concordia-server/mgiza-aligner/mgiza/mgizapp/src/logprob.cpp

150 lines
4.6 KiB
C++
Raw Normal View History

2017-01-21 17:07:36 +01:00
/*
EGYPT Toolkit for Statistical Machine Translation
Written by Yaser Al-Onaizan, Jan Curin, Michael Jahr, Kevin Knight, John Lafferty, Dan Melamed, David Purdy, Franz Och, Noah Smith, and David Yarowsky.
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.
*/
// Routines to perform integer exponential arithmetic.
// A number x is represented as n, where x = b**n.
// It is assumed that b > 1, something like b = 1.001;
#include "logprob.h"
#include <cstdlib>
#include <cstdio>
#include <iostream>
#include <fstream>
#include <string>
double *LogProb::ntof = NULL; // Tables will be initialized
int *LogProb::addtbl = NULL; // in Initialize function.
int *LogProb::subtbl = NULL; //
const int LogProb::max_2byte_integer = 32767;
const int LogProb::min_2byte_integer = -32768;
const double LogProb::b = 1.001; // a logarithm basis
const double LogProb::logb2 = log(b);
//const int LogProb::nmax = round(78.0E0 * log(1.0E1) / logb2);
const int LogProb::nmax = round(300.0E0 * log(1.0E1) / logb2);
const int LogProb::nmin = -nmax;
const int LogProb::tblbnd = round(log((b-1.0E0)/2.0E0)/logb2);
const int LogProb::zeron = round(pow((double)-2, (double)23));
const int LogProb::onen = 0;
const int LogProb::infn = onen - zeron;
const int LogProb::initialized = LogProb::Initialize();
const LogProb LogProb::zero(0);
const LogProb LogProb::one(1);
const LogProb LogProb::minus2(1e-2);
const LogProb LogProb::minus4(1e-4);
const LogProb LogProb::minus6(1e-6);
const LogProb LogProb::minus8(1e-8);
const LogProb LogProb::minus10(1e-10);
const LogProb LogProb::minus12(1e-12);
const LogProb LogProb::minus14(1e-14);
const LogProb LogProb::minus16(1e-16);
// static table initialization function
int LogProb::Initialize()
{
int nbytes = sizeof(double)*(nmax-nmin+1) + sizeof(int)*(0-tblbnd+1);
std::cerr << nbytes << " bytes used for LogProb tables (C++ version)\n";
ntof = new double[nmax-nmin+1];
addtbl = new int[-tblbnd+1];
subtbl = new int[-tblbnd+1];
// char filename[257];
// string filename ;
// ifstream ifs;
// ifs.open(filename.c_str());
// if (!ifs)
// {
int i;
std::cerr << "Building integer logs conversion tables\n";
ntof[0] = 0 ;
for (i=nmin+1; i<=nmax; ++i) {
double x = i;
ntof[i-nmin] = exp(x*logb2);
}
for (i=tblbnd; i<=0; ++i) {
double x = 1.0 + pow(b, i);
addtbl[i-tblbnd] = round(log(x)/logb2);
}
double sqrtb = exp(0.5*logb2);
for (i=0; i<=-tblbnd; ++i) {
double x = sqrtb * pow(b, i) - 1.0;
subtbl[i] = round(log(x)/logb2);
}
// if (toolsRoot)
// {
// ofstream ofs(filename.c_str());
// if (!ofs)
// cerr << "Could not write LogProb data to " << filename << endl;
// else
// {
// ofs.write((const char *)ntof, sizeof(double) * (nmax-nmin+1));
// ofs.write((const char *)addtbl, sizeof(int) * (-tblbnd+1));
// ofs.write((const char *)subtbl, sizeof(int) * (-tblbnd+1));
// }
// }
// }
// else
// {
// ifs.read((char *)ntof, sizeof(double) * (nmax - nmin + 1));
// ifs.read((char *)addtbl, sizeof(int) * (-tblbnd+1));
// ifs.read((char *)subtbl, sizeof(int) * (-tblbnd+1));
// }
return 1;
}
void LogProb::FreeTables()
{
delete [] addtbl;
delete [] subtbl;
delete [] ntof;
}
//---------------------------------------------------------------------------
// Aritmetic operators
//---------------------------------------------------------------------------
// Subtract two logarithm numbers. Use the following method:
// b**n - b**m = b**m( b**(n-m) - 1 ), assuming n >= m.
LogProb& LogProb::operator-=(const LogProb &subs)
{
if (subs.logr == zeron)
return *this;
int a = logr - subs.logr;
if (a <= 0) {
if (a < 0) {
std::cerr << "WARNING(logprob): Invalid arguments to nsub" <<(*this)<< " " << subs << std::endl;
//abort();
}
logr = zeron;
return *this;
}
if (a > -tblbnd)
return *this;
logr = subs.logr + subtbl[a];
return *this;
}