diff options
Diffstat (limited to 'decoder')
-rw-r--r-- | decoder/apply_models.cc | 21 | ||||
-rw-r--r-- | decoder/bottom_up_parser.cc | 11 | ||||
-rw-r--r-- | decoder/decoder.cc | 52 | ||||
-rw-r--r-- | decoder/decoder.h | 2 | ||||
-rw-r--r-- | decoder/hg_intersect.cc | 3 | ||||
-rw-r--r-- | decoder/inside_outside.h | 4 | ||||
-rw-r--r-- | decoder/scfg_translator.cc | 9 | ||||
-rw-r--r-- | decoder/sentence_metadata.h | 2 |
8 files changed, 60 insertions, 44 deletions
diff --git a/decoder/apply_models.cc b/decoder/apply_models.cc index 60b6f498..aec31a4f 100644 --- a/decoder/apply_models.cc +++ b/decoder/apply_models.cc @@ -13,6 +13,7 @@ #include <boost/functional/hash.hpp> +#include "verbose.h" #include "hg.h" #include "ff.h" @@ -170,7 +171,7 @@ public: out(*o), D(in.nodes_.size()), pop_limit_(pop_limit) { - cerr << " Applying feature functions (cube pruning, pop_limit = " << pop_limit_ << ')' << endl; + if (!SILENT) cerr << " Applying feature functions (cube pruning, pop_limit = " << pop_limit_ << ')' << endl; node_states_.reserve(kRESERVE_NUM_NODES); } @@ -181,14 +182,16 @@ public: int every = 1; if (num_nodes > 100) every = 10; assert(in.nodes_[pregoal].out_edges_.size() == 1); - cerr << " "; + if (!SILENT) cerr << " "; for (int i = 0; i < in.nodes_.size(); ++i) { - if (i % every == 0) cerr << '.'; + if (!SILENT && i % every == 0) cerr << '.'; KBest(i, i == goal_id); } - cerr << endl; - cerr << " Best path: " << log(D[goal_id].front()->vit_prob_) - << "\t" << log(D[goal_id].front()->est_prob_) << endl; + if (!SILENT) { + cerr << endl; + cerr << " Best path: " << log(D[goal_id].front()->vit_prob_) + << "\t" << log(D[goal_id].front()->est_prob_) << endl; + } out.PruneUnreachable(D[goal_id].front()->node_index_); FreeAll(); } @@ -379,12 +382,12 @@ struct NoPruningRescorer { int every = 1; if (num_nodes > 100) every = 10; assert(in.nodes_[pregoal].out_edges_.size() == 1); - cerr << " "; + if (!SILENT) cerr << " "; for (int i = 0; i < in.nodes_.size(); ++i) { - if (i % every == 0) cerr << '.'; + if (!SILENT && i % every == 0) cerr << '.'; ProcessOneNode(i, i == goal_id); } - cerr << endl; + if (!SILENT) cerr << endl; } private: diff --git a/decoder/bottom_up_parser.cc b/decoder/bottom_up_parser.cc index 88514b82..2d945222 100644 --- a/decoder/bottom_up_parser.cc +++ b/decoder/bottom_up_parser.cc @@ -10,6 +10,7 @@ #include "hg.h" #include "array2d.h" #include "tdict.h" +#include "verbose.h" using namespace std; @@ -17,7 +18,7 @@ struct ParserStats { ParserStats() : active_items(), passive_items() {} void Reset() { active_items=0; passive_items=0; } void Report() { - cerr << " ACTIVE ITEMS: " << active_items << "\tPASSIVE ITEMS: " << passive_items << endl; + if (!SILENT) cerr << " ACTIVE ITEMS: " << active_items << "\tPASSIVE ITEMS: " << passive_items << endl; } int active_items; int passive_items; @@ -172,7 +173,7 @@ PassiveChart::PassiveChart(const string& goal, for (int i = 0; i < grammars_.size(); ++i) act_chart_[i] = new ActiveChart(forest, *this); if (!kGOAL) kGOAL = TD::Convert("Goal") * -1; - cerr << " Goal category: [" << goal << ']' << endl; + if (!SILENT) cerr << " Goal category: [" << goal << ']' << endl; } void PassiveChart::ApplyRule(const int i, @@ -241,9 +242,9 @@ bool PassiveChart::Parse() { for (int gi = 0; gi < grammars_.size(); ++gi) act_chart_[gi]->SeedActiveChart(*grammars_[gi]); - cerr << " "; + if (!SILENT) cerr << " "; for (int l=1; l<input_.size()+1; ++l) { - cerr << '.'; + if (!SILENT) cerr << '.'; for (int i=0; i<input_.size() + 1 - l; ++i) { int j = i + l; for (int gi = 0; gi < grammars_.size(); ++gi) { @@ -278,7 +279,7 @@ bool PassiveChart::Parse() { } } } - cerr << endl; + if (!SILENT) cerr << endl; if (GoalFound()) forest_->PruneUnreachable(forest_->nodes_.size() - 1); diff --git a/decoder/decoder.cc b/decoder/decoder.cc index 2c53ff62..e0967e40 100644 --- a/decoder/decoder.cc +++ b/decoder/decoder.cc @@ -10,6 +10,7 @@ #include "filelib.h" #include "fdict.h" #include "timing_stats.h" +#include "verbose.h" #include "translator.h" #include "phrasebased_translator.h" @@ -51,6 +52,8 @@ static bool verbose_feature_functions=true; namespace Hack { void MaxTrans(const Hypergraph& in, int beam_size); } namespace NgramCache { void Clear(); } +DecoderObserver::~DecoderObserver() {} +void DecoderObserver::NotifyDecodingStart(const SentenceMetadata& smeta) {} void DecoderObserver::NotifySourceParseFailure(const SentenceMetadata&) {} void DecoderObserver::NotifyTranslationForest(const SentenceMetadata&, Hypergraph*) {} void DecoderObserver::NotifyAlignmentFailure(const SentenceMetadata&) {} @@ -639,15 +642,17 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { if (sgml.find("id") != sgml.end()) sent_id = atoi(sgml["id"].c_str()); - cerr << "\nINPUT: "; - if (buf.size() < 100) - cerr << buf << endl; - else { - size_t x = buf.rfind(" ", 100); - if (x == string::npos) x = 100; - cerr << buf.substr(0, x) << " ..." << endl; + if (!SILENT) { + cerr << "\nINPUT: "; + if (buf.size() < 100) + cerr << buf << endl; + else { + size_t x = buf.rfind(" ", 100); + if (x == string::npos) x = 100; + cerr << buf.substr(0, x) << " ..." << endl; + } + cerr << " id = " << sent_id << endl; } - cerr << " id = " << sent_id << endl; string to_translate; Lattice ref; ParseTranslatorInputLattice(buf, &to_translate, &ref); @@ -655,6 +660,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { //FIXME: should get the avg. or max source length of the input lattice (like Lattice::dist_(start,end)); but this is only used to scale beam parameters (optionally) anyway so fidelity isn't important. const bool has_ref = ref.size() > 0; SentenceMetadata smeta(sent_id, ref); + o->NotifyDecodingStart(smeta); Hypergraph forest; // -LM forest translator->ProcessMarkupHints(sgml); Timer t("Translation"); @@ -664,7 +670,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { translator->SentenceComplete(); if (!translation_successful) { - cerr << " NO PARSE FOUND.\n"; + if (!SILENT) cerr << " NO PARSE FOUND.\n"; o->NotifySourceParseFailure(smeta); o->NotifyDecodingComplete(smeta); return false; @@ -672,7 +678,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { const bool show_tree_structure=conf.count("show_tree_structure"); const bool show_features=conf.count("show_features"); - forest_stats(forest," -LM forest",show_tree_structure,show_features,feature_weights,oracle.show_derivation); + if (!SILENT) forest_stats(forest," -LM forest",show_tree_structure,show_features,feature_weights,oracle.show_derivation); if (conf.count("show_expected_length")) { const PRPair<double, double> res = Inside<PRPair<double, double>, @@ -714,7 +720,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { &lm_forest); forest.swap(lm_forest); forest.Reweight(feature_weights); - forest_stats(forest," +LM forest",show_tree_structure,show_features,feature_weights,oracle.show_derivation); + if (!SILENT) forest_stats(forest," +LM forest",show_tree_structure,show_features,feature_weights,oracle.show_derivation); } maybe_prune(forest,conf,"beam_prune","density_prune","+LM",srclen); @@ -729,22 +735,22 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { Hypergraph fsa_forest; assert(fsa_ffs.size()==1); ApplyFsaBy cfg(str("apply_fsa_by",conf),pop_limit); - cerr << "FSA rescoring with "<<cfg<<" "<<fsa_ffs[0]->describe()<<endl; + if (!SILENT) cerr << "FSA rescoring with "<<cfg<<" "<<fsa_ffs[0]->describe()<<endl; ApplyFsaModels(hgcfg,smeta,*fsa_ffs[0],feature_weights,cfg,&fsa_forest); forest.swap(fsa_forest); forest.Reweight(feature_weights); - forest_stats(forest," +FSA forest",show_tree_structure,show_features,feature_weights,oracle.show_derivation); + if (!SILENT) forest_stats(forest," +FSA forest",show_tree_structure,show_features,feature_weights,oracle.show_derivation); } // Oracle Rescoring if(get_oracle_forest) { Oracle oc=oracle.ComputeOracle(smeta,&forest,FeatureVector(feature_weights),10,conf["forest_output"].as<std::string>()); - cerr << " +Oracle BLEU forest (nodes/edges): " << forest.nodes_.size() << '/' << forest.edges_.size() << endl; - cerr << " +Oracle BLEU (paths): " << forest.NumberOfPaths() << endl; + if (!SILENT) cerr << " +Oracle BLEU forest (nodes/edges): " << forest.nodes_.size() << '/' << forest.edges_.size() << endl; + if (!SILENT) cerr << " +Oracle BLEU (paths): " << forest.NumberOfPaths() << endl; oc.hope.Print(cerr," +Oracle BLEU"); oc.fear.Print(cerr," -Oracle BLEU"); //Add 1-best translation (trans) to psuedo-doc vectors - oracle.IncludeLastScore(&cerr); + if (!SILENT) oracle.IncludeLastScore(&cerr); } o->NotifyTranslationForest(smeta, &forest); @@ -752,7 +758,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { if (conf.count("forest_output") && !has_ref) { ForestWriter writer(str("forest_output",conf), sent_id); if (FileExists(writer.fname_)) { - cerr << " Unioning...\n"; + if (!SILENT) cerr << " Unioning...\n"; Hypergraph new_hg; { ReadFile rf(writer.fname_); @@ -812,15 +818,15 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { HypergraphIO::WriteAsCFG(forest); if (has_ref) { if (HG::Intersect(ref, &forest)) { - cerr << " Constr. forest (nodes/edges): " << forest.nodes_.size() << '/' << forest.edges_.size() << endl; - cerr << " Constr. forest (paths): " << forest.NumberOfPaths() << endl; + if (!SILENT) cerr << " Constr. forest (nodes/edges): " << forest.nodes_.size() << '/' << forest.edges_.size() << endl; + if (!SILENT) cerr << " Constr. forest (paths): " << forest.NumberOfPaths() << endl; if (crf_uniform_empirical) { - cerr << " USING UNIFORM WEIGHTS\n"; + if (!SILENT) cerr << " USING UNIFORM WEIGHTS\n"; for (int i = 0; i < forest.edges_.size(); ++i) forest.edges_[i].edge_prob_=prob_t::One(); } else { forest.Reweight(feature_weights); - cerr << " Constr. VitTree: " << ViterbiFTree(forest) << endl; + if (!SILENT) cerr << " Constr. VitTree: " << ViterbiFTree(forest) << endl; } if (conf.count("show_partition")) { const prob_t z = Inside<prob_t, EdgeProb>(forest); @@ -830,7 +836,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { if (conf.count("forest_output")) { ForestWriter writer(str("forest_output",conf), sent_id); if (FileExists(writer.fname_)) { - cerr << " Unioning...\n"; + if (!SILENT) cerr << " Unioning...\n"; Hypergraph new_hg; { ReadFile rf(writer.fname_); @@ -893,7 +899,7 @@ bool DecoderImpl::Decode(const string& input, DecoderObserver* o) { if (conf.count("graphviz")) forest.PrintGraphviz(); } else { o->NotifyAlignmentFailure(smeta); - cerr << " REFERENCE UNREACHABLE.\n"; + if (!SILENT) cerr << " REFERENCE UNREACHABLE.\n"; if (write_gradient) { cout << endl << flush; } diff --git a/decoder/decoder.h b/decoder/decoder.h index 5dd6e1aa..2a1a43ce 100644 --- a/decoder/decoder.h +++ b/decoder/decoder.h @@ -12,6 +12,8 @@ struct Hypergraph; struct DecoderImpl; struct DecoderObserver { + virtual ~DecoderObserver(); + virtual void NotifyDecodingStart(const SentenceMetadata& smeta); virtual void NotifySourceParseFailure(const SentenceMetadata& smeta); virtual void NotifyTranslationForest(const SentenceMetadata& smeta, Hypergraph* hg); virtual void NotifyAlignmentFailure(const SentenceMetadata& semta); diff --git a/decoder/hg_intersect.cc b/decoder/hg_intersect.cc index a3a2cbd9..81350cb0 100644 --- a/decoder/hg_intersect.cc +++ b/decoder/hg_intersect.cc @@ -5,6 +5,7 @@ #include "fast_lexical_cast.hpp" #include <boost/functional/hash.hpp> +#include "verbose.h" #include "tdict.h" #include "hg.h" #include "trule.h" @@ -50,7 +51,7 @@ struct RuleFilter { }; static bool FastLinearIntersect(const Lattice& target, Hypergraph* hg) { - cerr << " Fast linear-chain intersection...\n"; + if (!SILENT) cerr << " Fast linear-chain intersection...\n"; vector<bool> prune(hg->edges_.size(), false); set<int> cov; map<const TRule*, TRulePtr> inverted_rules; diff --git a/decoder/inside_outside.h b/decoder/inside_outside.h index e6289176..dc96f1a9 100644 --- a/decoder/inside_outside.h +++ b/decoder/inside_outside.h @@ -1,5 +1,5 @@ -#ifndef _INSIDE_H_ -#define _INSIDE_H_ +#ifndef _INSIDE_OUTSIDE_H_ +#define _INSIDE_OUTSIDE_H_ #include <vector> #include <algorithm> diff --git a/decoder/scfg_translator.cc b/decoder/scfg_translator.cc index 08276c71..4e6cc226 100644 --- a/decoder/scfg_translator.cc +++ b/decoder/scfg_translator.cc @@ -14,6 +14,7 @@ #include "sentence_metadata.h" #include "tdict.h" #include "viterbi.h" +#include "verbose.h" #define foreach BOOST_FOREACH #define reverse_foreach BOOST_REVERSE_FOREACH @@ -100,7 +101,7 @@ struct SCFGTranslatorImpl { LatticeTools::ConvertTextOrPLF(input, &lattice); smeta->SetSourceLength(lattice.size()); if (add_pass_through_rules){ - cerr << "Adding pass through grammar" << endl; + if (!SILENT) cerr << "Adding pass through grammar" << endl; PassThroughGrammar* g = new PassThroughGrammar(lattice, default_nt, ctf_iterations_); g->SetGrammarName("PassThrough"); glist.push_back(GrammarPtr(g)); @@ -109,13 +110,13 @@ struct SCFGTranslatorImpl { if(printGrammarsUsed) cerr << "Using grammar::" << glist[gi]->GetGrammarName() << endl; } - cerr << "First pass parse... " << endl; + if (!SILENT) cerr << "First pass parse... " << endl; ExhaustiveBottomUpParser parser(goal, glist); if (!parser.Parse(lattice, forest)){ - cerr << "parse failed." << endl; + if (!SILENT) cerr << "parse failed." << endl; return false; } else { - cerr << "parse succeeded." << endl; + if (!SILENT) cerr << "parse succeeded." << endl; } forest->Reweight(weights); if (use_ctf_) { diff --git a/decoder/sentence_metadata.h b/decoder/sentence_metadata.h index 593019c8..c9a78dd3 100644 --- a/decoder/sentence_metadata.h +++ b/decoder/sentence_metadata.h @@ -13,6 +13,8 @@ struct SentenceMetadata { trg_len_(ref.size()), ref_(has_reference_ ? &ref : NULL) {} + int GetSentenceId() const { return sent_id_; } + // this should be called by the Translator object after // it has parsed the source void SetSourceLength(int sl) { src_len_ = sl; } |