File size: 9,474 Bytes
158b61b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 |
////////////////////////////////////////////////////////////
//
// generate set of target candidates for confusion net
//
////////////////////////////////////////////////////////////
#include <numeric>
#include "moses/Word.h"
#include "moses/Phrase.h"
#include "moses/ConfusionNet.h"
#include "moses/WordsRange.h"
#include "moses/TranslationModel/PhraseDictionaryTree.h"
using namespace Moses;
#if 0
// Generates all tuples from n indexes with ranges 0 to card[j]-1, respectively..
// Input: number of indexes and ranges: ranges[0] ... ranges[num_idx-1]
// Output: number of tuples and monodimensional array of tuples.
// Reference: mixed-radix generation algorithm (D. E. Knuth, TAOCP v. 4.2)
size_t GenerateTuples(unsigned num_idx,unsigned* ranges,unsigned *&tuples)
{
unsigned* single_tuple= new unsigned[num_idx+1];
unsigned num_tuples=1;
for (unsigned k=0; k<num_idx; ++k) {
num_tuples *= ranges[k];
single_tuple[k]=0;
}
tuples=new unsigned[num_idx * num_tuples];
// we need this additional element for the last iteration
single_tuple[num_idx]=0;
unsigned j=0;
for (unsigned n=0; n<num_tuples; ++n) {
memcpy((void *)((tuples + n * num_idx)),(void *)single_tuple,num_idx * sizeof(unsigned));
j=0;
while (single_tuple[j]==ranges[j]-1) {
single_tuple[j]=0;
++j;
}
++single_tuple[j];
}
delete [] single_tuple;
return num_tuples;
}
typedef PhraseDictionaryTree::PrefixPtr PPtr;
typedef std::vector<PPtr> vPPtr;
typedef std::vector<std::vector<Factor const*> > mPhrase;
std::ostream& operator<<(std::ostream& out,const mPhrase& p)
{
for(size_t i=0; i<p.size(); ++i) {
out<<i<<" - ";
for(size_t j=0; j<p[i].size(); ++j)
out<<p[i][j]->ToString()<<" ";
out<<"|";
}
return out;
}
struct State {
vPPtr ptrs;
WordsRange range;
float score;
State() : range(0,0),score(0.0) {}
State(size_t b,size_t e,const vPPtr& v,float sc=0.0) : ptrs(v),range(b,e),score(sc) {}
size_t begin() const {
return range.GetStartPos();
}
size_t end() const {
return range.GetEndPos();
}
float GetScore() const {
return score;
}
};
std::ostream& operator<<(std::ostream& out,const State& s)
{
out<<"["<<s.ptrs.size()<<" ("<<s.begin()<<","<<s.end()<<") "<<s.GetScore()<<"]";
return out;
}
typedef std::map<mPhrase,float> E2Costs;
struct GCData {
const std::vector<PhraseDictionaryTree const*>& pdicts;
const std::vector<std::vector<float> >& weights;
std::vector<FactorType> inF,outF;
size_t distinctOutputFactors;
vPPtr root;
size_t totalTuples,distinctTuples;
GCData(const std::vector<PhraseDictionaryTree const*>& a,
const std::vector<std::vector<float> >& b)
: pdicts(a),weights(b),totalTuples(0),distinctTuples(0) {
CHECK(pdicts.size()==weights.size());
std::set<FactorType> distinctOutFset;
inF.resize(pdicts.size());
outF.resize(pdicts.size());
root.resize(pdicts.size());
for(size_t i=0; i<pdicts.size(); ++i) {
root[i]=pdicts[i]->GetRoot();
inF[i]=pdicts[i]->GetInputFactorType();
outF[i]=pdicts[i]->GetOutputFactorType();
distinctOutFset.insert(pdicts[i]->GetOutputFactorType());
}
distinctOutputFactors=distinctOutFset.size();
}
FactorType OutFT(size_t i) const {
return outF[i];
}
FactorType InFT(size_t i) const {
return inF[i];
}
size_t DistinctOutFactors() const {
return distinctOutputFactors;
}
const vPPtr& GetRoot() const {
return root;
}
};
typedef std::vector<Factor const*> vFactor;
typedef std::vector<std::pair<float,vFactor> > TgtCandList;
typedef std::vector<TgtCandList> OutputFactor2TgtCandList;
typedef std::vector<OutputFactor2TgtCandList*> Len2Cands;
void GeneratePerFactorTgtList(size_t factorType,PPtr pptr,GCData& data,Len2Cands& len2cands)
{
std::vector<FactorTgtCand> cands;
data.pdicts[factorType]->GetTargetCandidates(pptr,cands);
for(std::vector<FactorTgtCand>::const_iterator cand=cands.begin(); cand!=cands.end(); ++cand) {
CHECK(data.weights[factorType].size()==cand->second.size());
float costs=std::inner_product(data.weights[factorType].begin(),
data.weights[factorType].end(),
cand->second.begin(),
0.0);
size_t len=cand->first.size();
if(len>=len2cands.size()) len2cands.resize(len+1,0);
if(!len2cands[len]) len2cands[len]=new OutputFactor2TgtCandList(data.DistinctOutFactors());
OutputFactor2TgtCandList &outf2tcandlist=*len2cands[len];
outf2tcandlist[data.OutFT(factorType)].push_back(std::make_pair(costs,cand->first));
}
}
void GenerateTupleTgtCands(OutputFactor2TgtCandList& tCand,E2Costs& e2costs,GCData& data)
{
// check if candidates are non-empty
bool gotCands=1;
for(size_t j=0; gotCands && j<tCand.size(); ++j)
gotCands &= !tCand[j].empty();
if(gotCands) {
// enumerate tuples
CHECK(data.DistinctOutFactors()==tCand.size());
std::vector<unsigned> radix(data.DistinctOutFactors());
for(size_t i=0; i<tCand.size(); ++i) radix[i]=tCand[i].size();
unsigned *tuples=0;
size_t numTuples=GenerateTuples(radix.size(),&radix[0],tuples);
data.totalTuples+=numTuples;
for(size_t i=0; i<numTuples; ++i) {
mPhrase e(radix.size());
float costs=0.0;
for(size_t j=0; j<radix.size(); ++j) {
CHECK(tuples[radix.size()*i+j]<tCand[j].size());
std::pair<float,vFactor> const& mycand=tCand[j][tuples[radix.size()*i+j]];
e[j]=mycand.second;
costs+=mycand.first;
}
#ifdef DEBUG
bool mismatch=0;
for(size_t j=1; !mismatch && j<e.size(); ++j)
if(e[j].size()!=e[j-1].size()) mismatch=1;
CHECK(mismatch==0);
#endif
std::pair<E2Costs::iterator,bool> p=e2costs.insert(std::make_pair(e,costs));
if(p.second) ++data.distinctTuples;
else {
// entry known, take min of costs, alternative: sum probs
if(costs<p.first->second) p.first->second=costs;
}
}
delete [] tuples;
}
}
void GenerateCandidates_(E2Costs& e2costs,const vPPtr& nextP,GCData& data)
{
Len2Cands len2cands;
// generate candidates for each element of nextP:
for(size_t factorType=0; factorType<nextP.size(); ++factorType)
if(nextP[factorType])
GeneratePerFactorTgtList(factorType,nextP[factorType],data,len2cands);
// for each length: enumerate tuples, compute score, and insert in e2costs
for(size_t len=0; len<len2cands.size(); ++len) if(len2cands[len])
GenerateTupleTgtCands(*len2cands[len],e2costs,data);
}
void GenerateCandidates(const ConfusionNet& src,
const std::vector<PhraseDictionaryTree const*>& pdicts,
const std::vector<std::vector<float> >& weights,
int verbose)
{
GCData data(pdicts,weights);
std::vector<State> stack;
for(size_t i=0; i<src.GetSize(); ++i) stack.push_back(State(i,i,data.GetRoot()));
std::map<WordsRange,E2Costs> cov2E;
// std::cerr<<"start while loop. initial stack size: "<<stack.size()<<"\n";
while(!stack.empty()) {
State curr(stack.back());
stack.pop_back();
//std::cerr<<"processing state "<<curr<<" stack size: "<<stack.size()<<"\n";
CHECK(curr.end()<src.GetSize());
const ConfusionNet::Column &currCol=src[curr.end()];
for(size_t colidx=0; colidx<currCol.size(); ++colidx) {
const Word& w=currCol[colidx].first;
vPPtr nextP(curr.ptrs);
for(size_t j=0; j<nextP.size(); ++j)
nextP[j]=pdicts[j]->Extend(nextP[j],
w.GetFactor(data.InFT(j))->GetString());
bool valid=1;
for(size_t j=0; j<nextP.size(); ++j) if(!nextP[j]) {
valid=0;
break;
}
if(valid) {
if(curr.end()+1<src.GetSize())
stack.push_back(State(curr.begin(),curr.end()+1,nextP,
curr.GetScore()+currCol[colidx].second));
E2Costs &e2costs=cov2E[WordsRange(curr.begin(),curr.end()+1)];
GenerateCandidates_(e2costs,nextP,data);
}
}
// check if there are translations of one-word phrases ...
//if(curr.begin()==curr.end() && tCand.empty()) {}
} // end while(!stack.empty())
if(verbose) {
// print statistics for debugging purposes
std::cerr<<"tuple stats: total: "<<data.totalTuples
<<" distinct: "<<data.distinctTuples<<" ("
<<(data.distinctTuples/(0.01*data.totalTuples))
<<"%)\n";
std::cerr<<"per coverage set:\n";
for(std::map<WordsRange,E2Costs>::const_iterator i=cov2E.begin();
i!=cov2E.end(); ++i) {
std::cerr<<i->first<<" -- distinct cands: "
<<i->second.size()<<"\n";
}
std::cerr<<"\n\n";
}
if(verbose>10) {
std::cerr<<"full list:\n";
for(std::map<WordsRange,E2Costs>::const_iterator i=cov2E.begin();
i!=cov2E.end(); ++i) {
std::cerr<<i->first<<" -- distinct cands: "
<<i->second.size()<<"\n";
for(E2Costs::const_iterator j=i->second.begin(); j!=i->second.end(); ++j)
std::cerr<<j->first<<" -- "<<j->second<<"\n";
}
}
}
#else
void GenerateCandidates(const ConfusionNet&,
const std::vector<PhraseDictionaryTree const*>&,
const std::vector<std::vector<float> >&,
int)
{
std::cerr<<"ERROR: GenerateCandidates is currently broken\n";
}
#endif
|