cppjieba_rb 0.2.1

data/ext/cppjieba/dict/user.dict.utf8

@@ -0,0 +1,4 @@
+云计算
+韩玉鉴赏
+蓝翔 nz
+区块链 10 nz

data/ext/cppjieba/include/cppjieba/DictTrie.hpp

@@ -0,0 +1,227 @@
+#ifndef CPPJIEBA_DICT_TRIE_HPP
+#define CPPJIEBA_DICT_TRIE_HPP
+
+#include <iostream>
+#include <fstream>
+#include <map>
+#include <string>
+#include <cstring>
+#include <cstdlib>
+#include <stdint.h>
+#include <cmath>
+#include <limits>
+#include "limonp/StringUtil.hpp"
+#include "limonp/Logging.hpp"
+#include "Unicode.hpp"
+#include "Trie.hpp"
+
+namespace cppjieba {
+
+using namespace limonp;
+
+const double MIN_DOUBLE = -3.14e+100;
+const double MAX_DOUBLE = 3.14e+100;
+const size_t DICT_COLUMN_NUM = 3;
+const char* const UNKNOWN_TAG = "";
+
+class DictTrie {
+ public:
+  enum UserWordWeightOption {
+    WordWeightMin,
+    WordWeightMedian,
+    WordWeightMax,
+  }; // enum UserWordWeightOption
+
+  DictTrie(const string& dict_path, const string& user_dict_paths = "", UserWordWeightOption user_word_weight_opt = WordWeightMedian) {
+    Init(dict_path, user_dict_paths, user_word_weight_opt);
+  }
+
+  ~DictTrie() {
+    delete trie_;
+  }
+
+  bool InsertUserWord(const string& word, const string& tag = UNKNOWN_TAG) {
+    DictUnit node_info;
+    if (!MakeNodeInfo(node_info, word, user_word_default_weight_, tag)) {
+      return false;
+    }
+    active_node_infos_.push_back(node_info);
+    trie_->InsertNode(node_info.word, &active_node_infos_.back());
+    return true;
+  }
+
+  const DictUnit* Find(RuneStrArray::const_iterator begin, RuneStrArray::const_iterator end) const {
+    return trie_->Find(begin, end);
+  }
+
+  void Find(RuneStrArray::const_iterator begin, 
+        RuneStrArray::const_iterator end, 
+        vector<struct Dag>&res,
+        size_t max_word_len = MAX_WORD_LENGTH) const {
+    trie_->Find(begin, end, res, max_word_len);
+  }
+
+  bool IsUserDictSingleChineseWord(const Rune& word) const {
+    return IsIn(user_dict_single_chinese_word_, word);
+  }
+
+  double GetMinWeight() const {
+    return min_weight_;
+  }
+
+ private:
+  void Init(const string& dict_path, const string& user_dict_paths, UserWordWeightOption user_word_weight_opt) {
+    LoadDict(dict_path);
+    freq_sum_ = CalcFreqSum(static_node_infos_);
+    CalculateWeight(static_node_infos_, freq_sum_);
+    SetStaticWordWeights(user_word_weight_opt);
+
+    if (user_dict_paths.size()) {
+      LoadUserDict(user_dict_paths);
+    }
+    Shrink(static_node_infos_);
+    CreateTrie(static_node_infos_);
+  }
+  
+  void CreateTrie(const vector<DictUnit>& dictUnits) {
+    assert(dictUnits.size());
+    vector<Unicode> words;
+    vector<const DictUnit*> valuePointers;
+    for (size_t i = 0 ; i < dictUnits.size(); i ++) {
+      words.push_back(dictUnits[i].word);
+      valuePointers.push_back(&dictUnits[i]);
+    }
+
+    trie_ = new Trie(words, valuePointers);
+  }
+
+  void LoadUserDict(const string& filePaths) {
+    vector<string> files = limonp::Split(filePaths, "|;");
+    size_t lineno = 0;
+    for (size_t i = 0; i < files.size(); i++) {
+      ifstream ifs(files[i].c_str());
+      XCHECK(ifs.is_open()) << "open " << files[i] << " failed"; 
+      string line;
+      DictUnit node_info;
+      vector<string> buf;
+      for (; getline(ifs, line); lineno++) {
+        if (line.size() == 0) {
+          continue;
+        }
+        buf.clear();
+        Split(line, buf, " ");
+        DictUnit node_info;
+        if(buf.size() == 1){
+          MakeNodeInfo(node_info, 
+                buf[0], 
+                user_word_default_weight_,
+                UNKNOWN_TAG);
+        } else if (buf.size() == 2) {
+          MakeNodeInfo(node_info, 
+                buf[0], 
+                user_word_default_weight_,
+                buf[1]);
+        } else if (buf.size() == 3) {
+          int freq = atoi(buf[1].c_str());
+          assert(freq_sum_ > 0.0);
+          double weight = log(1.0 * freq / freq_sum_);
+          MakeNodeInfo(node_info, buf[0], weight, buf[2]);
+        }
+        static_node_infos_.push_back(node_info);
+        if (node_info.word.size() == 1) {
+          user_dict_single_chinese_word_.insert(node_info.word[0]);
+        }
+      }
+    }
+  }
+
+  bool MakeNodeInfo(DictUnit& node_info,
+        const string& word, 
+        double weight, 
+        const string& tag) {
+    if (!DecodeRunesInString(word, node_info.word)) {
+      XLOG(ERROR) << "Decode " << word << " failed.";
+      return false;
+    }
+    node_info.weight = weight;
+    node_info.tag = tag;
+    return true;
+  }
+
+  void LoadDict(const string& filePath) {
+    ifstream ifs(filePath.c_str());
+    XCHECK(ifs.is_open()) << "open " << filePath << " failed.";
+    string line;
+    vector<string> buf;
+
+    DictUnit node_info;
+    for (size_t lineno = 0; getline(ifs, line); lineno++) {
+      Split(line, buf, " ");
+      XCHECK(buf.size() == DICT_COLUMN_NUM) << "split result illegal, line:" << line;
+      MakeNodeInfo(node_info, 
+            buf[0], 
+            atof(buf[1].c_str()), 
+            buf[2]);
+      static_node_infos_.push_back(node_info);
+    }
+  }
+
+  static bool WeightCompare(const DictUnit& lhs, const DictUnit& rhs) {
+    return lhs.weight < rhs.weight;
+  }
+
+  void SetStaticWordWeights(UserWordWeightOption option) {
+    XCHECK(!static_node_infos_.empty());
+    vector<DictUnit> x = static_node_infos_;
+    sort(x.begin(), x.end(), WeightCompare);
+    min_weight_ = x[0].weight;
+    max_weight_ = x[x.size() - 1].weight;
+    median_weight_ = x[x.size() / 2].weight;
+    switch (option) {
+     case WordWeightMin:
+       user_word_default_weight_ = min_weight_;
+       break;
+     case WordWeightMedian:
+       user_word_default_weight_ = median_weight_;
+       break;
+     default:
+       user_word_default_weight_ = max_weight_;
+       break;
+    }
+  }
+
+  double CalcFreqSum(const vector<DictUnit>& node_infos) const {
+    double sum = 0.0;
+    for (size_t i = 0; i < node_infos.size(); i++) {
+      sum += node_infos[i].weight;
+    }
+    return sum;
+  }
+
+  void CalculateWeight(vector<DictUnit>& node_infos, double sum) const {
+    assert(sum > 0.0);
+    for (size_t i = 0; i < node_infos.size(); i++) {
+      DictUnit& node_info = node_infos[i];
+      assert(node_info.weight > 0.0);
+      node_info.weight = log(double(node_info.weight)/sum);
+    }
+  }
+
+  void Shrink(vector<DictUnit>& units) const {
+    vector<DictUnit>(units.begin(), units.end()).swap(units);
+  }
+
+  vector<DictUnit> static_node_infos_;
+  deque<DictUnit> active_node_infos_; // must not be vector
+  Trie * trie_;
+
+  double freq_sum_;
+  double min_weight_;
+  double max_weight_;
+  double median_weight_;
+  double user_word_default_weight_;
+  unordered_set<Rune> user_dict_single_chinese_word_;
+};
+}
+
+#endif

data/ext/cppjieba/include/cppjieba/FullSegment.hpp

@@ -0,0 +1,93 @@
+#ifndef CPPJIEBA_FULLSEGMENT_H
+#define CPPJIEBA_FULLSEGMENT_H
+
+#include <algorithm>
+#include <set>
+#include <cassert>
+#include "limonp/Logging.hpp"
+#include "DictTrie.hpp"
+#include "SegmentBase.hpp"
+#include "Unicode.hpp"
+
+namespace cppjieba {
+class FullSegment: public SegmentBase {
+ public:
+  FullSegment(const string& dictPath) {
+    dictTrie_ = new DictTrie(dictPath);
+    isNeedDestroy_ = true;
+  }
+  FullSegment(const DictTrie* dictTrie)
+    : dictTrie_(dictTrie), isNeedDestroy_(false) {
+    assert(dictTrie_);
+  }
+  ~FullSegment() {
+    if (isNeedDestroy_) {
+      delete dictTrie_;
+    }
+  }
+  void Cut(const string& sentence, 
+        vector<string>& words) const {
+    vector<Word> tmp;
+    Cut(sentence, tmp);
+    GetStringsFromWords(tmp, words);
+  }
+  void Cut(const string& sentence, 
+        vector<Word>& words) const {
+    PreFilter pre_filter(symbols_, sentence);
+    PreFilter::Range range;
+    vector<WordRange> wrs;
+    wrs.reserve(sentence.size()/2);
+    while (pre_filter.HasNext()) {
+      range = pre_filter.Next();
+      Cut(range.begin, range.end, wrs);
+    }
+    words.clear();
+    words.reserve(wrs.size());
+    GetWordsFromWordRanges(sentence, wrs, words);
+  }
+  void Cut(RuneStrArray::const_iterator begin, 
+        RuneStrArray::const_iterator end, 
+        vector<WordRange>& res) const {
+    //resut of searching in trie tree
+    LocalVector<pair<size_t, const DictUnit*> > tRes;
+
+    //max index of res's words
+    int maxIdx = 0;
+
+    // always equals to (uItr - begin)
+    int uIdx = 0;
+
+    //tmp variables
+    int wordLen = 0;
+    assert(dictTrie_);
+    vector<struct Dag> dags;
+    dictTrie_->Find(begin, end, dags);
+    for (size_t i = 0; i < dags.size(); i++) {
+      for (size_t j = 0; j < dags[i].nexts.size(); j++) {
+        size_t nextoffset = dags[i].nexts[j].first;
+        assert(nextoffset < dags.size());
+        const DictUnit* du = dags[i].nexts[j].second;
+        if (du == NULL) {
+          if (dags[i].nexts.size() == 1 && maxIdx <= uIdx) {
+            WordRange wr(begin + i, begin + nextoffset);
+            res.push_back(wr);
+          }
+        } else {
+          wordLen = du->word.size();
+          if (wordLen >= 2 || (dags[i].nexts.size() == 1 && maxIdx <= uIdx)) {
+            WordRange wr(begin + i, begin + nextoffset);
+            res.push_back(wr);
+          }
+        }
+        maxIdx = uIdx + wordLen > maxIdx ? uIdx + wordLen : maxIdx;
+      }
+      uIdx++;
+    }
+  }
+ private:
+  const DictTrie* dictTrie_;
+  bool isNeedDestroy_;
+};
+}
+
+#endif

data/ext/cppjieba/include/cppjieba/HMMModel.hpp

@@ -0,0 +1,129 @@
+#ifndef CPPJIEBA_HMMMODEL_H
+#define CPPJIEBA_HMMMODEL_H
+
+#include "limonp/StringUtil.hpp"
+#include "Trie.hpp"
+
+namespace cppjieba {
+
+using namespace limonp;
+typedef unordered_map<Rune, double> EmitProbMap;
+
+struct HMMModel {
+  /*
+   * STATUS:
+   * 0: HMMModel::B, 1: HMMModel::E, 2: HMMModel::M, 3:HMMModel::S
+   * */
+  enum {B = 0, E = 1, M = 2, S = 3, STATUS_SUM = 4};
+
+  HMMModel(const string& modelPath) {
+    memset(startProb, 0, sizeof(startProb));
+    memset(transProb, 0, sizeof(transProb));
+    statMap[0] = 'B';
+    statMap[1] = 'E';
+    statMap[2] = 'M';
+    statMap[3] = 'S';
+    emitProbVec.push_back(&emitProbB);
+    emitProbVec.push_back(&emitProbE);
+    emitProbVec.push_back(&emitProbM);
+    emitProbVec.push_back(&emitProbS);
+    LoadModel(modelPath);
+  }
+  ~HMMModel() {
+  }
+  void LoadModel(const string& filePath) {
+    ifstream ifile(filePath.c_str());
+    XCHECK(ifile.is_open()) << "open " << filePath << " failed";
+    string line;
+    vector<string> tmp;
+    vector<string> tmp2;
+    //Load startProb
+    XCHECK(GetLine(ifile, line));
+    Split(line, tmp, " ");
+    XCHECK(tmp.size() == STATUS_SUM);
+    for (size_t j = 0; j< tmp.size(); j++) {
+      startProb[j] = atof(tmp[j].c_str());
+    }
+
+    //Load transProb
+    for (size_t i = 0; i < STATUS_SUM; i++) {
+      XCHECK(GetLine(ifile, line));
+      Split(line, tmp, " ");
+      XCHECK(tmp.size() == STATUS_SUM);
+      for (size_t j =0; j < STATUS_SUM; j++) {
+        transProb[i][j] = atof(tmp[j].c_str());
+      }
+    }
+
+    //Load emitProbB
+    XCHECK(GetLine(ifile, line));
+    XCHECK(LoadEmitProb(line, emitProbB));
+
+    //Load emitProbE
+    XCHECK(GetLine(ifile, line));
+    XCHECK(LoadEmitProb(line, emitProbE));
+
+    //Load emitProbM
+    XCHECK(GetLine(ifile, line));
+    XCHECK(LoadEmitProb(line, emitProbM));
+
+    //Load emitProbS
+    XCHECK(GetLine(ifile, line));
+    XCHECK(LoadEmitProb(line, emitProbS));
+  }
+  double GetEmitProb(const EmitProbMap* ptMp, Rune key, 
+        double defVal)const {
+    EmitProbMap::const_iterator cit = ptMp->find(key);
+    if (cit == ptMp->end()) {
+      return defVal;
+    }
+    return cit->second;
+  }
+  bool GetLine(ifstream& ifile, string& line) {
+    while (getline(ifile, line)) {
+      Trim(line);
+      if (line.empty()) {
+        continue;
+      }
+      if (StartsWith(line, "#")) {
+        continue;
+      }
+      return true;
+    }
+    return false;
+  }
+  bool LoadEmitProb(const string& line, EmitProbMap& mp) {
+    if (line.empty()) {
+      return false;
+    }
+    vector<string> tmp, tmp2;
+    Unicode unicode;
+    Split(line, tmp, ",");
+    for (size_t i = 0; i < tmp.size(); i++) {
+      Split(tmp[i], tmp2, ":");
+      if (2 != tmp2.size()) {
+        XLOG(ERROR) << "emitProb illegal.";
+        return false;
+      }
+      if (!DecodeRunesInString(tmp2[0], unicode) || unicode.size() != 1) {
+        XLOG(ERROR) << "TransCode failed.";
+        return false;
+      }
+      mp[unicode[0]] = atof(tmp2[1].c_str());
+    }
+    return true;
+  }
+
+  char statMap[STATUS_SUM];
+  double startProb[STATUS_SUM];
+  double transProb[STATUS_SUM][STATUS_SUM];
+  EmitProbMap emitProbB;
+  EmitProbMap emitProbE;
+  EmitProbMap emitProbM;
+  EmitProbMap emitProbS;
+  vector<EmitProbMap* > emitProbVec;
+}; // struct HMMModel
+
+} // namespace cppjieba
+
+#endif

data/ext/cppjieba/include/cppjieba/HMMSegment.hpp

@@ -0,0 +1,190 @@
+#ifndef CPPJIBEA_HMMSEGMENT_H
+#define CPPJIBEA_HMMSEGMENT_H
+
+#include <iostream>
+#include <fstream>
+#include <memory.h>
+#include <cassert>
+#include "HMMModel.hpp"
+#include "SegmentBase.hpp"
+
+namespace cppjieba {
+class HMMSegment: public SegmentBase {
+ public:
+  HMMSegment(const string& filePath)
+  : model_(new HMMModel(filePath)), isNeedDestroy_(true) {
+  }
+  HMMSegment(const HMMModel* model) 
+  : model_(model), isNeedDestroy_(false) {
+  }
+  ~HMMSegment() {
+    if (isNeedDestroy_) {
+      delete model_;
+    }
+  }
+
+  void Cut(const string& sentence, 
+        vector<string>& words) const {
+    vector<Word> tmp;
+    Cut(sentence, tmp);
+    GetStringsFromWords(tmp, words);
+  }
+  void Cut(const string& sentence, 
+        vector<Word>& words) const {
+    PreFilter pre_filter(symbols_, sentence);
+    PreFilter::Range range;
+    vector<WordRange> wrs;
+    wrs.reserve(sentence.size()/2);
+    while (pre_filter.HasNext()) {
+      range = pre_filter.Next();
+      Cut(range.begin, range.end, wrs);
+    }
+    words.clear();
+    words.reserve(wrs.size());
+    GetWordsFromWordRanges(sentence, wrs, words);
+  }
+  void Cut(RuneStrArray::const_iterator begin, RuneStrArray::const_iterator end, vector<WordRange>& res) const {
+    RuneStrArray::const_iterator left = begin;
+    RuneStrArray::const_iterator right = begin;
+    while (right != end) {
+      if (right->rune < 0x80) {
+        if (left != right) {
+          InternalCut(left, right, res);
+        }
+        left = right;
+        do {
+          right = SequentialLetterRule(left, end);
+          if (right != left) {
+            break;
+          }
+          right = NumbersRule(left, end);
+          if (right != left) {
+            break;
+          }
+          right ++;
+        } while (false);
+        WordRange wr(left, right - 1);
+        res.push_back(wr);
+        left = right;
+      } else {
+        right++;
+      }
+    }
+    if (left != right) {
+      InternalCut(left, right, res);
+    }
+  }
+ private:
+  // sequential letters rule
+  RuneStrArray::const_iterator SequentialLetterRule(RuneStrArray::const_iterator begin, RuneStrArray::const_iterator end) const {
+    Rune x = begin->rune;
+    if (('a' <= x && x <= 'z') || ('A' <= x && x <= 'Z')) {
+      begin ++;
+    } else {
+      return begin;
+    }
+    while (begin != end) {
+      x = begin->rune;
+      if (('a' <= x && x <= 'z') || ('A' <= x && x <= 'Z') || ('0' <= x && x <= '9')) {
+        begin ++;
+      } else {
+        break;
+      }
+    }
+    return begin;
+  }
+  //
+  RuneStrArray::const_iterator NumbersRule(RuneStrArray::const_iterator begin, RuneStrArray::const_iterator end) const {
+    Rune x = begin->rune;
+    if ('0' <= x && x <= '9') {
+      begin ++;
+    } else {
+      return begin;
+    }
+    while (begin != end) {
+      x = begin->rune;
+      if ( ('0' <= x && x <= '9') || x == '.') {
+        begin++;
+      } else {
+        break;
+      }
+    }
+    return begin;
+  }
+  void InternalCut(RuneStrArray::const_iterator begin, RuneStrArray::const_iterator end, vector<WordRange>& res) const {
+    vector<size_t> status;
+    Viterbi(begin, end, status);
+
+    RuneStrArray::const_iterator left = begin;
+    RuneStrArray::const_iterator right;
+    for (size_t i = 0; i < status.size(); i++) {
+      if (status[i] % 2) { //if (HMMModel::E == status[i] || HMMModel::S == status[i])
+        right = begin + i + 1;
+        WordRange wr(left, right - 1);
+        res.push_back(wr);
+        left = right;
+      }
+    }
+  }
+
+  void Viterbi(RuneStrArray::const_iterator begin, 
+        RuneStrArray::const_iterator end, 
+        vector<size_t>& status) const {
+    size_t Y = HMMModel::STATUS_SUM;
+    size_t X = end - begin;
+
+    size_t XYSize = X * Y;
+    size_t now, old, stat;
+    double tmp, endE, endS;
+
+    vector<int> path(XYSize);
+    vector<double> weight(XYSize);
+
+    //start
+    for (size_t y = 0; y < Y; y++) {
+      weight[0 + y * X] = model_->startProb[y] + model_->GetEmitProb(model_->emitProbVec[y], begin->rune, MIN_DOUBLE);
+      path[0 + y * X] = -1;
+    }
+
+    double emitProb;
+
+    for (size_t x = 1; x < X; x++) {
+      for (size_t y = 0; y < Y; y++) {
+        now = x + y*X;
+        weight[now] = MIN_DOUBLE;
+        path[now] = HMMModel::E; // warning
+        emitProb = model_->GetEmitProb(model_->emitProbVec[y], (begin+x)->rune, MIN_DOUBLE);
+        for (size_t preY = 0; preY < Y; preY++) {
+          old = x - 1 + preY * X;
+          tmp = weight[old] + model_->transProb[preY][y] + emitProb;
+          if (tmp > weight[now]) {
+            weight[now] = tmp;
+            path[now] = preY;
+          }
+        }
+      }
+    }
+
+    endE = weight[X-1+HMMModel::E*X];
+    endS = weight[X-1+HMMModel::S*X];
+    stat = 0;
+    if (endE >= endS) {
+      stat = HMMModel::E;
+    } else {
+      stat = HMMModel::S;
+    }
+
+    status.resize(X);
+    for (int x = X -1 ; x >= 0; x--) {
+      status[x] = stat;
+      stat = path[x + stat*X];
+    }
+  }
+
+  const HMMModel* model_;
+  bool isNeedDestroy_;
+}; // class HMMSegment
+
+} // namespace cppjieba
+
+#endif