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
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
|
/*
* Data.cpp
* met - Minimum Error Training
*
* Created by Nicola Bertoldi on 13/05/08.
*
*/
#include <algorithm>
#include "util/check.hh"
#include <cmath>
#include <fstream>
#include "Data.h"
#include "FileStream.h"
#include "Scorer.h"
#include "ScorerFactory.h"
#include "Util.h"
Data::Data()
: theScorer(NULL),
number_of_scores(0),
_sparse_flag(false),
scoredata(NULL),
featdata(NULL) {}
Data::Data(Scorer& ptr)
: theScorer(&ptr),
score_type(theScorer->getName()),
number_of_scores(0),
_sparse_flag(false),
scoredata(new ScoreData(*theScorer)),
featdata(new FeatureData)
{
TRACE_ERR("Data::score_type " << score_type << std::endl);
TRACE_ERR("Data::Scorer type from Scorer: " << theScorer->getName() << endl);
}
Data::~Data() {
if (featdata) {
delete featdata;
featdata = NULL;
}
if (scoredata) {
delete scoredata;
scoredata = NULL;
}
}
//ADDED BY TS
void Data::remove_duplicates() {
uint nSentences = featdata->size();
assert(scoredata->size() == nSentences);
for (uint s=0; s < nSentences; s++) {
FeatureArray& feat_array = featdata->get(s);
ScoreArray& score_array = scoredata->get(s);
assert(feat_array.size() == score_array.size());
//serves as a hash-map:
std::map<double, std::vector<uint> > lookup;
uint end_pos = feat_array.size() - 1;
uint nRemoved = 0;
for (uint k=0; k <= end_pos; k++) {
const FeatureStats& cur_feats = feat_array.get(k);
double sum = 0.0;
for (uint l=0; l < cur_feats.size(); l++)
sum += cur_feats.get(l);
if (lookup.find(sum) != lookup.end()) {
//std::cerr << "hit" << std::endl;
std::vector<uint>& cur_list = lookup[sum];
uint l=0;
for (l=0; l < cur_list.size(); l++) {
uint j=cur_list[l];
if (cur_feats == feat_array.get(j)
&& score_array.get(k) == score_array.get(j)) {
if (k < end_pos) {
feat_array.swap(k,end_pos);
score_array.swap(k,end_pos);
k--;
}
end_pos--;
nRemoved++;
break;
}
}
if (l == lookup[sum].size())
cur_list.push_back(k);
}
else
lookup[sum].push_back(k);
// for (uint j=0; j < k; j++) {
// if (feat_array.get(k) == feat_array.get(j)
// && score_array.get(k) == score_array.get(j)) {
// if (k < end_pos) {
// feat_array.swap(k,end_pos);
// score_array.swap(k,end_pos);
// k--;
// }
// end_pos--;
// nRemoved++;
// break;
// }
// }
}
std::cerr << "removed " << nRemoved << "/" << feat_array.size() << std::endl;
if (nRemoved > 0) {
feat_array.resize(end_pos+1);
score_array.resize(end_pos+1);
}
}
}
//END_ADDED
void Data::loadnbest(const std::string &file)
{
TRACE_ERR("loading nbest from " << file << std::endl);
FeatureStats featentry;
ScoreStats scoreentry;
std::string sentence_index;
inputfilestream inp(file); // matches a stream with a file. Opens the file
if (!inp.good())
throw runtime_error("Unable to open: " + file);
std::string substring, subsubstring, stringBuf;
std::string theSentence;
std::string::size_type loc;
while (getline(inp,stringBuf,'\n')) {
if (stringBuf.empty()) continue;
// TRACE_ERR("stringBuf: " << stringBuf << std::endl);
getNextPound(stringBuf, substring, "|||"); //first field
sentence_index = substring;
getNextPound(stringBuf, substring, "|||"); //second field
theSentence = substring;
// adding statistics for error measures
featentry.reset();
scoreentry.clear();
theScorer->prepareStats(sentence_index, theSentence, scoreentry);
scoredata->add(scoreentry, sentence_index);
getNextPound(stringBuf, substring, "|||"); //third field
// examine first line for name of features
if (!existsFeatureNames()) {
std::string stringsupport=substring;
std::string features="";
std::string tmpname="";
size_t tmpidx=0;
while (!stringsupport.empty()) {
// TRACE_ERR("Decompounding: " << substring << std::endl);
getNextPound(stringsupport, subsubstring);
// string ending with ":" are skipped, because they are the names of the features
if ((loc = subsubstring.find_last_of(":")) != subsubstring.length()-1) {
features+=tmpname+"_"+stringify(tmpidx)+" ";
tmpidx++;
}
// ignore sparse feature name
else if (subsubstring.find("_") != string::npos) {
// also ignore its value
getNextPound(stringsupport, subsubstring);
}
// update current feature name
else {
tmpidx=0;
tmpname=subsubstring.substr(0,subsubstring.size() - 1);
}
}
featdata->setFeatureMap(features);
}
// adding features
while (!substring.empty()) {
// TRACE_ERR("Decompounding: " << substring << std::endl);
getNextPound(substring, subsubstring);
// no ':' -> feature value that needs to be stored
if ((loc = subsubstring.find_last_of(":")) != subsubstring.length()-1) {
featentry.add(ConvertStringToFeatureStatsType(subsubstring));
}
// sparse feature name? store as well
else if (subsubstring.find("_") != string::npos) {
std::string name = subsubstring;
getNextPound(substring, subsubstring);
featentry.addSparse( name, atof(subsubstring.c_str()) );
_sparse_flag = true;
}
}
//cerr << "number of sparse features: " << featentry.getSparse().size() << endl;
featdata->add(featentry,sentence_index);
}
inp.close();
}
// TODO
void Data::mergeSparseFeatures() {
std::cerr << "ERROR: sparse features can only be trained with pairwise ranked optimizer (PRO), not traditional MERT\n";
exit(1);
}
void Data::createShards(size_t shard_count, float shard_size, const string& scorerconfig,
std::vector<Data>& shards)
{
CHECK(shard_count);
CHECK(shard_size >= 0);
CHECK(shard_size <= 1);
size_t data_size = scoredata->size();
CHECK(data_size == featdata->size());
shard_size *= data_size;
for (size_t shard_id = 0; shard_id < shard_count; ++shard_id) {
vector<size_t> shard_contents;
if (shard_size == 0) {
//split into roughly equal size shards
size_t shard_start = floor(0.5 + shard_id * (float)data_size / shard_count);
size_t shard_end = floor(0.5 + (shard_id+1) * (float)data_size / shard_count);
for (size_t i = shard_start; i < shard_end; ++i) {
shard_contents.push_back(i);
}
} else {
//create shards by randomly sampling
for (size_t i = 0; i < floor(shard_size+0.5); ++i) {
shard_contents.push_back(rand() % data_size);
}
}
Scorer* scorer = ScorerFactory::getScorer(score_type, scorerconfig);
shards.push_back(Data(*scorer));
shards.back().score_type = score_type;
shards.back().number_of_scores = number_of_scores;
shards.back()._sparse_flag = _sparse_flag;
for (size_t i = 0; i < shard_contents.size(); ++i) {
shards.back().featdata->add(featdata->get(shard_contents[i]));
shards.back().scoredata->add(scoredata->get(shard_contents[i]));
}
//cerr << endl;
}
}
|