plComputableObjectList createClusterJointDist( const Variables& variables, const Cluster& cluster ) {
plComputableObjectList cndProbTab;
const Variable& var = variables[*cluster.begin()];
const DistValues probs = createNBUniVarProbTab( var.cardinality() ) ;
const plProbTable probTab( var, probs, true );
cndProbTab *= probTab;
DistValueMat clusterProbTables = createClusterProbTables( variables, cluster );
//./bVariables clustVars;
// clustVars ^= var;
Cluster::const_iterator curr, next;
curr = cluster.begin(); next = curr + 1;
DistValueMat::const_iterator ptIt = clusterProbTables.begin();
while ( next != cluster.end() ) {
plDistributionTable distTab_Xi_Z( variables[*next], variables[*curr] );
for (size_t h = 0; h < variables[*curr].cardinality(); ++h) {
distTab_Xi_Z.push( plProbTable( variables[*next], ptIt->at(h)), (int)h );
}
cndProbTab *= distTab_Xi_Z;
++curr; ++next; ++ptIt;
}
return cndProbTab; //plJointDistribution( clustVars, cndProbTab );
}
plComputableObjectList createNBCndTable(Variables& variables, Variable& latentVar) {
plComputableObjectList dataCndTable;
DistValueMat probTableXZ = createNBProbTables(variables, latentVar);
for (size_t x = 0; x < variables.size(); ++x) {
plDistributionTable distTab_Xi_Z(variables[x], latentVar);
for (size_t h = 0; h < latentVar.cardinality(); ++h) {
distTab_Xi_Z.push( plProbTable(variables[x], probTableXZ[x][h]), (int)h );
}
dataCndTable *= distTab_Xi_Z; // adds the conditional table to result
}
return dataCndTable;
}
/** BN vertex file: see SINGLE format.
* BN distribution file: see above.
*/
void BayesGraphLoad::operator()( std::shared_ptr<Graph> graph,
const std::string labPosFileName,
const std::string vertexFileName,
const std::string distributionFileName,
const std::string cndDataFileName,
const std::string dataFileName
) const {
BOOST_LOG_TRIVIAL(trace) << "begin loading label..." << std::endl;
LabPosMap labPosMap = readLabPos(labPosFileName);
BOOST_LOG_TRIVIAL(trace) << "end loading label..." << std::endl;
std::ifstream vertexFile(vertexFileName.c_str()), distributionFile(distributionFileName.c_str()), dataFile(dataFileName.c_str());
CSVIterator<std::string> vertexLine(vertexFile); ++vertexLine; // skips header.
BOOST_LOG_TRIVIAL(trace) << "begin loading vertices\n";
for( ; vertexLine != CSVIterator<std::string>(); ++vertexLine ) {
size_t id = boost::lexical_cast<size_t>( (*vertexLine)[0] );
int level = boost::lexical_cast<int>( (*vertexLine)[2] );
int cardinality = boost::lexical_cast<int>( (*vertexLine)[3] );
// printf("id: %lu, level: %d, card: %d, label: %s, pos: %d", id, level, cardinality, label.c_str(), position);
std::string label = labPosMap[id].first;
int position = labPosMap[id].second;
create_graph_node( graph, cardinality, label, position, level );
}
Graph& graphRef = *graph;
BOOST_LOG_TRIVIAL(trace) << "end loading vertices: " << boost::num_vertices(graphRef) << "\n\n";
BOOST_LOG_TRIVIAL(trace) << "begin loading dist\n";
CSVIterator<std::string> distributionLine(distributionFile);
while ( distributionLine != CSVIterator<std::string>() ) {
plVariablesConjunction variables; // holds child variables
plComputableObjectList jointDistri;
size_t latentId = boost::lexical_cast<size_t>( (*distributionLine)[BN_LATENT_ID] );
size_t nbrChildren = boost::lexical_cast<size_t>( (*distributionLine)[NBR_CHILDREN] );
Node& latentNode = graphRef[ latentId ]; ++distributionLine; // reads next line.
std::vector< plProbValue > probValuesZ;
for ( size_t latentVal = 0; latentVal < latentNode.variable.cardinality(); ++latentVal) { // loads probability table for the latent var
probValuesZ.push_back( boost::lexical_cast<plProbValue>( (*distributionLine)[latentVal] ) );
}
const plProbTable probTabZ(latentNode.variable, probValuesZ); ++distributionLine;
for ( size_t child = 0; child < nbrChildren; ++child ) {
size_t childId = boost::lexical_cast<size_t>( (*distributionLine)[BN_LATENT_ID] ); ++distributionLine;
Node& childNode = graphRef[ childId ]; variables ^= childNode.variable;
plDistributionTable distTab_Xi_Z ( childNode.variable, latentNode.variable );
for ( size_t latentVal = 0; latentVal < latentNode.variable.cardinality(); ++latentVal ) {
std::vector< plProbValue > probValuesXiZ_vals;
for ( size_t childVal = 0; childVal < childNode.variable.cardinality(); ++childVal ) {
probValuesXiZ_vals.push_back( boost::lexical_cast<plProbValue>( (*distributionLine)[childVal] ) );
}
distTab_Xi_Z.push( plProbTable( childNode.variable, probValuesXiZ_vals), (int)latentVal );
++distributionLine;
}
jointDistri *= distTab_Xi_Z; // adds the conditional table to result
boost::add_edge( latentId, childId, graphRef );
}
auto jd = ( probTabZ * jointDistri );
++distributionLine;
latentNode.set_joint_distribution(
plJointDistribution(latentNode.variable ^ variables, probTabZ * jointDistri) );
}
distributionFile.close();
vertexFile.close();
set_data(*graph, cndDataFileName, dataFileName);
}
