src/lc.cpp

   1 /*  This file is part of libDAI - http://www.libdai.org/
   2  *
   3  *  libDAI is licensed under the terms of the GNU General Public License version
   4  *  2, or (at your option) any later version. libDAI is distributed without any
   5  *  warranty. See the file COPYING for more details.
   6  *
   7  *  Copyright (C) 2006-2009  Joris Mooij  [joris dot mooij at libdai dot org]
   8  *  Copyright (C) 2006-2007  Radboud University Nijmegen, The Netherlands
   9  */
  10
  11
  12 #include <iostream>
  13 #include <algorithm>
  14 #include <map>
  15 #include <set>
  16 #include <dai/lc.h>
  17 #include <dai/util.h>
  18 #include <dai/alldai.h>
  19
  20
  21 namespace dai {
  22
  23
  24 using namespace std;
  25
  26
  27 const char *LC::Name = "LC";
  28
  29
  30 void LC::setProperties( const PropertySet &opts ) {
  31     DAI_ASSERT( opts.hasKey("tol") );
  32     DAI_ASSERT( opts.hasKey("maxiter") );
  33     DAI_ASSERT( opts.hasKey("verbose") );
  34     DAI_ASSERT( opts.hasKey("cavity") );
  35     DAI_ASSERT( opts.hasKey("updates") );
  36
  37     props.tol = opts.getStringAs<double>("tol");
  38     props.maxiter = opts.getStringAs<size_t>("maxiter");
  39     props.verbose = opts.getStringAs<size_t>("verbose");
  40     props.cavity = opts.getStringAs<Properties::CavityType>("cavity");
  41     props.updates = opts.getStringAs<Properties::UpdateType>("updates");
  42     if( opts.hasKey("cavainame") )
  43         props.cavainame = opts.getStringAs<string>("cavainame");
  44     if( opts.hasKey("cavaiopts") )
  45         props.cavaiopts = opts.getStringAs<PropertySet>("cavaiopts");
  46     if( opts.hasKey("reinit") )
  47         props.reinit = opts.getStringAs<bool>("reinit");
  48     if( opts.hasKey("damping") )
  49         props.damping = opts.getStringAs<double>("damping");
  50     else
  51         props.damping = 0.0;
  52 }
  53
  54
  55 PropertySet LC::getProperties() const {
  56     PropertySet opts;
  57     opts.Set( "tol", props.tol );
  58     opts.Set( "maxiter", props.maxiter );
  59     opts.Set( "verbose", props.verbose );
  60     opts.Set( "cavity", props.cavity );
  61     opts.Set( "updates", props.updates );
  62     opts.Set( "cavainame", props.cavainame );
  63     opts.Set( "cavaiopts", props.cavaiopts );
  64     opts.Set( "reinit", props.reinit );
  65     opts.Set( "damping", props.damping );
  66     return opts;
  67 }
  68
  69
  70 string LC::printProperties() const {
  71     stringstream s( stringstream::out );
  72     s << "[";
  73     s << "tol=" << props.tol << ",";
  74     s << "maxiter=" << props.maxiter << ",";
  75     s << "verbose=" << props.verbose << ",";
  76     s << "cavity=" << props.cavity << ",";
  77     s << "updates=" << props.updates << ",";
  78     s << "cavainame=" << props.cavainame << ",";
  79     s << "cavaiopts=" << props.cavaiopts << ",";
  80     s << "reinit=" << props.reinit << ",";
  81     s << "damping=" << props.damping << "]";
  82     return s.str();
  83 }
  84
  85
  86 LC::LC( const FactorGraph & fg, const PropertySet &opts ) : DAIAlgFG(fg), _pancakes(), _cavitydists(), _phis(), _beliefs(), _maxdiff(0.0), _iters(0), props() {
  87     setProperties( opts );
  88
  89     // create pancakes
  90     _pancakes.resize( nrVars() );
  91
  92     // create cavitydists
  93     for( size_t i=0; i < nrVars(); i++ )
  94         _cavitydists.push_back(Factor( delta(i) ));
  95
  96     // create phis
  97     _phis.reserve( nrVars() );
  98     for( size_t i = 0; i < nrVars(); i++ ) {
  99         _phis.push_back( vector<Factor>() );
 100         _phis[i].reserve( nbV(i).size() );
 101         foreach( const Neighbor &I, nbV(i) )
 102             _phis[i].push_back( Factor( factor(I).vars() / var(i) ) );
 103     }
 104
 105     // create beliefs
 106     _beliefs.reserve( nrVars() );
 107     for( size_t i=0; i < nrVars(); i++ )
 108         _beliefs.push_back(Factor(var(i)));
 109 }
 110
 111
 112 string LC::identify() const {
 113     return string(Name) + printProperties();
 114 }
 115
 116
 117 void LC::CalcBelief (size_t i) {
 118     _beliefs[i] = _pancakes[i].marginal(var(i));
 119 }
 120
 121
 122 double LC::CalcCavityDist (size_t i, const std::string &name, const PropertySet &opts) {
 123     Factor Bi;
 124     double maxdiff = 0;
 125
 126     if( props.verbose >= 2 )
 127         cerr << "Initing cavity " << var(i) << "(" << delta(i).size() << " vars, " << delta(i).nrStates() << " states)" << endl;
 128
 129     if( props.cavity == Properties::CavityType::UNIFORM )
 130         Bi = Factor(delta(i));
 131     else {
 132         InfAlg *cav = newInfAlg( name, *this, opts );
 133         cav->makeCavity( i );
 134
 135         if( props.cavity == Properties::CavityType::FULL )
 136             Bi = calcMarginal( *cav, cav->fg().delta(i), props.reinit );
 137         else if( props.cavity == Properties::CavityType::PAIR )
 138             Bi = calcMarginal2ndO( *cav, cav->fg().delta(i), props.reinit );
 139         else if( props.cavity == Properties::CavityType::PAIR2 ) {
 140             vector<Factor> pairbeliefs = calcPairBeliefsNew( *cav, cav->fg().delta(i), props.reinit );
 141             for( size_t ij = 0; ij < pairbeliefs.size(); ij++ )
 142                 Bi *= pairbeliefs[ij];
 143         }
 144         maxdiff = cav->maxDiff();
 145         delete cav;
 146     }
 147     Bi.normalize();
 148     _cavitydists[i] = Bi;
 149
 150     return maxdiff;
 151 }
 152
 153
 154 double LC::InitCavityDists( const std::string &name, const PropertySet &opts ) {
 155     double tic = toc();
 156
 157     if( props.verbose >= 1 ) {
 158         cerr << Name << "::InitCavityDists:  ";
 159         if( props.cavity == Properties::CavityType::UNIFORM )
 160             cerr << "Using uniform initial cavity distributions" << endl;
 161         else if( props.cavity == Properties::CavityType::FULL )
 162             cerr << "Using full " << name << opts << "...";
 163         else if( props.cavity == Properties::CavityType::PAIR )
 164             cerr << "Using pairwise " << name << opts << "...";
 165         else if( props.cavity == Properties::CavityType::PAIR2 )
 166             cerr << "Using pairwise(new) " << name << opts << "...";
 167     }
 168
 169     double maxdiff = 0.0;
 170     for( size_t i = 0; i < nrVars(); i++ ) {
 171         double md = CalcCavityDist(i, name, opts);
 172         if( md > maxdiff )
 173             maxdiff = md;
 174     }
 175
 176     if( props.verbose >= 1 ) {
 177         cerr << Name << "::InitCavityDists used " << toc() - tic << " seconds." << endl;
 178     }
 179
 180     return maxdiff;
 181 }
 182
 183
 184 long LC::SetCavityDists( std::vector<Factor> &Q ) {
 185     if( props.verbose >= 1 )
 186         cerr << Name << "::SetCavityDists:  Setting initial cavity distributions" << endl;
 187     if( Q.size() != nrVars() )
 188         return -1;
 189     for( size_t i = 0; i < nrVars(); i++ ) {
 190         if( _cavitydists[i].vars() != Q[i].vars() ) {
 191             return i+1;
 192         } else
 193             _cavitydists[i] = Q[i];
 194     }
 195     return 0;
 196 }
 197
 198
 199 void LC::init() {
 200     for( size_t i = 0; i < nrVars(); ++i )
 201         foreach( const Neighbor &I, nbV(i) )
 202             if( props.updates == Properties::UpdateType::SEQRND )
 203                 _phis[i][I.iter].randomize();
 204             else
 205                 _phis[i][I.iter].fill(1.0);
 206 }
 207
 208
 209 Factor LC::NewPancake (size_t i, size_t _I, bool & hasNaNs) {
 210     size_t I = nbV(i)[_I];
 211     Factor piet = _pancakes[i];
 212
 213     // recalculate _pancake[i]
 214     VarSet Ivars = factor(I).vars();
 215     Factor A_I;
 216     for( VarSet::const_iterator k = Ivars.begin(); k != Ivars.end(); k++ )
 217         if( var(i) != *k )
 218             A_I *= (_pancakes[findVar(*k)] * factor(I).inverse()).marginal( Ivars / var(i), false );
 219     if( Ivars.size() > 1 )
 220         A_I ^= (1.0 / (Ivars.size() - 1));
 221     Factor A_Ii = (_pancakes[i] * factor(I).inverse() * _phis[i][_I].inverse()).marginal( Ivars / var(i), false );
 222     Factor quot = A_I / A_Ii;
 223     if( props.damping != 0.0 )
 224         quot = (quot^(1.0 - props.damping)) * (_phis[i][_I]^props.damping);
 225
 226     piet *= quot / _phis[i][_I].normalized();
 227     _phis[i][_I] = quot.normalized();
 228
 229     piet.normalize();
 230
 231     if( piet.hasNaNs() ) {
 232         cerr << Name << "::NewPancake(" << i << ", " << _I << "):  has NaNs!" << endl;
 233         hasNaNs = true;
 234     }
 235
 236     return piet;
 237 }
 238
 239
 240 double LC::run() {
 241     if( props.verbose >= 1 )
 242         cerr << "Starting " << identify() << "...";
 243     if( props.verbose >= 2 )
 244         cerr << endl;
 245
 246     double tic = toc();
 247     Diffs diffs(nrVars(), 1.0);
 248
 249     double md = InitCavityDists( props.cavainame, props.cavaiopts );
 250     if( md > _maxdiff )
 251         _maxdiff = md;
 252
 253     for( size_t i = 0; i < nrVars(); i++ ) {
 254         _pancakes[i] = _cavitydists[i];
 255
 256         foreach( const Neighbor &I, nbV(i) ) {
 257             _pancakes[i] *= factor(I);
 258             if( props.updates == Properties::UpdateType::SEQRND )
 259               _pancakes[i] *= _phis[i][I.iter];
 260         }
 261
 262         _pancakes[i].normalize();
 263
 264         CalcBelief(i);
 265     }
 266
 267     vector<Factor> old_beliefs;
 268     for(size_t i=0; i < nrVars(); i++ )
 269         old_beliefs.push_back(belief(i));
 270
 271     bool hasNaNs = false;
 272     for( size_t i=0; i < nrVars(); i++ )
 273         if( _pancakes[i].hasNaNs() ) {
 274             hasNaNs = true;
 275             break;
 276         }
 277     if( hasNaNs ) {
 278         cerr << Name << "::run:  initial _pancakes has NaNs!" << endl;
 279         return 1.0;
 280     }
 281
 282     size_t nredges = nrEdges();
 283     vector<Edge> update_seq;
 284     update_seq.reserve( nredges );
 285     for( size_t i = 0; i < nrVars(); ++i )
 286         foreach( const Neighbor &I, nbV(i) )
 287             update_seq.push_back( Edge( i, I.iter ) );
 288
 289     // do several passes over the network until maximum number of iterations has
 290     // been reached or until the maximum belief difference is smaller than tolerance
 291     for( _iters=0; _iters < props.maxiter && diffs.maxDiff() > props.tol; _iters++ ) {
 292         // Sequential updates
 293         if( props.updates == Properties::UpdateType::SEQRND )
 294             random_shuffle( update_seq.begin(), update_seq.end() );
 295
 296         for( size_t t=0; t < nredges; t++ ) {
 297             size_t i = update_seq[t].first;
 298             size_t _I = update_seq[t].second;
 299             _pancakes[i] = NewPancake( i, _I, hasNaNs);
 300             if( hasNaNs )
 301                 return 1.0;
 302             CalcBelief( i );
 303         }
 304
 305         // compare new beliefs with old ones
 306         for(size_t i=0; i < nrVars(); i++ ) {
 307             diffs.push( dist( belief(i), old_beliefs[i], Prob::DISTLINF ) );
 308             old_beliefs[i] = belief(i);
 309         }
 310
 311         if( props.verbose >= 3 )
 312             cerr << Name << "::run:  maxdiff " << diffs.maxDiff() << " after " << _iters+1 << " passes" << endl;
 313     }
 314
 315     if( diffs.maxDiff() > _maxdiff )
 316         _maxdiff = diffs.maxDiff();
 317
 318     if( props.verbose >= 1 ) {
 319         if( diffs.maxDiff() > props.tol ) {
 320             if( props.verbose == 1 )
 321                 cerr << endl;
 322                 cerr << Name << "::run:  WARNING: not converged within " << props.maxiter << " passes (" << toc() - tic << " seconds)...final maxdiff:" << diffs.maxDiff() << endl;
 323         } else {
 324             if( props.verbose >= 2 )
 325                 cerr << Name << "::run:  ";
 326                 cerr << "converged in " << _iters << " passes (" << toc() - tic << " seconds)." << endl;
 327         }
 328     }
 329
 330     return diffs.maxDiff();
 331 }
 332
 333
 334 } // end of namespace dai