summaryrefslogtreecommitdiff
path: root/extractor/precomputation.cc
blob: e29018c2a581ffb4e6bb72cdfdcc772798841abc (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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
#include "precomputation.h"

#include <iostream>
#include <queue>

#include "data_array.h"
#include "suffix_array.h"

using namespace std;

namespace extractor {

int Precomputation::FIRST_NONTERMINAL = -1;
int Precomputation::SECOND_NONTERMINAL = -2;

Precomputation::Precomputation(
    shared_ptr<SuffixArray> suffix_array, int num_frequent_patterns,
    int num_super_frequent_patterns, int max_rule_span,
    int max_rule_symbols, int min_gap_size,
    int max_frequent_phrase_len, int min_frequency) {
  vector<int> data = suffix_array->GetData()->GetData();
  vector<vector<int> > frequent_patterns = FindMostFrequentPatterns(
      suffix_array, data, num_frequent_patterns, max_frequent_phrase_len,
      min_frequency);

  unordered_set<vector<int>, VectorHash> frequent_patterns_set;
  unordered_set<vector<int>, VectorHash> super_frequent_patterns_set;
  for (size_t i = 0; i < frequent_patterns.size(); ++i) {
    frequent_patterns_set.insert(frequent_patterns[i]);
    if (i < num_super_frequent_patterns) {
      super_frequent_patterns_set.insert(frequent_patterns[i]);
    }
  }

  vector<tuple<int, int, int> > matchings;
  for (size_t i = 0; i < data.size(); ++i) {
    if (data[i] == DataArray::END_OF_LINE) {
      AddCollocations(matchings, data, max_rule_span, min_gap_size,
          max_rule_symbols);
      matchings.clear();
      continue;
    }
    vector<int> pattern;
    for (int j = 1; j <= max_frequent_phrase_len && i + j <= data.size(); ++j) {
      pattern.push_back(data[i + j - 1]);
      if (frequent_patterns_set.count(pattern)) {
        int is_super_frequent = super_frequent_patterns_set.count(pattern);
        matchings.push_back(make_tuple(i, j, is_super_frequent));
      } else {
        // If the current pattern is not frequent, any longer pattern having the
        // current pattern as prefix will not be frequent.
        break;
      }
    }
  }
}

Precomputation::Precomputation() {}

Precomputation::~Precomputation() {}

vector<vector<int> > Precomputation::FindMostFrequentPatterns(
    shared_ptr<SuffixArray> suffix_array, const vector<int>& data,
    int num_frequent_patterns, int max_frequent_phrase_len, int min_frequency) {
  vector<int> lcp = suffix_array->BuildLCPArray();
  vector<int> run_start(max_frequent_phrase_len);

  priority_queue<pair<int, pair<int, int> > > heap;
  for (size_t i = 1; i < lcp.size(); ++i) {
    for (int len = lcp[i]; len < max_frequent_phrase_len; ++len) {
      int frequency = i - run_start[len];
      // TODO(pauldb): Only add patterns that don't span across multiple
      // sentences.
      if (frequency >= min_frequency) {
        heap.push(make_pair(frequency,
            make_pair(suffix_array->GetSuffix(run_start[len]), len + 1)));
      }
      run_start[len] = i;
    }
  }

  vector<vector<int> > frequent_patterns;
  while (frequent_patterns.size() < num_frequent_patterns && !heap.empty()) {
    int start = heap.top().second.first;
    int len = heap.top().second.second;
    heap.pop();

    vector<int> pattern(data.begin() + start, data.begin() + start + len);
    if (find(pattern.begin(), pattern.end(), DataArray::END_OF_LINE) ==
        pattern.end()) {
      frequent_patterns.push_back(pattern);
    }
  }
  return frequent_patterns;
}

void Precomputation::AddCollocations(
    const vector<tuple<int, int, int> >& matchings, const vector<int>& data,
    int max_rule_span, int min_gap_size, int max_rule_symbols) {
  for (size_t i = 0; i < matchings.size(); ++i) {
    int start1, size1, is_super1;
    tie(start1, size1, is_super1) = matchings[i];

    for (size_t j = i + 1; j < matchings.size(); ++j) {
      int start2, size2, is_super2;
      tie(start2, size2, is_super2) = matchings[j];
      if (start2 - start1 >= max_rule_span) {
        break;
      }

      if (start2 - start1 - size1 >= min_gap_size
          && start2 + size2 - start1 <= max_rule_span
          && size1 + size2 + 1 <= max_rule_symbols) {
        vector<int> pattern(data.begin() + start1,
            data.begin() + start1 + size1);
        pattern.push_back(Precomputation::FIRST_NONTERMINAL);
        pattern.insert(pattern.end(), data.begin() + start2,
            data.begin() + start2 + size2);
        AddStartPositions(collocations[pattern], start1, start2);

        if (is_super2) {
          pattern.push_back(Precomputation::SECOND_NONTERMINAL);
          for (size_t k = j + 1; k < matchings.size(); ++k) {
            int start3, size3, is_super3;
            tie(start3, size3, is_super3) = matchings[k];
            if (start3 - start1 >= max_rule_span) {
              break;
            }

            if (start3 - start2 - size2 >= min_gap_size
                && start3 + size3 - start1 <= max_rule_span
                && size1 + size2 + size3 + 2 <= max_rule_symbols
                && (is_super1 || is_super3)) {
              pattern.insert(pattern.end(), data.begin() + start3,
                  data.begin() + start3 + size3);
              AddStartPositions(collocations[pattern], start1, start2, start3);
              pattern.erase(pattern.end() - size3);
            }
          }
        }
      }
    }
  }
}

void Precomputation::AddStartPositions(
    vector<int>& positions, int pos1, int pos2) {
  positions.push_back(pos1);
  positions.push_back(pos2);
}

void Precomputation::AddStartPositions(
    vector<int>& positions, int pos1, int pos2, int pos3) {
  positions.push_back(pos1);
  positions.push_back(pos2);
  positions.push_back(pos3);
}

void Precomputation::WriteBinary(const fs::path& filepath) const {
  FILE* file = fopen(filepath.string().c_str(), "w");

  // TODO(pauldb): Refactor this code.
  int size = collocations.size();
  fwrite(&size, sizeof(int), 1, file);
  for (auto entry: collocations) {
    size = entry.first.size();
    fwrite(&size, sizeof(int), 1, file);
    fwrite(entry.first.data(), sizeof(int), size, file);

    size = entry.second.size();
    fwrite(&size, sizeof(int), 1, file);
    fwrite(entry.second.data(), sizeof(int), size, file);
  }
}

const Index& Precomputation::GetCollocations() const {
  return collocations;
}

} // namespace extractor