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
|
#ifndef _DTRAIN_SAMPLE_NET_H_
#define _DTRAIN_SAMPLE_NET_H_
#include "kbest.h"
#include "score_net_interface.h"
namespace dtrain
{
struct ScoredKbest : public DecoderObserver
{
const size_t k_;
size_t feature_count_, effective_sz_;
vector<ScoredHyp> samples_;
PerSentenceBleuScorer* scorer_;
vector<Ngrams>* ref_ngs_;
vector<size_t>* ref_ls_;
bool dont_score;
string viterbiTreeStr_, viterbiRules_;
ScoredKbest(const size_t k, PerSentenceBleuScorer* scorer) :
k_(k), scorer_(scorer), dont_score(false) {}
virtual void
NotifyTranslationForest(const SentenceMetadata& /*smeta*/, Hypergraph* hg)
{
samples_.clear(); effective_sz_ = feature_count_ = 0;
KBest::KBestDerivations<vector<WordID>, ESentenceTraversal,
KBest::FilterUnique, prob_t, EdgeProb> kbest(*hg, k_);
for (size_t i = 0; i < k_; ++i) {
const KBest::KBestDerivations<vector<WordID>, ESentenceTraversal,
KBest::FilterUnique, prob_t, EdgeProb>::Derivation* d =
kbest.LazyKthBest(hg->nodes_.size() - 1, i);
if (!d) break;
ScoredHyp h;
h.w = d->yield;
h.f = d->feature_values;
h.model = log(d->score);
h.rank = i;
if (!dont_score)
h.gold = scorer_->Score(h.w, *ref_ngs_, *ref_ls_);
samples_.push_back(h);
effective_sz_++;
feature_count_ += h.f.size();
viterbiTreeStr_ = hg->show_viterbi_tree(false);
ostringstream ss;
ViterbiRules(*hg, &ss);
viterbiRules_ = ss.str();
}
}
vector<ScoredHyp>* GetSamples() { return &samples_; }
inline void SetReference(vector<Ngrams>& ngs, vector<size_t>& ls)
{
ref_ngs_ = &ngs;
ref_ls_ = &ls;
}
inline size_t GetFeatureCount() { return feature_count_; }
inline size_t GetSize() { return effective_sz_; }
inline string GetViterbiTreeStr() { return viterbiTreeStr_; }
inline string GetViterbiRules() { return viterbiRules_; }
};
} // namespace
#endif
|
