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
|
#ifndef _DTRAIN_KBESTGET_H_
#define _DTRAIN_KBESTGET_H_
#include "kbest.h" // cdec
#include "verbose.h"
#include "viterbi.h"
#include "ff_register.h"
#include "decoder.h"
#include "weights.h"
#include "logval.h"
using namespace std;
namespace dtrain
{
typedef double score_t;
struct ScoredHyp
{
vector<WordID> w;
SparseVector<double> f;
score_t model;
score_t score;
unsigned rank;
};
struct LocalScorer
{
unsigned N_;
vector<score_t> w_;
virtual score_t
Score(vector<WordID>& hyp, vector<WordID>& ref, const unsigned rank)=0;
void Reset() {} // only for approx bleu
inline void
Init(unsigned N, vector<score_t> weights)
{
assert(N > 0);
N_ = N;
if (weights.empty()) for (unsigned i = 0; i < N_; i++) w_.push_back(1./N_);
else w_ = weights;
}
inline score_t
brevity_penaly(const unsigned hyp_len, const unsigned ref_len)
{
if (hyp_len > ref_len) return 1;
return exp(1 - (score_t)ref_len/hyp_len);
}
};
struct HypSampler : public DecoderObserver
{
LocalScorer* scorer_;
vector<WordID>* ref_;
virtual vector<ScoredHyp>* GetSamples()=0;
inline void SetScorer(LocalScorer* scorer) { scorer_ = scorer; }
inline void SetRef(vector<WordID>& ref) { ref_ = &ref; }
};
///////////////////////////////////////////////////////////////////////////////
struct KBestGetter : public HypSampler
{
const unsigned k_;
const string filter_type_;
vector<ScoredHyp> s_;
KBestGetter(const unsigned k, const string filter_type) :
k_(k), filter_type_(filter_type) {}
virtual void
NotifyTranslationForest(const SentenceMetadata& smeta, Hypergraph* hg)
{
KBest(*hg);
}
vector<ScoredHyp>* GetSamples() { return &s_; }
void
KBest(const Hypergraph& forest)
{
if (filter_type_ == "unique") {
KBestUnique(forest);
} else if (filter_type_ == "no") {
KBestNoFilter(forest);
}
}
void
KBestUnique(const Hypergraph& forest)
{
s_.clear();
KBest::KBestDerivations<vector<WordID>, ESentenceTraversal,
KBest::FilterUnique, prob_t, EdgeProb> kbest(forest, k_);
for (unsigned i = 0; i < k_; ++i) {
const KBest::KBestDerivations<vector<WordID>, ESentenceTraversal, KBest::FilterUnique,
prob_t, EdgeProb>::Derivation* d =
kbest.LazyKthBest(forest.nodes_.size() - 1, i);
if (!d) break;
ScoredHyp h;
h.w = d->yield;
h.f = d->feature_values;
h.model = d->score.as_float();
// DEBUG
cout << i+1 << ". "<< h.model << endl;
// /DEBUG
h.rank = i;
h.score = scorer_->Score(h.w, *ref_, i);
s_.push_back(h);
}
}
void
KBestNoFilter(const Hypergraph& forest)
{
s_.clear();
KBest::KBestDerivations<vector<WordID>, ESentenceTraversal> kbest(forest, k_);
for (unsigned i = 0; i < k_; ++i) {
const KBest::KBestDerivations<vector<WordID>, ESentenceTraversal>::Derivation* d =
kbest.LazyKthBest(forest.nodes_.size() - 1, i);
if (!d) break;
ScoredHyp h;
h.w = d->yield;
h.f = d->feature_values;
h.model = d->score.as_float();
h.rank = i;
h.score = scorer_->Score(h.w, *ref_, i);
s_.push_back(h);
}
}
};
} // namespace
#endif
|
