double BayesNet::query( const DataPoint& dpt ) const
{
for( int dvi = 0; dvi < dpt.getNumValues(); ++dvi )
// Can classify only nominal values
mlAssert( dpt.getValueType(dvi) == AT_Enum );
int attrc = dpt.getNumValues() - 1;
Node* qnode = getNode(attrc);
DataPoint qdpt = dpt;
std::vector<double> vlogp = std::vector<double>( qnode->getNumValues(), 0 );
// Compute probability for every possible value
double sump = 0;
double maxlogp = -DBL_MAX;
int maxv = 0;
for( int vi = 0; vi < qnode->getNumValues(); ++vi )
{
qdpt.setEnumValue( attrc, vi );
vlogp[vi] = queryLog(qdpt);
sump += exp( vlogp[vi] );
if( vlogp[vi] > maxlogp )
{
maxlogp = vlogp[vi];
maxv = vi;
}
}
// Compute scaled probability of this instance
sump -= exp( vlogp[maxv] );
double logp = vlogp[ dpt.getEnumValue(attrc) ] - maxlogp - log( 1 + sump/exp( vlogp[maxv] ) );
double prob = exp(logp);
return prob;
}
void selectAlignments(std::vector<std::vector<Chain*>>& dst, DbAlignment*** alignments,
int32_t* alignments_lengths, Chain** queries, int32_t queries_length,
float threshold) {
dst.resize(queries_length);
fprintf(stderr, "** Selecting alignments with median threshold: %.2f **\n", threshold);
std::vector<ThreadPoolTask*> thread_tasks(queries_length, nullptr);
for (int32_t i = 0; i < queries_length; ++i) {
if (alignments_lengths[i] == 0) {
continue;
}
auto thread_data = new ThreadSelectionData(dst[i], alignments[i],
alignments_lengths[i], queries[i], threshold);
thread_tasks[i] = threadPoolSubmit(threadSelectAlignments, (void*) thread_data);
}
for (int32_t i = 0; i < queries_length; ++i) {
threadPoolTaskWait(thread_tasks[i]);
threadPoolTaskDelete(thread_tasks[i]);
queryLog(i + 1, queries_length);
}
fprintf(stderr, "\n\n");
}
