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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
|
/*
* Filter a grammar in striped format
*/
#include <iostream>
#include <string>
#include <map>
#include <vector>
#include <utility>
#include <cstdlib>
#include <fstream>
#include <tr1/unordered_map>
#include "suffix_tree.h"
#include "sparse_vector.h"
#include "sentence_pair.h"
#include "extract.h"
#include "fdict.h"
#include "tdict.h"
#include "lex_trans_tbl.h"
#include "filelib.h"
#include <boost/shared_ptr.hpp>
#include <boost/functional/hash.hpp>
#include <boost/program_options.hpp>
#include <boost/program_options/variables_map.hpp>
using namespace std;
using namespace std::tr1;
namespace po = boost::program_options;
static const size_t MAX_LINE_LENGTH = 64000000;
typedef unordered_map<vector<WordID>, RuleStatistics, boost::hash<vector<WordID> > > ID2RuleStatistics;
void InitCommandLine(int argc, char** argv, po::variables_map* conf) {
po::options_description opts("Configuration options");
opts.add_options()
("test_set,t", po::value<string>(), "Filter for this test set")
("top_e_given_f,n", po::value<size_t>()->default_value(30), "Keep top N rules, according to p(e|f). 0 for all")
("help,h", "Print this help message and exit");
po::options_description clo("Command line options");
po::options_description dcmdline_options;
dcmdline_options.add(opts);
po::store(parse_command_line(argc, argv, dcmdline_options), *conf);
po::notify(*conf);
if (conf->count("help") || conf->count("test_set")==0) {
cerr << "\nUsage: filter_grammar -t TEST-SET.fr [-options] < grammar\n";
cerr << dcmdline_options << endl;
exit(1);
}
}
namespace {
inline bool IsWhitespace(char c) { return c == ' ' || c == '\t'; }
inline bool IsBracket(char c){return c == '[' || c == ']';}
inline void SkipWhitespace(const char* buf, int* ptr) {
while (buf[*ptr] && IsWhitespace(buf[*ptr])) { ++(*ptr); }
}
}
int ReadPhraseUntilDividerOrEnd(const char* buf, const int sstart, const int end, vector<WordID>* p) {
static const WordID kDIV = TD::Convert("|||");
int ptr = sstart;
while(ptr < end) {
while(ptr < end && IsWhitespace(buf[ptr])) { ++ptr; }
int start = ptr;
while(ptr < end && !IsWhitespace(buf[ptr])) { ++ptr; }
if (ptr == start) {cerr << "Warning! empty token.\n"; return ptr; }
const WordID w = TD::Convert(string(buf, start, ptr - start));
if((IsBracket(buf[start]) and IsBracket(buf[ptr-1])) or( w == kDIV))
p->push_back(1 * w);
else {
if (w == kDIV) return ptr;
p->push_back(w);
}
}
return ptr;
}
void ParseLine(const char* buf, vector<WordID>* cur_key, ID2RuleStatistics* counts) {
static const WordID kDIV = TD::Convert("|||");
counts->clear();
int ptr = 0;
while(buf[ptr] != 0 && buf[ptr] != '\t') { ++ptr; }
if (buf[ptr] != '\t') {
cerr << "Missing tab separator between key and value!\n INPUT=" << buf << endl;
exit(1);
}
cur_key->clear();
// key is: "[X] ||| word word word"
int tmpp = ReadPhraseUntilDividerOrEnd(buf, 0, ptr, cur_key);
cur_key->push_back(kDIV);
ReadPhraseUntilDividerOrEnd(buf, tmpp, ptr, cur_key);
++ptr;
int start = ptr;
int end = ptr;
int state = 0; // 0=reading label, 1=reading count
vector<WordID> name;
while(buf[ptr] != 0) {
while(buf[ptr] != 0 && buf[ptr] != '|') { ++ptr; }
if (buf[ptr] == '|') {
++ptr;
if (buf[ptr] == '|') {
++ptr;
if (buf[ptr] == '|') {
++ptr;
end = ptr - 3;
while (end > start && IsWhitespace(buf[end-1])) { --end; }
if (start == end) {
cerr << "Got empty token!\n LINE=" << buf << endl;
exit(1);
}
switch (state) {
case 0: ++state; name.clear(); ReadPhraseUntilDividerOrEnd(buf, start, end, &name); break;
case 1: --state; (*counts)[name].ParseRuleStatistics(buf, start, end); break;
default: cerr << "Can't happen\n"; abort();
}
SkipWhitespace(buf, &ptr);
start = ptr;
}
}
}
}
end=ptr;
while (end > start && IsWhitespace(buf[end-1])) { --end; }
if (end > start) {
switch (state) {
case 0: ++state; name.clear(); ReadPhraseUntilDividerOrEnd(buf, start, end, &name); break;
case 1: --state; (*counts)[name].ParseRuleStatistics(buf, start, end); break;
default: cerr << "Can't happen\n"; abort();
}
}
}
struct SourceFilter {
// return true to keep the rule, otherwise false
virtual bool Matches(const vector<WordID>& key) const = 0;
virtual ~SourceFilter() {}
};
struct DumbSuffixTreeFilter : SourceFilter {
DumbSuffixTreeFilter(const string& corpus) :
kDIV(TD::Convert("|||")) {
cerr << "Build suffix tree from test set in " << corpus << endl;
assert(FileExists(corpus));
ReadFile rfts(corpus);
istream& testSet = *rfts.stream();
char* buf = new char[MAX_LINE_LENGTH];
AnnotatedParallelSentence sent;
/* process the data set to build suffix tree
*/
while(!testSet.eof()) {
testSet.getline(buf, MAX_LINE_LENGTH);
if (buf[0] == 0) continue;
//hack to read in the test set using AnnotatedParallelSentence
strcat(buf," ||| fake ||| 0-0");
sent.ParseInputLine(buf);
//add each successive suffix to the tree
for(int i=0; i<sent.f_len; i++)
root.InsertPath(sent.f, i, sent.f_len - 1);
}
delete[] buf;
}
virtual bool Matches(const vector<WordID>& key) const {
const Node<int>* curnode = &root;
const int ks = key.size() - 1;
for(int i=0; i < ks; i++) {
const string& word = TD::Convert(key[i]);
if (key[i] == kDIV || (word[0] == '[' && word[word.size() - 1] == ']')) { // non-terminal
curnode = &root;
} else if (curnode) {
curnode = curnode->Extend(key[i]);
if (!curnode) return false;
}
}
return true;
}
const WordID kDIV;
Node<int> root;
};
int main(int argc, char** argv){
po::variables_map conf;
InitCommandLine(argc, argv, &conf);
const int max_options = conf["top_e_given_f"].as<size_t>();;
istream& unscored_grammar = cin;
cerr << "Loading test set " << conf["test_set"].as<string>() << "...\n";
boost::shared_ptr<SourceFilter> filter;
filter.reset(new DumbSuffixTreeFilter(conf["test_set"].as<string>()));
cerr << "Filtering...\n";
//score unscored grammar
char* buf = new char[MAX_LINE_LENGTH];
ID2RuleStatistics acc, cur_counts;
vector<WordID> key, cur_key,temp_key;
int line = 0;
multimap<float, ID2RuleStatistics::const_iterator> options;
const int kCOUNT = FD::Convert("CFE");
while(!unscored_grammar.eof())
{
++line;
options.clear();
unscored_grammar.getline(buf, MAX_LINE_LENGTH);
if (buf[0] == 0) continue;
ParseLine(buf, &cur_key, &cur_counts);
if (!filter || filter->Matches(cur_key)) {
// sort by counts
for (ID2RuleStatistics::const_iterator it = cur_counts.begin(); it != cur_counts.end(); ++it) {
options.insert(make_pair(-it->second.counts.value(kCOUNT), it));
}
int ocount = 0;
cout << TD::GetString(cur_key) << '\t';
bool first = true;
for (multimap<float,ID2RuleStatistics::const_iterator>::iterator it = options.begin(); it != options.end(); ++it) {
if (first) { first = false; } else { cout << " ||| "; }
cout << TD::GetString(it->second->first) << " ||| " << it->second->second;
++ocount;
if (ocount == max_options) break;
}
cout << endl;
}
}
}
|
