summaryrefslogtreecommitdiff
path: root/utils/synutils/tsuruoka_maxent.h
blob: b5a87404f62dd383f2443848d071b25b5d925348 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
/*
 * tsuruoka_maxent.h
 *
 */

#ifndef TSURUOKA_MAXENT_H_
#define TSURUOKA_MAXENT_H_

#include "utility.h"
#include "stringlib.h"
#include "maxent-3.0/maxent.h"

#include <assert.h>
#include <vector>
#include <string>
#include <string.h>
#include <tr1/unordered_map>

using namespace std;

typedef std::tr1::unordered_map<std::string, int> Map;
typedef std::tr1::unordered_map<std::string, int>::iterator Iterator;

struct Tsuruoka_Maxent {
  Tsuruoka_Maxent(const char* pszModelFName) {
    if (pszModelFName != NULL) {
      m_pModel = new ME_Model();
      m_pModel->load_from_file(pszModelFName);
    } else
      m_pModel = NULL;
  }

  ~Tsuruoka_Maxent() {
    if (m_pModel != NULL) delete m_pModel;
  }

  void fnTrain(const char* pszInstanceFName, const char* pszAlgorithm,
               const char* pszModelFName, int iNumIteration) {
    assert(strcmp(pszAlgorithm, "l1") == 0 || strcmp(pszAlgorithm, "l2") == 0 ||
           strcmp(pszAlgorithm, "sgd") == 0 ||
           strcmp(pszAlgorithm, "SGD") == 0);
    FILE* fpIn = fopen(pszInstanceFName, "r");

    ME_Model* pModel = new ME_Model();

    char* pszLine = new char[100001];
    int iNumInstances = 0;
    int iLen;
    while (!feof(fpIn)) {
      pszLine[0] = '\0';
      fgets(pszLine, 20000, fpIn);
      if (strlen(pszLine) == 0) {
        continue;
      }

      iLen = strlen(pszLine);
      while (iLen > 0 && pszLine[iLen - 1] > 0 && pszLine[iLen - 1] < 33) {
        pszLine[iLen - 1] = '\0';
        iLen--;
      }

      iNumInstances++;

      ME_Sample* pmes = new ME_Sample();

      char* p = strrchr(pszLine, ' ');
      assert(p != NULL);
      p[0] = '\0';
      p++;
      vector<string> vecContext;
      SplitOnWhitespace(string(pszLine), &vecContext);

      pmes->label = string(p);
      for (size_t i = 0; i < vecContext.size(); i++)
        pmes->add_feature(vecContext[i]);
      pModel->add_training_sample((*pmes));
      if (iNumInstances % 100000 == 0)
        fprintf(stdout, "......Reading #Instances: %1d\n", iNumInstances);
      delete pmes;
    }
    fprintf(stdout, "......Reading #Instances: %1d\n", iNumInstances);
    fclose(fpIn);

    if (strcmp(pszAlgorithm, "l1") == 0)
      pModel->use_l1_regularizer(1.0);
    else if (strcmp(pszAlgorithm, "l2") == 0)
      pModel->use_l2_regularizer(1.0);
    else
      pModel->use_SGD();

    pModel->train();
    pModel->save_to_file(pszModelFName);

    delete pModel;
    fprintf(stdout, "......Finished Training\n");
    fprintf(stdout, "......Model saved as %s\n", pszModelFName);
    delete[] pszLine;
  }

  double fnEval(const char* pszContext, const char* pszOutcome) const {
    vector<string> vecContext;
    ME_Sample* pmes = new ME_Sample();
    SplitOnWhitespace(string(pszContext), &vecContext);

    for (size_t i = 0; i < vecContext.size(); i++)
      pmes->add_feature(vecContext[i]);
    vector<double> vecProb = m_pModel->classify(*pmes);
    delete pmes;
    int iLableID = m_pModel->get_class_id(pszOutcome);
    return vecProb[iLableID];
  }
  void fnEval(const char* pszContext,
              vector<pair<string, double> >& vecOutput) const {
    vector<string> vecContext;
    ME_Sample* pmes = new ME_Sample();
    SplitOnWhitespace(string(pszContext), &vecContext);

    vecOutput.clear();

    for (size_t i = 0; i < vecContext.size(); i++)
      pmes->add_feature(vecContext[i]);
    vector<double> vecProb = m_pModel->classify(*pmes);

    for (size_t i = 0; i < vecProb.size(); i++) {
      string label = m_pModel->get_class_label(i);
      vecOutput.push_back(make_pair(label, vecProb[i]));
    }
    delete pmes;
  }
  void fnEval(const char* pszContext, vector<double>& vecOutput) const {
    vector<string> vecContext;
    ME_Sample* pmes = new ME_Sample();
    SplitOnWhitespace(string(pszContext), &vecContext);

    vecOutput.clear();

    for (size_t i = 0; i < vecContext.size(); i++)
      pmes->add_feature(vecContext[i]);
    vector<double> vecProb = m_pModel->classify(*pmes);

    for (size_t i = 0; i < vecProb.size(); i++) {
      string label = m_pModel->get_class_label(i);
      vecOutput.push_back(vecProb[i]);
    }
    delete pmes;
  }
  int fnGetClassId(const string& strLabel) const {
    return m_pModel->get_class_id(strLabel);
  }

 private:
  ME_Model* m_pModel;
};

#endif /* TSURUOKA_MAXENT_H_ */