void IncrementalQuoteGraph::AddNodes() {
TIntSet::TIter QuoteIdsEnd = NewQuotes.EndI();
for (TIntSet::TIter QuoteId = NewQuotes.BegI(); QuoteId < QuoteIdsEnd; QuoteId++) {
if (!QGraph->IsNode(QuoteId.GetKey()))
QGraph->AddNode(QuoteId.GetKey());
}
}
void TAGMFit::InitNodeData() {
TSnap::DelSelfEdges(G);
NIDComVH.Gen(G->GetNodes());
for (TUNGraph::TNodeI NI = G->BegNI(); NI < G->EndNI(); NI++) {
NIDComVH.AddDat(NI.GetId());
}
TAGMUtil::GetNodeMembership(NIDComVH, CIDNSetV);
GetEdgeJointCom();
LambdaV.Gen(CIDNSetV.Len());
for (int c = 0; c < CIDNSetV.Len(); c++) {
int MaxE = (CIDNSetV[c].Len()) * (CIDNSetV[c].Len() - 1) / 2;
if (MaxE < 2) {
LambdaV[c] = MaxLambda;
}
else{
LambdaV[c] = -log((double) (MaxE - ComEdgesV[c]) / MaxE);
}
if (LambdaV[c] > MaxLambda) { LambdaV[c] = MaxLambda; }
if (LambdaV[c] < MinLambda) { LambdaV[c] = MinLambda; }
}
NIDCIDPrS.Gen(G->GetNodes() * 10);
for (int c = 0; c < CIDNSetV.Len(); c++) {
for (TIntSet::TIter SI = CIDNSetV[c].BegI(); SI < CIDNSetV[c].EndI(); SI++) {
NIDCIDPrS.AddKey(TIntPr(SI.GetKey(), c));
}
}
}
double TAGMFit::SelectLambdaSum(const TFltV& NewLambdaV, const TIntSet& ComK) {
double Result = 0.0;
for (TIntSet::TIter SI = ComK.BegI(); SI < ComK.EndI(); SI++) {
IAssert(NewLambdaV[SI.GetKey()] >= 0);
Result += NewLambdaV[SI.GetKey()];
}
return Result;
}
void TAGMUtil::GetNodeMembership(THash<TInt,TIntSet >& NIDComVH, const TVec<TIntSet>& CmtyVV) {
for (int i = 0; i < CmtyVV.Len(); i++) {
int CID = i;
for (TIntSet::TIter SI = CmtyVV[i].BegI(); SI < CmtyVV[i].EndI(); SI++) {
int NID = SI.GetKey();
NIDComVH.AddDat(NID).AddKey(CID);
}
}
}
// YES I COPIED AND PASTED CODE my section leader would be so ashamed :D
void LSH::MinHash(THash<TMd5Sig, TIntSet>& ShingleToQuoteIds,
TVec<THash<TIntV, TIntSet> >& SignatureBandBuckets) {
TRnd RandomGenerator; // TODO: make this "more random" by incorporating time
for (int i = 0; i < NumBands; ++i) {
THash < TInt, TIntV > Inverted; // (QuoteID, QuoteSignatureForBand)
THash < TIntV, TIntSet > BandBuckets; // (BandSignature, QuoteIDs)
for (int j = 0; j < BandSize; ++j) {
// Create new signature
TVec < TMd5Sig > Signature;
ShingleToQuoteIds.GetKeyV(Signature);
Signature.Shuffle(RandomGenerator);
// Place in bucket - not very efficient
int SigLen = Signature.Len();
for (int k = 0; k < SigLen; ++k) {
TIntSet CurSet = ShingleToQuoteIds.GetDat(Signature[k]);
for (TIntSet::TIter l = CurSet.BegI(); l < CurSet.EndI(); l++) {
TInt Key = l.GetKey();
if (Inverted.IsKey(Key)) {
TIntV CurSignature = Inverted.GetDat(Key);
if (CurSignature.Len() <= j) {
CurSignature.Add(k);
Inverted.AddDat(Key, CurSignature);
}
} else {
TIntV NewSignature;
NewSignature.Add(k);
Inverted.AddDat(Key, NewSignature);
}
}
}
}
TIntV InvertedKeys;
Inverted.GetKeyV(InvertedKeys);
TInt InvertedLen = InvertedKeys.Len();
for (int k = 0; k < InvertedLen; ++k) {
TIntSet Bucket;
TIntV Signature = Inverted.GetDat(InvertedKeys[k]);
if (BandBuckets.IsKey(Signature)) {
Bucket = BandBuckets.GetDat(Signature);
}
Bucket.AddKey(InvertedKeys[k]);
BandBuckets.AddDat(Signature, Bucket);
}
SignatureBandBuckets.Add(BandBuckets);
Err("%d out of %d band signatures computed\n", i + 1, NumBands);
}
Err("Minhash step complete!\n");
}
void TIncrementalClustering::KeepAtMostOneChildPerNode(PNGraph& G, TQuoteBase *QB, TDocBase *DB) {
TIntSet::TIter EndNode = AffectedNodes.EndI();
for (TIntSet::TIter NodeId = AffectedNodes.BegI(); NodeId < EndNode; NodeId++) {
TNGraph::TNodeI Node = G->GetNI(NodeId.GetKey());
TQuote SourceQuote;
if (QB->GetQuote(Node.GetId(), SourceQuote)) {
TInt NodeDegree = Node.GetOutDeg();
if (NodeDegree > 1) {
TFlt MaxScore = 0;
TInt MaxNodeId = 0;
TIntV NodeV;
// first pass: check to see if we are pointing to any old nodes - if so, they get higher
// priority over the new ones for edge selection.
bool ContainsOldNode = false;
for (int i = 0; i < NodeDegree; ++i) {
if (!NewQuotes.IsKey(Node.GetOutNId(i))) {
ContainsOldNode = true;
}
}
// modified edge selection: filter out new nodes if old ones exist.
for (int i = 0; i < NodeDegree; ++i) {
TInt CurNode = Node.GetOutNId(i);
NodeV.Add(CurNode);
TQuote DestQuote;
if (QB->GetQuote(CurNode, DestQuote)) {
TFlt EdgeScore = 0;
if (!ContainsOldNode || !NewQuotes.IsKey(Node.GetOutNId(i))) {
EdgeScore = ComputeEdgeScore(SourceQuote, DestQuote, DB);
}
if (EdgeScore > MaxScore) {
MaxScore = EdgeScore;
MaxNodeId = CurNode;
}
}
}
// remove all other edges, backwards to prevent indexing fail
for (int i = 0; i < NodeV.Len(); i++) {
if (NodeV[i] != MaxNodeId) {
G->DelEdge(Node.GetId(), NodeV[i]);
}
}
//printf("Out degree: %d out of %d\n", Node.GetOutDeg(), NodeDegree.Val);
}
}
}
fprintf(stderr, "finished deleting edges\n");
}
// Gradient of likelihood for P_c.
void TAGMFit::GradLogLForLambda(TFltV& GradV) {
GradV.Gen(LambdaV.Len());
TFltV SumEdgeProbsV(LambdaV.Len());
for (int e = 0; e < EdgeComVH.Len(); e++) {
TIntSet& JointCom = EdgeComVH[e];
double LambdaSum = SelectLambdaSum(JointCom);
double Puv = 1 - exp(- LambdaSum);
if (JointCom.Len() == 0) { Puv = PNoCom; }
for (TIntSet::TIter SI = JointCom.BegI(); SI < JointCom.EndI(); SI++) {
SumEdgeProbsV[SI.GetKey()] += (1 - Puv) / Puv;
}
}
for (int k = 0; k < LambdaV.Len(); k++) {
int MaxEk = CIDNSetV[k].Len() * (CIDNSetV[k].Len() - 1) / 2;
int NotEdgesInCom = MaxEk - ComEdgesV[k];
GradV[k] = SumEdgeProbsV[k] - (double) NotEdgesInCom;
if (LambdaV[k] > 0.0 && RegCoef > 0.0) { //if regularization exists
GradV[k] -= RegCoef;
}
}
}
/// Newton method: DEPRECATED
int TAGMFast::MLENewton(const double& Thres, const int& MaxIter, const TStr PlotNm) {
TExeTm ExeTm;
int iter = 0, PrevIter = 0;
TIntFltPrV IterLV;
double PrevL = TFlt::Mn, CurL;
TUNGraph::TNodeI UI;
TIntV NIdxV;
G->GetNIdV(NIdxV);
int CID, UID, NewtonIter;
double Fuc, PrevFuc, Grad, H;
while(iter < MaxIter) {
NIdxV.Shuffle(Rnd);
for (int ui = 0; ui < F.Len(); ui++, iter++) {
if (! PlotNm.Empty() && iter % G->GetNodes() == 0) {
IterLV.Add(TIntFltPr(iter, Likelihood(false)));
}
UID = NIdxV[ui];
//find set of candidate c (we only need to consider c to which a neighbor of u belongs to)
TIntSet CIDSet;
UI = G->GetNI(UID);
if (UI.GetDeg() == 0) { //if the node is isolated, clear its membership and skip
if (! F[UID].Empty()) { F[UID].Clr(); }
continue;
}
for (int e = 0; e < UI.GetDeg(); e++) {
if (HOVIDSV[UID].IsKey(UI.GetNbrNId(e))) { continue; }
TIntFltH& NbhCIDH = F[UI.GetNbrNId(e)];
for (TIntFltH::TIter CI = NbhCIDH.BegI(); CI < NbhCIDH.EndI(); CI++) {
CIDSet.AddKey(CI.GetKey());
}
}
for (TIntFltH::TIter CI = F[UID].BegI(); CI < F[UID].EndI(); CI++) { //remove the community membership which U does not share with its neighbors
if (! CIDSet.IsKey(CI.GetKey())) {
DelCom(UID, CI.GetKey());
}
}
if (CIDSet.Empty()) { continue; }
for (TIntSet::TIter CI = CIDSet.BegI(); CI < CIDSet.EndI(); CI++) {
CID = CI.GetKey();
//optimize for UID, CID
//compute constants
TFltV AlphaKV(UI.GetDeg());
for (int e = 0; e < UI.GetDeg(); e++) {
if (HOVIDSV[UID].IsKey(UI.GetNbrNId(e))) { continue; }
AlphaKV[e] = (1 - PNoCom) * exp(- DotProduct(UID, UI.GetNbrNId(e)) + GetCom(UI.GetNbrNId(e), CID) * GetCom(UID, CID));
IAssertR(AlphaKV[e] <= 1.0, TStr::Fmt("AlphaKV=%f, %f, %f", AlphaKV[e].Val, PNoCom.Val, GetCom(UI.GetNbrNId(e), CID)));
}
Fuc = GetCom(UID, CID);
PrevFuc = Fuc;
Grad = GradientForOneVar(AlphaKV, UID, CID, Fuc), H = 0.0;
if (Grad <= 1e-3 && Grad >= -0.1) { continue; }
NewtonIter = 0;
while (NewtonIter++ < 10) {
Grad = GradientForOneVar(AlphaKV, UID, CID, Fuc), H = 0.0;
H = HessianForOneVar(AlphaKV, UID, CID, Fuc);
if (Fuc == 0.0 && Grad <= 0.0) { Grad = 0.0; }
if (fabs(Grad) < 1e-3) { break; }
if (H == 0.0) { Fuc = 0.0; break; }
double NewtonStep = - Grad / H;
if (NewtonStep < -0.5) { NewtonStep = - 0.5; }
Fuc += NewtonStep;
if (Fuc < 0.0) { Fuc = 0.0; }
}
if (Fuc == 0.0) {
DelCom(UID, CID);
}
else {
AddCom(UID, CID, Fuc);
}
}
}
if (iter - PrevIter >= 2 * G->GetNodes() && iter > 10000) {
PrevIter = iter;
CurL = Likelihood();
if (PrevL > TFlt::Mn && ! PlotNm.Empty()) {
printf("\r%d iterations, Likelihood: %f, Diff: %f", iter, CurL, CurL - PrevL);
}
fflush(stdout);
if (CurL - PrevL <= Thres * fabs(PrevL)) { break; }
else { PrevL = CurL; }
}
}
if (! PlotNm.Empty()) {
printf("\nMLE for Lambda completed with %d iterations(%s)\n", iter, ExeTm.GetTmStr());
TGnuPlot::PlotValV(IterLV, PlotNm + ".likelihood_Q");
}
return iter;
}
