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 );
}
DistValueMat createClusterProbTables( const Variables& variables, const Cluster& cluster ) {
DistValueMat distValueMat;
Cluster::const_iterator curr, next;
curr = cluster.begin(); next = curr + 1;
while ( next != cluster.end() ) {
DistValueVec X_Z;
for (size_t i = 0; i < variables[*curr].cardinality(); ++i) {
DistValues X_Z_i = createNBUniVarProbTab( variables[*next].cardinality() );
X_Z_i[i] = 1.0 / (1.0 - RATE);
normalizeDistValues(X_Z_i);
X_Z.push_back(X_Z_i);
}
distValueMat.push_back(X_Z);
++curr; ++next;
}
return distValueMat;
}
bool ClusterDbscan::ClusterNoise(pClusterPoint& p, Cluster& neighbors)
{
if (m_result.size() > 1)
{
std::vector<unsigned int> cluster_count;
cluster_count.resize(m_result.size(), 0);
for (ClusterIter iter = neighbors.begin();
iter != neighbors.end(); ++iter)
{
if ((*iter)->noise)
continue;
int index = m_result.find_(*iter);
if (index == -1)
continue;
cluster_count[index]++;
}
//find max
auto result = std::max_element(
cluster_count.begin(), cluster_count.end());
if (*result)
{
int index = std::distance(cluster_count.begin(), result);
m_result[index].push_back(p);
p->noise = false;
return true;
}
else return false;
}
else
{
Cluster cluster;
cluster.push_back(p);
p->noise = false;
m_result.push_back(cluster);
return true;
}
}
void ClusterDbscan::ExpandCluster(pClusterPoint& p, Cluster& neighbors, Cluster& cluster)
{
cluster.push_back(p);
p->noise = false;
for (ClusterIter iter = neighbors.begin();
iter != neighbors.end(); ++iter)
{
pClusterPoint p2 = *iter;
if (!p2->visited)
{
p2->visited = true;
Cluster neighbors2 = GetNeighbors(p2, m_eps, m_intw);
if (neighbors2.size() >= m_size)
neighbors.join(neighbors2);
}
if (!m_result.find(p2))
{
cluster.push_back(p2);
p2->noise = false;
}
}
}
inline void MatchingDynamicClusterer::find_matches( const Cluster &step_cluster, vector<int> &matches )
{
int size_step = (int)step_cluster.size();
if( size_step < MIN_CLUSTER_SIZE )
{
return;
}
DynamicClustering::iterator dit;
DynamicClustering::iterator dend = m_dynamic.end();
int dyn_index = 0;
for( dit = m_dynamic.begin() ; dit != dend; dit++, dyn_index++ )
{
// Dead?
if( m_death_age > 0 && m_dynamic[dyn_index].is_dead( m_step, m_death_age ) )
{
continue;
}
Cluster& front = (*dit).front();
int size_front = front.size();
set<NODE> tmp;
set_intersection(step_cluster.begin(), step_cluster.end(), front.begin(), front.end(), insert_iterator< set < NODE > > (tmp,tmp.begin()) );
int inter = tmp.size();
if( inter == 0 )
{
continue;
}
#ifdef SIM_OVERLAP
double sim = (double)(inter)/min(size_step,size_front);
#else
double sim = (double)(inter)/(size_step+size_front-inter);
#endif
if( sim > m_threshold )
{
matches.push_back( dyn_index );
}
}
}
