blob: d836ad4fc1b903dc48b1d558b2b2271a23260e34 [file] [log] [blame]
/*
Copyright (C) 2000,2001 Franz Josef Och (RWTH Aachen - Lehrstuhl fuer Informatik VI)
This file is part of GIZA++ ( extension of GIZA ).
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307,
USA.
*/
#ifndef transpair_modelhmm_h_fjo_defined
#define transpair_modelhmm_h_fjo_defined
#include "Array2.h"
#include "defs.h"
#include "Vector.h"
#include "NTables.h"
#include "ATables.h"
#include "TTables.h"
#include "alignment.h"
#include <cmath>
#include "transpair_model2.h"
#include "ForwardBackward.h"
#include "hmm.h"
class transpair_modelhmm : public transpair_model2
{
public:
typedef transpair_modelhmm simpler_transpair_model;
HMMNetwork*net;
transpair_modelhmm(const Vector<WordIndex>&es, const Vector<WordIndex>&fs, const tmodel<COUNT, PROB>&tTable,
const amodel<PROB>&aTable,const amodel<PROB>&,const nmodel<PROB>&,
double, double,const hmm*h)
: transpair_model2(es,fs,tTable,aTable),net(h->makeHMMNetwork(es,fs,0))
{}
~transpair_modelhmm() { delete net; }
int modelnr()const{return 6;}
LogProb scoreOfMove(const alignment&a, WordIndex _new_i, WordIndex j,double=-1.0)const
{
int new_i=_new_i;
LogProb change=1.0;
int old_i=a(j);
if (old_i == new_i)
change=1.0;
else
{
int theJ=j-1;
old_i--;
new_i--;
int jj=j-1;
while(jj>0&&a(jj)==0)
jj--;
int theIPrev= (jj>0)?(a(jj)-1):0;
if( j>1&&a(j-1)==0 )
theIPrev+=l;
if( old_i==-1 ){old_i = theIPrev;if(old_i<int(l))old_i+=l;}
if( new_i==-1 ){new_i = theIPrev;if(new_i<int(l))new_i+=l;}
int theIPrevOld=theIPrev,theIPrevNew=theIPrev;
if( theJ==0 )
{
change*=net->getAlphainit(new_i)/net->getAlphainit(old_i);
}
do
{
if( new_i!=old_i )
{
change*=net->nodeProb(new_i,theJ)/net->nodeProb(old_i,theJ);
}
if( theJ>0)
change*=net->outProb(theJ,theIPrevNew,new_i)/net->outProb(theJ,theIPrevOld,old_i);
theIPrevOld=old_i;
theIPrevNew=new_i;
theJ++;
if( theJ<int(m) && a(theJ+1)==0 )
{
if( new_i<int(l)) new_i+=l;
if( old_i<int(l)) old_i+=l;
}
} while( theJ<int(m) && a(theJ+1)==0 );
if(theJ==int(m))
{
change*=net->getBetainit(new_i)/net->getBetainit(old_i);
}
else
{
new_i=a(theJ+1)-1;
if( new_i==-1)
new_i=theIPrevNew;
change*=net->outProb(theJ,theIPrevNew,new_i)/net->outProb(theJ,theIPrevOld,new_i);
}
}
return change;
}
LogProb scoreOfAlignmentForChange(const alignment&)const
{return -1.0; }
LogProb scoreOfSwap(const alignment&a, WordIndex j1, WordIndex j2,double=-1.0)const
{
return _scoreOfSwap(a,j1,j2);
}
LogProb _scoreOfMove(const alignment&a, WordIndex new_i, WordIndex j,double=-1.0)const
{
alignment b(a);
b.set(j, new_i);
LogProb a_prob=prob_of_target_and_alignment_given_source(a);
LogProb b_prob=prob_of_target_and_alignment_given_source(b);
if( a_prob )
return b_prob/a_prob;
else if( b_prob )
return 1e20;
else
return 1.0;
}
LogProb _scoreOfSwap(const alignment&a, WordIndex j1, WordIndex j2,double=-1.0)const
{
WordIndex aj1=a(j1),aj2=a(j2);
if( aj1==aj2 )
return 1.0;
LogProb a_prob=prob_of_target_and_alignment_given_source(a);
/*alignment b(a);
b.set(j1, a(j2));
b.set(j2, a(j1));
LogProb b_prob=prob_of_target_and_alignment_given_source(b);*/
const_cast<alignment&>(a).set(j1,aj2);
const_cast<alignment&>(a).set(j2,aj1);
LogProb b_prob=prob_of_target_and_alignment_given_source(a);
const_cast<alignment&>(a).set(j1,aj1);
const_cast<alignment&>(a).set(j2,aj2);
if( a_prob )
return b_prob/a_prob;
else if( b_prob )
return 1e20;
else
return 1.0;
}
inline friend ostream&operator<<(ostream&out, const transpair_modelhmm&)
{
return out << "NO-OUTPUT for transpair_modelhmm\n";
}
LogProb prob_of_target_and_alignment_given_source(const alignment&al,bool verbose=0)const
{
double prob=1.0;
int theIPrev=0;
for(unsigned int j=1;j<=m;j++)
{
int theJ=j-1;
int theI=al(j)-1;
if( theI==-1 )
theI=(theIPrev%l)+l;
prob*=net->nodeProb(theI,theJ);
if( verbose )
cout << "NP " << net->nodeProb(theI,theJ) << ' ';
if( j==1 )
{
prob*=net->getAlphainit(theI);
if( verbose )
cout << "AP0 " << net->getAlphainit(theI) << ' ';
}
else
{
prob*=net->outProb(theJ,theIPrev,theI);
if( verbose )
cout << "AP1 " << net->outProb(theJ,theIPrev,theI) << ' ';
}
theIPrev=theI;
if( j==m )
{
prob*=net->getBetainit(theI);
if( verbose )
cout << "AP2 " << net->getBetainit(theI) << ' ';
}
if( verbose )
cout << "j:"<<theJ<<" i:"<<theI << "; ";
}
if( verbose )
cout << '\n';
return prob*net->finalMultiply;
}
void computeScores(const alignment&al,vector<double>&d)const
{
double prob1=1.0,prob2=1.0;
int theIPrev=0;
for(unsigned int j=1;j<=m;j++)
{
int theJ=j-1;
int theI=al(j)-1;
if( theI==-1 )
theI=(theIPrev%l)+l;
prob1*=net->nodeProb(theI,theJ);
if( j==1 )
{
prob2*=net->getAlphainit(theI);
}
else
{
prob2*=net->outProb(theJ,theIPrev,theI);
}
theIPrev=theI;
if( j==m )
{
prob2*=net->getBetainit(theI);
}
}
d.push_back(prob1);
d.push_back(prob2);
}
bool isSubOptimal()const{return 0;}
};
#endif