pop();
if(now->next == NULL) {
continue;
}
for(typename TrieNodeType::NextMap::iterator iter = now->next->begin(); iter != now->next->end(); iter++) {
back = now->fail;
while(back != NULL) {
if(back->next && (backiter = back->next->find(iter->first)) != back->next->end())
{
iter->second->fail = backiter->second;
break;
}
back = back->fail;
}
if(back == NULL) {
iter->second->fail = _root;
}
que.push(iter->second);
}
}
}
private:
void _createTrie(const KeysContainerType& keys, const ValueContainerType& valuePointers)
{
@ -105,22 +213,22 @@ namespace CppJieba
{
TrieNodeType* ptNode = _root;
typename TrieNodeType::KeyMapType::const_iterator kmIter;
typename TrieNodeType::NextMap::const_iterator kmIter;
for(typename KeyContainerType::const_iterator citer = key.begin(); citer != key.end(); citer++)
{
if(NULL == ptNode->ptKeyMap)
if(NULL == ptNode->next)
{
ptNode->ptKeyMap = new typename TrieNodeType::KeyMapType;
ptNode->next = new typename TrieNodeType::NextMap;
}
kmIter = ptNode->ptKeyMap->find(*citer);
if(ptNode->ptKeyMap->end() == kmIter)
kmIter = ptNode->next->find(*citer);
if(ptNode->next->end() == kmIter)
{
TrieNodeType * nextNode = new TrieNodeType;
nextNode->ptKeyMap = NULL;
nextNode->next = NULL;
nextNode->ptValue = NULL;
(*ptNode->ptKeyMap)[*citer] = nextNode;
(*ptNode->next)[*citer] = nextNode;
ptNode = nextNode;
}
else
@ -136,14 +244,14 @@ namespace CppJieba
{
return;
}
if(node->ptKeyMap)
if(node->next)
{
typename TrieNodeType::KeyMapType::iterator it;
for(it = node->ptKeyMap->begin(); it != node->ptKeyMap->end(); it++)
typename TrieNodeType::NextMap::iterator it;
for(it = node->next->begin(); it != node->next->end(); it++)
{
_deleteNode(it->second);
}
delete node->ptKeyMap;
delete node->next;
}
delete node;
}

View File

@ -21,7 +21,6 @@ TEST(MixSegmentTest, Test1)
ASSERT_EQ(words, vector<string>(res, res + sizeof(res)/sizeof(res[0])));
ASSERT_TRUE(segment.cut(str2, words));
ASSERT_EQ(words, vector<string>(res2, res2 + sizeof(res2)/sizeof(res2[0])));
//exit(0);
}
TEST(MixSegmentTest, NoUserDict)

View File

@ -1,4 +1,5 @@
#include "src/DictTrie.hpp"
#include "src/MPSegment.hpp"
#include "gtest/gtest.h"
using namespace CppJieba;
@ -64,3 +65,14 @@ TEST(DictTrieTest, UserDict)
res << *unit;
ASSERT_EQ("[\"20113\", \"35745\", \"31639\"] x -2.975", res);
}
TEST(DictTrieTest, automation)
{
DictTrie trie(DICT_FILE, "../test/testdata/userdict.utf8");
//string word = "yasherhs";
string word = "abcderf";
Unicode unicode;
ASSERT_TRUE(TransCode::decode(word, unicode));
vector<struct SegmentChar> res;
trie.find(unicode.begin(), unicode.end(), res);
}