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
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
|
#include <sys/stat.h>
#include <boost/foreach.hpp>
#include "line_splitter.hh"
#include "storing.hh"
#include "StoreTarget.h"
#include "StoreVocab.h"
#include "moses/Util.h"
#include "moses/InputFileStream.h"
using namespace std;
namespace Moses
{
///////////////////////////////////////////////////////////////////////
void Node::Add(Table &table, const SourcePhrase &sourcePhrase, size_t pos)
{
if (pos < sourcePhrase.size()) {
uint64_t vocabId = sourcePhrase[pos];
Node *child;
Children::iterator iter = m_children.find(vocabId);
if (iter == m_children.end()) {
// New node. Write other children then discard them
BOOST_FOREACH(Children::value_type &valPair, m_children) {
Node &otherChild = valPair.second;
otherChild.Write(table);
}
m_children.clear();
// create new node
child = &m_children[vocabId];
assert(!child->done);
child->key = key + (vocabId << pos);
}
else {
child = &iter->second;
}
child->Add(table, sourcePhrase, pos + 1);
}
else {
// this node was written previously 'cos it has rules
done = true;
}
}
void Node::Write(Table &table)
{
//cerr << "START write " << done << " " << key << endl;
BOOST_FOREACH(Children::value_type &valPair, m_children) {
Node &child = valPair.second;
child.Write(table);
}
if (!done) {
// save
Entry sourceEntry;
sourceEntry.value = NONE;
sourceEntry.key = key;
//Put into table
table.Insert(sourceEntry);
}
}
///////////////////////////////////////////////////////////////////////
void createProbingPT(const std::string &phrasetable_path,
const std::string &basepath, int num_scores, int num_lex_scores,
bool log_prob, int max_cache_size, bool scfg)
{
std::cerr << "Starting..." << std::endl;
//Get basepath and create directory if missing
mkdir(basepath.c_str(), S_IRWXU | S_IRWXG | S_IROTH | S_IXOTH);
StoreTarget storeTarget(basepath);
//Get uniq lines:
unsigned long uniq_entries = countUniqueSource(phrasetable_path);
//Source phrase vocabids
StoreVocab<uint64_t> sourceVocab(basepath + "/source_vocabids");
//Read the file
util::FilePiece filein(phrasetable_path.c_str());
//Init the probing hash table
size_t size = Table::Size(uniq_entries, 1.2);
char * mem = new char[size];
memset(mem, 0, size);
Table sourceEntries(mem, size);
std::priority_queue<CacheItem*, std::vector<CacheItem*>, CacheItemOrderer> cache;
float totalSourceCount = 0;
//Keep track of the size of each group of target phrases
size_t line_num = 0;
//Read everything and processs
std::string prevSource;
Node sourcePhrases;
sourcePhrases.done = true;
sourcePhrases.key = 0;
while (true) {
try {
//Process line read
line_text line;
line = splitLine(filein.ReadLine(), scfg);
//cerr << "line=" << line.source_phrase << endl;
++line_num;
if (line_num % 1000000 == 0) {
std::cerr << line_num << " " << std::flush;
}
//Add source phrases to vocabularyIDs
add_to_map(sourceVocab, line.source_phrase);
if (prevSource.empty()) {
// 1st line
prevSource = line.source_phrase.as_string();
storeTarget.Append(line, log_prob, scfg);
}
else if (prevSource == line.source_phrase) {
//If we still have the same line, just append to it:
storeTarget.Append(line, log_prob, scfg);
}
else {
assert(prevSource != line.source_phrase);
//Create a new entry even
// save
uint64_t targetInd = storeTarget.Save();
// next line
storeTarget.Append(line, log_prob, scfg);
//Create an entry for the previous source phrase:
Entry sourceEntry;
sourceEntry.value = targetInd;
//The key is the sum of hashes of individual words bitshifted by their position in the phrase.
//Probably not entirerly correct, but fast and seems to work fine in practise.
std::vector<uint64_t> vocabid_source = getVocabIDs(prevSource);
if (scfg) {
// storing prefixes?
sourcePhrases.Add(sourceEntries, vocabid_source);
}
sourceEntry.key = getKey(vocabid_source);
/*
cerr << "prevSource=" << prevSource << flush
<< " vocabids=" << Debug(vocabid_source) << flush
<< " key=" << sourceEntry.key << endl;
*/
//Put into table
sourceEntries.Insert(sourceEntry);
// update cache - CURRENT source phrase, not prev
if (max_cache_size) {
std::string countStr = line.counts.as_string();
countStr = Trim(countStr);
if (!countStr.empty()) {
std::vector<float> toks = Tokenize<float>(countStr);
//cerr << "CACHE:" << line.source_phrase << " " << countStr << " " << toks[1] << endl;
if (toks.size() >= 2) {
totalSourceCount += toks[1];
// compute key for CURRENT source
std::vector<uint64_t> currVocabidSource = getVocabIDs(line.source_phrase.as_string());
uint64_t currKey = getKey(currVocabidSource);
CacheItem *item = new CacheItem(
Trim(line.source_phrase.as_string()),
currKey,
toks[1]);
cache.push(item);
if (max_cache_size > 0 && cache.size() > max_cache_size) {
cache.pop();
}
}
}
}
//Set prevLine
prevSource = line.source_phrase.as_string();
}
}
catch (util::EndOfFileException e) {
std::cerr
<< "Reading phrase table finished, writing remaining files to disk."
<< std::endl;
//After the final entry is constructed we need to add it to the phrase_table
//Create an entry for the previous source phrase:
uint64_t targetInd = storeTarget.Save();
Entry sourceEntry;
sourceEntry.value = targetInd;
//The key is the sum of hashes of individual words. Probably not entirerly correct, but fast
std::vector<uint64_t> vocabid_source = getVocabIDs(prevSource);
sourceEntry.key = getKey(vocabid_source);
//Put into table
sourceEntries.Insert(sourceEntry);
break;
}
}
sourcePhrases.Write(sourceEntries);
storeTarget.SaveAlignment();
serialize_table(mem, size, (basepath + "/probing_hash.dat"));
sourceVocab.Save();
serialize_cache(cache, (basepath + "/cache"), totalSourceCount);
delete[] mem;
//Write configfile
std::ofstream configfile;
configfile.open((basepath + "/config").c_str());
configfile << "API_VERSION\t" << API_VERSION << '\n';
configfile << "uniq_entries\t" << uniq_entries << '\n';
configfile << "num_scores\t" << num_scores << '\n';
configfile << "num_lex_scores\t" << num_lex_scores << '\n';
configfile << "log_prob\t" << log_prob << '\n';
configfile.close();
}
size_t countUniqueSource(const std::string &path)
{
size_t ret = 0;
InputFileStream strme(path);
std::string line, prevSource;
while (std::getline(strme, line)) {
std::vector<std::string> toks = TokenizeMultiCharSeparator(line, "|||");
assert(toks.size() != 0);
if (prevSource != toks[0]) {
prevSource = toks[0];
++ret;
}
}
return ret;
}
void serialize_cache(
std::priority_queue<CacheItem*, std::vector<CacheItem*>, CacheItemOrderer> &cache,
const std::string &path, float totalSourceCount)
{
std::vector<const CacheItem*> vec(cache.size());
size_t ind = cache.size() - 1;
while (!cache.empty()) {
const CacheItem *item = cache.top();
vec[ind] = item;
cache.pop();
--ind;
}
std::ofstream os(path.c_str());
os << totalSourceCount << std::endl;
for (size_t i = 0; i < vec.size(); ++i) {
const CacheItem *item = vec[i];
os << item->count << "\t" << item->sourceKey << "\t" << item->source << std::endl;
delete item;
}
os.close();
}
uint64_t getKey(const std::vector<uint64_t> &vocabid_source)
{
return getKey(vocabid_source.data(), vocabid_source.size());
}
std::vector<uint64_t> CreatePrefix(const std::vector<uint64_t> &vocabid_source, size_t endPos)
{
assert(endPos < vocabid_source.size());
std::vector<uint64_t> ret(endPos + 1);
for (size_t i = 0; i <= endPos; ++i) {
ret[i] = vocabid_source[i];
}
return ret;
}
}
|