// to get first few eigenvectors
void GetEigVec(const PUNGraph& Graph, const int& EigVecs, TFltV& EigValV, TVec<TFltV>& EigVecV) {
const int Nodes = Graph->GetNodes();
// Lanczos
TUNGraphMtx GraphMtx(Graph);
int CalcVals = int(2*EigVecs);
if (CalcVals > Nodes) { CalcVals = Nodes; }
TFltVV EigVecVV;
//while (EigValV.Len() < EigVecs && CalcVals < 10*EigVecs) {
try {
TSparseSVD::Lanczos(GraphMtx, EigVecs, 2*EigVecs, ssotFull, EigValV, EigVecVV, false); }
catch(...) {
printf("\n ***EXCEPTION: TRIED %d GOT %d values** \n", CalcVals, EigValV.Len()); }
if (EigValV.Len() < EigVecs) {
printf(" ***TRIED %d GOT %d values** \n", CalcVals, EigValV.Len()); }
// CalcVals += EigVecs;
//}
TFltIntPrV EigValIdV;
for (int i = 0; i < EigValV.Len(); i++) {
EigValIdV.Add(TFltIntPr(EigValV[i], i));
}
EigValIdV.Sort(false);
EigValV.Sort(false);
for (int v = 0; v < EigValIdV.Len(); v++) { // vector components are not sorted!!!
EigVecV.Add();
EigVecVV.GetCol(EigValIdV[v].Val2, EigVecV.Last());
}
IsAllValVNeg(EigVecV[0], true);
}
/////////////////////////////////////////////////
// Top2 Friends network
void TTop2FriendNet::SetTop2() {
Top2NIdH.Gen(Net->GetNodes());
TFltIntPrV WgtNIdV;
for (TWgtNet::TNodeI NI = Net->BegNI(); NI < Net->EndNI(); NI++) {
WgtNIdV.Clr(false);
for (int e = 0; e < NI.GetOutDeg(); e++) {
WgtNIdV.Add(TFltIntPr(NI.GetOutEDat(e), NI.GetOutNId(e)));
}
WgtNIdV.Shuffle(TInt::Rnd); // so that ties are broken randomly
WgtNIdV.Sort(false);
if (WgtNIdV.Len() == 0) { Top2NIdH.AddDat(NI.GetId(), TIntPr(-1, -1)); }
else if (WgtNIdV.Len() == 1) { Top2NIdH.AddDat(NI.GetId(), TIntPr(WgtNIdV[0].Val2, -1)); }
else if (WgtNIdV.Len() >= 2) {
Top2NIdH.AddDat(NI.GetId(), TIntPr(WgtNIdV[0].Val2, WgtNIdV[1].Val2)); }
}
// create union find structure
PNGraph Top1Net = GetTop1Net();
Top1UF = TUnionFind(Top1Net->GetNodes());
TCnComV CnComV;
TCnCom::GetWccs(Top1Net, CnComV);
for (TWgtNet::TNodeI NI = Net->BegNI(); NI < Net->EndNI(); NI++) {
Top1UF.Add(NI.GetId());
}
for (int c = 0; c < CnComV.Len(); c++) {
for (int i = 1; i < CnComV[c].Len(); i++) {
Top1UF.Union(CnComV[c][0], CnComV[c][i]); }
}
}
void TLwOntoGround::ClassifySpV(
const PBowSpV& QueryBowSpV, TSimTermIdPrV& SimTermIdPrV) const {
PBowSim BowSim=TBowSim::New(bstCos);
SimTermIdPrV.Clr();
for (int TermN=0; TermN<TermIdToConceptSpVH.Len(); TermN++){
int TermId=TermIdToConceptSpVH.GetKey(TermN);
PBowSpV ConceptSpV=TermIdToConceptSpVH[TermN];
double Sim=BowSim->GetSim(QueryBowSpV, ConceptSpV);
TStr TermNm = GetLwOnto()->GetTermBs()->GetTerm(TermId)->GetTermNm();
if (Sim > 0.0) { SimTermIdPrV.Add(TFltIntPr(Sim, TermId)); } //B: >0
}
SimTermIdPrV.Sort(false);
}
void GetSngVec(const PNGraph& Graph, const int& SngVecs, TFltV& SngValV, TVec<TFltV>& LeftSV, TVec<TFltV>& RightSV) {
const int Nodes = Graph->GetNodes();
SngValV.Clr();
LeftSV.Clr();
RightSV.Clr();
TFltVV LSingV, RSingV;
if (Nodes < 100) {
// perform full SVD
TFltVV AdjMtx(Nodes+1, Nodes+1);
TIntH NodeIdH;
// create adjecency matrix (1-based)
for (TNGraph::TNodeI NodeI = Graph->BegNI(); NodeI < Graph->EndNI(); NodeI++) {
NodeIdH.AddKey(NodeI.GetId()); }
for (TNGraph::TNodeI NodeI = Graph->BegNI(); NodeI < Graph->EndNI(); NodeI++) {
const int NodeId = NodeIdH.GetKeyId(NodeI.GetId())+1;
for (int e = 0; e < NodeI.GetOutDeg(); e++) {
const int DstNId = NodeIdH.GetKeyId(NodeI.GetOutNId(e))+1; // no self edges
if (NodeId != DstNId) AdjMtx.At(NodeId, DstNId) = 1;
}
}
try { // can fail to converge but results seem to be good
TSvd::Svd1Based(AdjMtx, LSingV, SngValV, RSingV);
} catch(...) {
printf("\n***No SVD convergence: G(%d, %d)\n", Nodes, Graph->GetEdges());
}
} else { // Lanczos
TNGraphMtx GraphMtx(Graph);
TSparseSVD::LanczosSVD(GraphMtx, SngVecs, 2*SngVecs, ssotFull, SngValV, LSingV, RSingV);
//TGAlg::SaveFullMtx(Graph, "adj_mtx.txt");
//TLAMisc::DumpTFltVVMjrSubMtrx(LSingV, LSingV.GetRows(), LSingV.GetCols(), "LSingV2.txt"); // save MTX
}
TFltIntPrV SngValIdV;
for (int i = 0; i < SngValV.Len(); i++) {
SngValIdV.Add(TFltIntPr(SngValV[i], i));
}
SngValIdV.Sort(false);
SngValV.Sort(false);
for (int v = 0; v < SngValIdV.Len(); v++) {
LeftSV.Add();
LSingV.GetCol(SngValIdV[v].Val2, LeftSV.Last());
RightSV.Add();
RSingV.GetCol(SngValIdV[v].Val2, RightSV.Last());
}
IsAllValVNeg(LeftSV[0], true);
IsAllValVNeg(RightSV[0], true);
}
void TAGMFast::NeighborComInit(const int InitComs) {
//initialize with best neighborhood communities (Gleich et.al. KDD'12)
F.Gen(G->GetNodes());
SumFV.Gen(InitComs);
NumComs = InitComs;
const int Edges = G->GetEdges();
//TIntFltH NCPhiH(F.Len());
TFltIntPrV NIdPhiV(F.Len(), 0);
TIntSet InvalidNIDS(F.Len());
TIntV ChosenNIDV(InitComs, 0); //FOR DEBUG
TExeTm RunTm;
//compute conductance of neighborhood community
for (int u = 0; u < F.Len(); u++) {
TIntSet NBCmty(G->GetNI(u).GetDeg() + 1);
double Phi;
if (G->GetNI(u).GetDeg() < 5) { //do not include nodes with too few degree
Phi = 1.0;
} else {
TAGMUtil::GetNbhCom(G, u, NBCmty);
IAssert(NBCmty.Len() == G->GetNI(u).GetDeg() + 1);
Phi = TAGMUtil::GetConductance(G, NBCmty, Edges);
}
//NCPhiH.AddDat(u, Phi);
NIdPhiV.Add(TFltIntPr(Phi, u));
}
NIdPhiV.Sort(true);
printf("conductance computation completed [%s]\n", RunTm.GetTmStr());
fflush(stdout);
//choose nodes with local minimum in conductance
int CurCID = 0;
for (int ui = 0; ui < NIdPhiV.Len(); ui++) {
int UID = NIdPhiV[ui].Val2;
fflush(stdout);
if (InvalidNIDS.IsKey(UID)) { continue; }
ChosenNIDV.Add(UID); //FOR DEBUG
//add the node and its neighbors to the current community
AddCom(UID, CurCID, 1.0);
TUNGraph::TNodeI NI = G->GetNI(UID);
fflush(stdout);
for (int e = 0; e < NI.GetDeg(); e++) {
AddCom(NI.GetNbrNId(e), CurCID, 1.0);
}
//exclude its neighbors from the next considerations
for (int e = 0; e < NI.GetDeg(); e++) {
InvalidNIDS.AddKey(NI.GetNbrNId(e));
}
CurCID++;
fflush(stdout);
if (CurCID >= NumComs) { break; }
}
if (NumComs > CurCID) {
printf("%d communities needed to fill randomly\n", NumComs - CurCID);
}
//assign a member to zero-member community (if any)
for (int c = 0; c < SumFV.Len(); c++) {
if (SumFV[c] == 0.0) {
int ComSz = 10;
for (int u = 0; u < ComSz; u++) {
int UID = Rnd.GetUniDevInt(G->GetNodes());
AddCom(UID, c, Rnd.GetUniDev());
}
}
}
}
// Initialize node community memberships using best neighborhood communities (see D. Gleich et al. KDD'12).
void TAGMFit::NeighborComInit(const int InitComs) {
CIDNSetV.Gen(InitComs);
const int Edges = G->GetEdges();
TFltIntPrV NIdPhiV(G->GetNodes(), 0);
TIntSet InvalidNIDS(G->GetNodes());
TIntV ChosenNIDV(InitComs, 0); //FOR DEBUG
TExeTm RunTm;
//compute conductance of neighborhood community
TIntV NIdV;
G->GetNIdV(NIdV);
for (int u = 0; u < NIdV.Len(); u++) {
TIntSet NBCmty(G->GetNI(NIdV[u]).GetDeg() + 1);
double Phi;
if (G->GetNI(NIdV[u]).GetDeg() < 5) { //do not include nodes with too few degree
Phi = 1.0;
} else {
TAGMUtil::GetNbhCom(G, NIdV[u], NBCmty);
IAssert(NBCmty.Len() == G->GetNI(NIdV[u]).GetDeg() + 1);
Phi = TAGMUtil::GetConductance(G, NBCmty, Edges);
}
NIdPhiV.Add(TFltIntPr(Phi, NIdV[u]));
}
NIdPhiV.Sort(true);
printf("conductance computation completed [%s]\n", RunTm.GetTmStr());
fflush(stdout);
//choose nodes with local minimum in conductance
int CurCID = 0;
for (int ui = 0; ui < NIdPhiV.Len(); ui++) {
int UID = NIdPhiV[ui].Val2;
fflush(stdout);
if (InvalidNIDS.IsKey(UID)) { continue; }
ChosenNIDV.Add(UID); //FOR DEBUG
//add the node and its neighbors to the current community
CIDNSetV[CurCID].AddKey(UID);
TUNGraph::TNodeI NI = G->GetNI(UID);
fflush(stdout);
for (int e = 0; e < NI.GetDeg(); e++) {
CIDNSetV[CurCID].AddKey(NI.GetNbrNId(e));
}
//exclude its neighbors from the next considerations
for (int e = 0; e < NI.GetDeg(); e++) {
InvalidNIDS.AddKey(NI.GetNbrNId(e));
}
CurCID++;
fflush(stdout);
if (CurCID >= InitComs) { break; }
}
if (InitComs > CurCID) {
printf("%d communities needed to fill randomly\n", InitComs - CurCID);
}
//assign a member to zero-member community (if any)
for (int c = 0; c < CIDNSetV.Len(); c++) {
if (CIDNSetV[c].Len() == 0) {
int ComSz = 10;
for (int u = 0; u < ComSz; u++) {
int UID = G->GetRndNI().GetId();
CIDNSetV[c].AddKey(UID);
}
}
}
InitNodeData();
SetDefaultPNoCom();
}
