polygon::polygon (const vertex_list& vert){
DEBUGF ('c', this);
vertex_list v;
GLfloat xavg = 0;
GLfloat yavg = 0;
for(size_t i = 0; i < vert.size(); i++){
xavg += vert[i].xpos; yavg += vert[i].ypos;
}
xavg /= vert.size(); yavg /= vert.size();
for(size_t i = 0; i < vert.size(); i++)
v.push_back({vert[i].xpos - xavg, vert[i].ypos - yavg});
vertices = v;
}
std::vector<double> dissimilarity_subsets_sp(
const adjacency_array<vid_t>& adj,
const vertex_list& s, const redwalk::MatrixUS& sp)
{
const idx_t nv = adj.vcount();
const idx_t ns = s.size();
//to check if a vertex is a target
std::vector<bool> is_targ(nv, false);
for(idx_t i = 0; i < ns; i++)
is_targ[s[i]] = true;
//Set up triplet list for sparse matrix
typedef Eigen::Triplet<redwalk::flt_t> Triplet;
std::vector<Triplet> tripletList;
tripletList.reserve(nv + 2 * adj.ecount()); //TODO: reserve exact or close to
//Build the tripletlist
for(idx_t i = 0; i < nv; i++)
{
tripletList.push_back(Triplet(i, i, redwalk::ONE_F));
if(!is_targ[i])
{
redwalk::flt_t val = redwalk::NEG_ONE_F / (redwalk::flt_t)adj.degree(i);
NeighborhoodList<vid_t> nbrs = adj[i];
NeighborhoodList<vid_t>::iterator nitr;
for(nitr = nbrs.begin(); nitr != nbrs.end(); nitr++)
tripletList.push_back(Triplet(i,*nitr,val));
}
}
//set up the random walk normalized laplacian
Eigen::SparseMatrix<redwalk::flt_t> L(nv, nv);
L.setFromTriplets(tripletList.begin(), tripletList.end());
L.makeCompressed();
//free the triplets
std::vector<Triplet>().swap(tripletList);
//factor the RWNL
Eigen::SparseLU<Eigen::SparseMatrix<redwalk::flt_t, Eigen::ColMajor>,
Eigen::COLAMDOrdering<int> > solver;
solver.analyzePattern(L);
solver.factorize(L);
//setup b
redwalk::MatrixF b(nv, ns);
b.fill(0);
for(idx_t i = 0; i < ns; i++)
b.row(s[i]) = sp.col(i).cast<redwalk::flt_t>();
//Solve
const redwalk::MatrixF x = solver.solve(b);
//Free b
b.resize(0,0);
//Free L
Eigen::SparseMatrix<redwalk::flt_t, Eigen::ColMajor>().swap(L);
//Allocate D - what will become a "dist"
//number of entries in contiguous symmetric matrix with no diagonal
std::vector<double> D((ns * ns - ns) >> 1, 0.0);
idx_t idx = 0;
redwalk::flt_t mspi, mspj;
for(idx_t i = 0; i < (ns-1); i++)
{
vid_t ti = s[i];
idx_t degi = adj.degree(ti);
NeighborhoodList<vid_t> nbrsi = adj[ti];
for(idx_t j = i+1; j < ns; j++)
{
vid_t tj = s[j];
idx_t degj = adj.degree(tj);
NeighborhoodList<vid_t> nbrsj = adj[tj];
mspi = 0; mspj = 0;
for(idx_t n = 0; n < degi; n++)
mspi += x(nbrsi[n],j);
mspi /= degi;
for(idx_t n = 0; n < degj; n++)
mspj += x(nbrsj[n],i);
mspj /= degj;
D[idx] = (mspi < mspj) ? mspi : mspj;
idx++;
}
}
return D;
}