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
|
#include <sstream>
#include <iostream>
#include <fstream>
#include <vector>
#include <cassert>
#include <cmath>
#include <tr1/memory>
#include <boost/program_options.hpp>
#include <boost/program_options/variables_map.hpp>
#include "verbose.h"
#include "hg.h"
#include "prob.h"
#include "inside_outside.h"
#include "ff_register.h"
#include "decoder.h"
#include "filelib.h"
#include "online_optimizer.h"
#include "fdict.h"
#include "weights.h"
#include "sparse_vector.h"
#include "sampler.h"
#ifdef HAVE_MPI
#include <boost/mpi/timer.hpp>
#include <boost/mpi.hpp>
namespace mpi = boost::mpi;
#endif
using namespace std;
namespace po = boost::program_options;
struct FComp {
const vector<double>& w_;
FComp(const vector<double>& w) : w_(w) {}
bool operator()(int a, int b) const {
return fabs(w_[a]) > fabs(w_[b]);
}
};
void ShowFeatures(const vector<double>& w) {
vector<int> fnums(w.size());
for (int i = 0; i < w.size(); ++i)
fnums[i] = i;
sort(fnums.begin(), fnums.end(), FComp(w));
for (vector<int>::iterator i = fnums.begin(); i != fnums.end(); ++i) {
if (w[*i]) cout << FD::Convert(*i) << ' ' << w[*i] << endl;
}
}
void ReadConfig(const string& ini, vector<string>* out) {
ReadFile rf(ini);
istream& in = *rf.stream();
while(in) {
string line;
getline(in, line);
if (!in) continue;
out->push_back(line);
}
}
void StoreConfig(const vector<string>& cfg, istringstream* o) {
ostringstream os;
for (int i = 0; i < cfg.size(); ++i) { os << cfg[i] << endl; }
o->str(os.str());
}
bool InitCommandLine(int argc, char** argv, po::variables_map* conf) {
po::options_description opts("Configuration options");
opts.add_options()
("input,i",po::value<string>(),"Corpus of source language sentences")
("weights,w",po::value<string>(),"Input feature weights file")
("decoder_config,c",po::value<string>(), "cdec.ini file");
po::options_description clo("Command line options");
clo.add_options()
("config", po::value<string>(), "Configuration file")
("help,h", "Print this help message and exit");
po::options_description dconfig_options, dcmdline_options;
dconfig_options.add(opts);
dcmdline_options.add(opts).add(clo);
po::store(parse_command_line(argc, argv, dcmdline_options), *conf);
if (conf->count("config")) {
ifstream config((*conf)["config"].as<string>().c_str());
po::store(po::parse_config_file(config, dconfig_options), *conf);
}
po::notify(*conf);
if (conf->count("help") || !conf->count("input") || !conf->count("decoder_config")) {
cerr << dcmdline_options << endl;
return false;
}
return true;
}
void ReadTrainingCorpus(const string& fname, int rank, int size, vector<string>* c, vector<int>* order) {
ReadFile rf(fname);
istream& in = *rf.stream();
string line;
int id = 0;
while(in) {
getline(in, line);
if (!in) break;
if (id % size == rank) {
c->push_back(line);
order->push_back(id);
}
++id;
}
}
static const double kMINUS_EPSILON = -1e-6;
struct TrainingObserver : public DecoderObserver {
void Reset() {
acc_exp.clear();
total_complete = 0;
}
virtual void NotifyDecodingStart(const SentenceMetadata& smeta) {
cur_model_exp.clear();
state = 1;
}
// compute model expectations, denominator of objective
virtual void NotifyTranslationForest(const SentenceMetadata& smeta, Hypergraph* hg) {
assert(state == 1);
state = 2;
const prob_t z = InsideOutside<prob_t,
EdgeProb,
SparseVector<prob_t>,
EdgeFeaturesAndProbWeightFunction>(*hg, &cur_model_exp);
cur_model_exp /= z;
acc_exp += cur_model_exp;
}
virtual void NotifyAlignmentForest(const SentenceMetadata& smeta, Hypergraph* hg) {
cerr << "IGNORING ALIGNMENT FOREST!\n";
}
virtual void NotifyDecodingComplete(const SentenceMetadata& smeta) {
if (state == 2) {
++total_complete;
}
}
void GetExpectations(SparseVector<double>* g) const {
g->clear();
for (SparseVector<prob_t>::const_iterator it = acc_exp.begin(); it != acc_exp.end(); ++it)
g->set_value(it->first, it->second);
}
int total_complete;
SparseVector<prob_t> cur_model_exp;
SparseVector<prob_t> acc_exp;
int state;
};
#ifdef HAVE_MPI
namespace boost { namespace mpi {
template<>
struct is_commutative<std::plus<SparseVector<double> >, SparseVector<double> >
: mpl::true_ { };
} } // end namespace boost::mpi
#endif
int main(int argc, char** argv) {
#ifdef HAVE_MPI
mpi::environment env(argc, argv);
mpi::communicator world;
const int size = world.size();
const int rank = world.rank();
#else
const int size = 1;
const int rank = 0;
#endif
if (size > 1) SetSilent(true); // turn off verbose decoder output
register_feature_functions();
po::variables_map conf;
if (!InitCommandLine(argc, argv, &conf))
return 1;
// load initial weights
Weights weights;
if (conf.count("weights"))
weights.InitFromFile(conf["weights"].as<string>());
vector<string> corpus;
vector<int> ids;
ReadTrainingCorpus(conf["input"].as<string>(), rank, size, &corpus, &ids);
assert(corpus.size() > 0);
vector<string> cdec_ini;
ReadConfig(conf["decoder_config"].as<string>(), &cdec_ini);
istringstream ini;
StoreConfig(cdec_ini, &ini);
Decoder decoder(&ini);
if (decoder.GetConf()["input"].as<string>() != "-") {
cerr << "cdec.ini must not set an input file\n";
return 1;
}
SparseVector<double> x;
weights.InitSparseVector(&x);
TrainingObserver observer;
weights.InitFromVector(x);
vector<double> lambdas;
weights.InitVector(&lambdas);
decoder.SetWeights(lambdas);
observer.Reset();
for (unsigned i = 0; i < corpus.size(); ++i) {
int id = ids[i];
decoder.SetId(id);
decoder.Decode(corpus[i], &observer);
}
SparseVector<double> local_exps, exps;
observer.GetExpectations(&local_exps);
#ifdef HAVE_MPI
reduce(world, local_exps, exps, std::plus<SparseVector<double> >(), 0);
#else
exps.swap(local_exps);
#endif
weights.InitFromVector(exps);
weights.InitVector(&lambdas);
ShowFeatures(lambdas);
return 0;
}
|
