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
|
#include "lm/model.hh"
#include <stdlib.h>
#define BOOST_TEST_MODULE ModelTest
#include <boost/test/unit_test.hpp>
#include <boost/test/floating_point_comparison.hpp>
namespace lm {
namespace ngram {
namespace {
#define StartTest(word, ngram, score) \
ret = model.FullScore( \
state, \
model.GetVocabulary().Index(word), \
out);\
BOOST_CHECK_CLOSE(score, ret.prob, 0.001); \
BOOST_CHECK_EQUAL(static_cast<unsigned int>(ngram), ret.ngram_length); \
BOOST_CHECK_EQUAL(std::min<unsigned char>(ngram, 5 - 1), out.valid_length_);
#define AppendTest(word, ngram, score) \
StartTest(word, ngram, score) \
state = out;
template <class M> void Starters(const M &model) {
FullScoreReturn ret;
Model::State state(model.BeginSentenceState());
Model::State out;
StartTest("looking", 2, -0.4846522);
// , probability plus <s> backoff
StartTest(",", 1, -1.383514 + -0.4149733);
// <unk> probability plus <s> backoff
StartTest("this_is_not_found", 0, -1.995635 + -0.4149733);
}
template <class M> void Continuation(const M &model) {
FullScoreReturn ret;
Model::State state(model.BeginSentenceState());
Model::State out;
AppendTest("looking", 2, -0.484652);
AppendTest("on", 3, -0.348837);
AppendTest("a", 4, -0.0155266);
AppendTest("little", 5, -0.00306122);
State preserve = state;
AppendTest("the", 1, -4.04005);
AppendTest("biarritz", 1, -1.9889);
AppendTest("not_found", 0, -2.29666);
AppendTest("more", 1, -1.20632);
AppendTest(".", 2, -0.51363);
AppendTest("</s>", 3, -0.0191651);
state = preserve;
AppendTest("more", 5, -0.00181395);
AppendTest("loin", 5, -0.0432557);
}
#define StatelessTest(word, provide, ngram, score) \
ret = model.FullScoreForgotState(indices + num_words - word, indices + num_words - word + provide, indices[num_words - word - 1], state); \
BOOST_CHECK_CLOSE(score, ret.prob, 0.001); \
BOOST_CHECK_EQUAL(static_cast<unsigned int>(ngram), ret.ngram_length); \
model.GetState(indices + num_words - word, indices + num_words - word + provide, before); \
ret = model.FullScore(before, indices[num_words - word - 1], out); \
BOOST_CHECK(state == out); \
BOOST_CHECK_CLOSE(score, ret.prob, 0.001); \
BOOST_CHECK_EQUAL(static_cast<unsigned int>(ngram), ret.ngram_length);
template <class M> void Stateless(const M &model) {
const char *words[] = {"<s>", "looking", "on", "a", "little", "the", "biarritz", "not_found", "more", ".", "</s>"};
const size_t num_words = sizeof(words) / sizeof(const char*);
// Silience "array subscript is above array bounds" when extracting end pointer.
WordIndex indices[num_words + 1];
for (unsigned int i = 0; i < num_words; ++i) {
indices[num_words - 1 - i] = model.GetVocabulary().Index(words[i]);
}
FullScoreReturn ret;
State state, out, before;
ret = model.FullScoreForgotState(indices + num_words - 1, indices + num_words, indices[num_words - 2], state);
BOOST_CHECK_CLOSE(-0.484652, ret.prob, 0.001);
StatelessTest(1, 1, 2, -0.484652);
// looking
StatelessTest(1, 2, 2, -0.484652);
// on
AppendTest("on", 3, -0.348837);
StatelessTest(2, 3, 3, -0.348837);
StatelessTest(2, 2, 3, -0.348837);
StatelessTest(2, 1, 2, -0.4638903);
// a
StatelessTest(3, 4, 4, -0.0155266);
// little
AppendTest("little", 5, -0.00306122);
StatelessTest(4, 5, 5, -0.00306122);
// the
AppendTest("the", 1, -4.04005);
StatelessTest(5, 5, 1, -4.04005);
// No context of the.
StatelessTest(5, 0, 1, -1.687872);
// biarritz
StatelessTest(6, 1, 1, -1.9889);
// not found
StatelessTest(7, 1, 0, -2.29666);
StatelessTest(7, 0, 0, -1.995635);
WordIndex unk[1];
unk[0] = 0;
model.GetState(unk, unk + 1, state);
BOOST_CHECK_EQUAL(0, state.valid_length_);
}
//const char *kExpectedOrderProbing[] = {"<unk>", ",", ".", "</s>", "<s>", "a", "also", "beyond", "biarritz", "call", "concerns", "consider", "considering", "for", "higher", "however", "i", "immediate", "in", "is", "little", "loin", "look", "looking", "more", "on", "screening", "small", "the", "to", "watch", "watching", "what", "would"};
class ExpectEnumerateVocab : public EnumerateVocab {
public:
ExpectEnumerateVocab() {}
void Add(WordIndex index, const StringPiece &str) {
BOOST_CHECK_EQUAL(seen.size(), index);
seen.push_back(std::string(str.data(), str.length()));
}
void Check(const base::Vocabulary &vocab) {
BOOST_CHECK_EQUAL(34ULL, seen.size());
BOOST_REQUIRE(!seen.empty());
BOOST_CHECK_EQUAL("<unk>", seen[0]);
for (WordIndex i = 0; i < seen.size(); ++i) {
BOOST_CHECK_EQUAL(i, vocab.Index(seen[i]));
}
}
void Clear() {
seen.clear();
}
std::vector<std::string> seen;
};
template <class ModelT> void LoadingTest() {
Config config;
config.arpa_complain = Config::NONE;
config.messages = NULL;
ExpectEnumerateVocab enumerate;
config.enumerate_vocab = &enumerate;
config.probing_multiplier = 2.0;
ModelT m("test.arpa", config);
enumerate.Check(m.GetVocabulary());
Starters(m);
Continuation(m);
Stateless(m);
}
BOOST_AUTO_TEST_CASE(probing) {
LoadingTest<Model>();
}
BOOST_AUTO_TEST_CASE(sorted) {
LoadingTest<SortedModel>();
}
BOOST_AUTO_TEST_CASE(trie) {
LoadingTest<TrieModel>();
}
template <class ModelT> void BinaryTest() {
Config config;
config.write_mmap = "test.binary";
config.messages = NULL;
ExpectEnumerateVocab enumerate;
config.enumerate_vocab = &enumerate;
{
ModelT copy_model("test.arpa", config);
enumerate.Check(copy_model.GetVocabulary());
enumerate.Clear();
}
config.write_mmap = NULL;
ModelT binary("test.binary", config);
enumerate.Check(binary.GetVocabulary());
Starters(binary);
Continuation(binary);
Stateless(binary);
unlink("test.binary");
}
BOOST_AUTO_TEST_CASE(write_and_read_probing) {
BinaryTest<Model>();
}
BOOST_AUTO_TEST_CASE(write_and_read_sorted) {
BinaryTest<SortedModel>();
}
BOOST_AUTO_TEST_CASE(write_and_read_trie) {
BinaryTest<TrieModel>();
}
} // namespace
} // namespace ngram
} // namespace lm
|
