//
// Determine if 2 sets contain the same elements.
//
bool DataFlowUtil::setEquals(const Assignments& a, const Assignments& b) {
// make sure sets are the same length
if (a.size() != b.size()) {
return false;
}
// ensure they contain the same elements
for (Assignments::const_iterator i = a.begin(); i != a.end(); ++i) {
const Assignment& test = *i;
if (b.count(test) < 1) {
return false;
}
}
return true;
}
Assignments get_state_clusters(const Subset &subset, const Assignments &states,
ParticleStatesTable *pst, double resolution) {
Vector<Ints> rotated(subset.size(), Ints(states.size(), -1));
for (unsigned int i = 0; i < states.size(); ++i) {
for (unsigned int j = 0; j < states[i].size(); ++j) {
rotated[j][i] = states[i][j];
}
}
for (unsigned int i = 0; i < rotated.size(); ++i) {
std::sort(rotated[i].begin(), rotated[i].end());
rotated[i].erase(std::unique(rotated[i].begin(), rotated[i].end()),
rotated[i].end());
}
Vector<Ints> clustering(states.size());
for (unsigned int i = 0; i < subset.size(); ++i) {
IMP::PointerMember<ParticleStates> ps =
pst->get_particle_states(subset[i]);
Ints c = get_state_clusters(subset[i], ps, rotated[i], resolution);
clustering[i] = c;
}
return filter_states(subset, states, clustering);
}
Paths CorrespondenceGenerator::generate()
{
Paths result;
auto boxA = a->bbox(), boxB = b->bbox();
QStringList nodesA, nodesB;
for (auto n : a->nodes) nodesA << n->id;
for (auto n : b->nodes) nodesB << n->id;
for (auto g : a->groups) for (auto nid : g) nodesA.removeAll(nid);
for (auto g : b->groups) for (auto nid : g) nodesB.removeAll(nid);
Structure::NodeGroups tmpA, tmpB;
for (auto nid : nodesA) tmpA.push_back( QVector<QString>() << nid );
for (auto nid : nodesB) tmpB.push_back( QVector<QString>() << nid );
for (auto g : a->groups) tmpA.push_back(g);
for (auto g : b->groups) tmpB.push_back(g);
a->groups = tmpA;
b->groups = tmpB;
/// Build similarity matrix of groups:
Eigen::MatrixXd similiarity = Eigen::MatrixXd::Ones(a->groups.size(), b->groups.size() + 1);
for (size_t i = 0; i < similiarity.rows(); i++){
double minVal = DBL_MAX;
for (size_t j = 0; j < similiarity.cols(); j++){
double val = 0;
if (j < b->groups.size())
{
Eigen::AlignedBox3d groupBoxA, groupBoxB;
for (auto sid : a->groups[i]) groupBoxA.extend(a->getNode(sid)->bbox(1.01));
for (auto sid : b->groups[j]) groupBoxB.extend(b->getNode(sid)->bbox(1.01));
Vector3 relativeCenterA = (groupBoxA.center() - boxA.min()).array() / boxA.sizes().array();
Vector3 relativeCenterB = (groupBoxB.center() - boxB.min()).array() / boxB.sizes().array();
val = (relativeCenterA - relativeCenterB).norm();
minVal = std::min(minVal, val);
}
similiarity(i, j) = val;
}
double maxVal = similiarity.row(i).maxCoeff();
if (maxVal == 0) maxVal = 1.0;
for (size_t j = 0; j < b->groups.size(); j++)
similiarity(i, j) = (similiarity(i, j) - minVal) / maxVal;
}
std::vector< std::vector<float> > data;
for (int i = 0; i < similiarity.rows(); i++){
std::vector<float> dataRow;
for (int j = 0; j < similiarity.cols(); j++)
dataRow.push_back(similiarity(i, j));
data.push_back(dataRow);
}
//if (false) showTableColorMap(data, true); // DEBUG
// Parameters:
double similarity_threshold = 0.4;
if (similiarity.rows() < 4) similarity_threshold = 1.0;
// Collect good candidates
Assignments candidates;
for (size_t i = 0; i < similiarity.rows(); i++){
QVector<size_t> candidate;
for (size_t j = 0; j < similiarity.cols(); j++){
if (similiarity(i, j) < similarity_threshold)
candidate << j;
}
candidates << candidate;
}
int count_threshold = 1;
Assignments assignments;
if (candidates.size() > 7)
{
debugBox(QString("%1 candidates. Too complex (lower similairty?)").arg(candidates.size()));
similarity_threshold = 0.1;
Assignments candidates;
for (size_t i = 0; i < similiarity.rows(); i++){
QVector<size_t> candidate;
for (size_t j = 0; j < similiarity.cols(); j++){
if (similiarity(i, j) < similarity_threshold)
candidate << j;
}
candidates << candidate;
}
cart_product(assignments, candidates, 10000);
count_threshold = 12;
}
else
{
cart_product(assignments, candidates);
}
// Filter assignments
Assignments filtered;
for (auto & a : assignments)
{
QMap<size_t, size_t> counts;
bool accept = true;
auto NOTHING_SEGMENT = similiarity.cols() - 1;
for (auto i : a) counts[i]++;
for (auto k : counts.keys())
{
if (k != NOTHING_SEGMENT && counts[k] > count_threshold){
accept = false;
break;
}
}
if (accept) filtered << a;
}
for (auto & assignment : filtered)
{
QVector < QStringList > landmarksA, landmarksB;
for (size_t i = 0; i < assignment.size(); i++)
{
if (assignment[i] == b->groups.size()) continue;
auto grpA = a->groups[i];
auto grpB = b->groups[assignment[i]];
// Resolve many-to-many
if (grpA.size() > 1 && grpB.size() > 1)
{
QVector<QString> tmpA, tmpB;
Eigen::AlignedBox3d groupBoxA, groupBoxB;
for (auto sid : grpA) groupBoxA.extend(a->getNode(sid)->bbox(1.01));
for (auto sid : grpB) groupBoxB.extend(b->getNode(sid)->bbox(1.01));
for (auto nodeid_i : grpA){
auto nodeA = a->getNode(nodeid_i);
QMap < QString, double > dists;
for (auto nodeid_j : grpB){
auto nodeB = b->getNode(nodeid_j);
Vector3 posA = (nodeA->position(Eigen::Vector4d(0.5, 0.5, 0, 0)) - boxA.min()).array() / boxA.sizes().array();
Vector3 posB = (nodeB->position(Eigen::Vector4d(0.5, 0.5, 0, 0)) - boxB.min()).array() / boxB.sizes().array();
dists[nodeid_j] = (posA - posB).norm();
}
auto nodeid_j = sortQMapByValue(dists).first().second;
landmarksA << (QStringList() << nodeid_i);
landmarksB << (QStringList() << nodeid_j);
}
}
else
{
landmarksA << QStringList::fromVector(grpA);
landmarksB << QStringList::fromVector(grpB);
}
}
result.push_back( qMakePair(landmarksA, landmarksB) );
}
return result;
}
