void TCliqueOverlap::Expand(const THashSet<TInt>& SUBG, THashSet<TInt>& CAND) { if (SUBG.Len()==0) { if (m_Q.Len() >= m_minMaxCliqueSize) { m_Q.Pack(); m_maxCliques->Add(m_Q); } return; } if (CAND.Len()==0) return; //Get u that maximaze CAND intersection with neighbours of vertex u int u = MaxNbrsInCANDNodeId(SUBG, CAND); //Get neighbours of node u THashSet<TInt> nbrsU; GetNbrs(u, nbrsU); //Get relative complement of nbrsU in CAND THashSet<TInt> EXT; GetRelativeComplement(CAND, nbrsU, EXT); while(EXT.Len() != 0) { int q = GetNodeIdWithMaxDeg(EXT); // m_Q.Add(q); // THashSet<TInt> nbrsQ; GetNbrs(q, nbrsQ); // THashSet<TInt> SUBGq; GetIntersection(SUBG, nbrsQ, SUBGq); // THashSet<TInt> CANDq; GetIntersection(CAND, nbrsQ, CANDq); // Expand(SUBGq, CANDq); // CAND.DelKey(q); m_Q.DelLast(); // EXT.Clr(); GetRelativeComplement(CAND, nbrsU, EXT); } }
void TCliqueOverlap::GetIntersection(const THashSet<TInt>& A, const THashSet<TInt>& B, THashSet<TInt>& C) { if (A.Len() < B.Len()) { for (THashSetKeyI<TInt> it=A.BegI(); it<A.EndI(); it++) if (B.IsKey(it.GetKey())) C.AddKey(it.GetKey()); } else { for (THashSetKeyI<TInt> it=B.BegI(); it<B.EndI(); it++) if (A.IsKey(it.GetKey())) C.AddKey(it.GetKey()); } }
int TCliqueOverlap::Intersection(const THashSet<TInt>& A, const THashSet<TInt>& B) { int n = 0; if (A.Len() < B.Len()) { for (THashSetKeyI<TInt> it=A.BegI(); it<A.EndI(); it++) if (B.IsKey(it.GetKey())) n++; } else { for (THashSetKeyI<TInt> it=B.BegI(); it<B.EndI(); it++) if (A.IsKey(it.GetKey())) n++; } return n; }
void LSH::ComputeSignatures(THashSet<TMd5Sig>& Shingles, THash<TMd5Sig, TIntV>& Signatures, int NumSignatures) { if (NumSignatures < 1) return; TRnd RandomGenerator; // TODO: make this "more random" by incorporating time TInt NumShingles = Shingles.Len(); for (int i = 0; i < NumSignatures; ++i) { // Create new signature TVec < TMd5Sig > Shuffle; Shingles.GetKeyV(Shuffle); Shuffle.Shuffle(RandomGenerator); for (int j = 0; j < NumShingles; j++) { TIntV Signature; Signatures.IsKeyGetDat(Shuffle[j], Signature); Signature.Add(j); Signatures.AddDat(Shuffle[j], Signature); } } Err("Computed %d signatures!\n", NumSignatures); }
void LSH::WordHashing(TQuoteBase* QuoteBase, THashSet<TMd5Sig>& Shingles) { Err("Hashing shingles using words...\n"); TIntV QuoteIds; QuoteBase->GetAllQuoteIds(QuoteIds); for (int qt = 0; qt < QuoteIds.Len(); qt++) { if (qt % 1000 == 0) { Err("%d out of %d completed\n", qt, QuoteIds.Len()); } TQuote Q; QuoteBase->GetQuote(QuoteIds[qt], Q); TStrV Content; Q.GetParsedContent(Content); int ContentLen = Content.Len(); for (int i = 0; i < ContentLen; i++) { const TMd5Sig ShingleMd5(Content[i]); Shingles.AddKey(ShingleMd5); } } Err("Done with word hashing! Number of shingles: %d\n", Shingles.Len()); }
void LSH::MinHash(TQuoteBase *QB, THashSet<TMd5Sig>& Shingles, TVec<THash<TMd5Sig, TIntSet> >& SignatureBandBuckets) { Err("Creating buckets...\n"); THash < TMd5Sig, TIntV > Signatures; ComputeSignatures(Shingles, Signatures, NumBands * BandSize); // bucket creation for (int i = 0; i < NumBands; ++i) { SignatureBandBuckets.Add(THash<TMd5Sig, TIntSet>()); } // bucket filling int NumShingles = Shingles.Len(); THash<TInt, TQuote> Quotes; QB->GetIdToTQuotes(Quotes); THash<TInt, TQuote>::TIter CurI = Quotes.BegI(); THash<TInt, TQuote>::TIter EndI = Quotes.EndI(); TQuote Q; // SKYFALL for (; CurI < EndI; CurI++) { Q = CurI.GetDat(); TStrV Content; Q.GetParsedContent(Content); TInt Id = Q.GetId(); // signature for quote int ContentLen = Content.Len(); TVec < TIntV > Signature; for (int i = 0; i < ContentLen; i++) { const TMd5Sig ShingleMd5(Content[i]); Signature.Add(Signatures.GetDat(ShingleMd5)); } // place in bucket if (ContentLen < WordWindow) { for (int i = 0; i < NumBands; ++i) { TStr Sig; for (int j = 0; j < BandSize; ++j) { int CurSig = i * BandSize + j; TInt min = NumShingles; for (int k = 0; k < ContentLen; k++) { if (Signature[k][CurSig] < min) { min = Signature[k][CurSig]; } } Sig += min.GetStr() + "-"; } //Err(Sig.CStr()); const TMd5Sig SigMd5(Sig); TIntSet Bucket; SignatureBandBuckets[i].IsKeyGetDat(SigMd5, Bucket); Bucket.AddKey(Id); SignatureBandBuckets[i].AddDat(SigMd5, Bucket); } } else { } } Err("Minhash step complete!\n"); }
void BigMain(int argc, char* argv[]) { TExeTm ExeTm; Env = TEnv(argc, argv, TNotify::StdNotify); Env.PrepArgs("QuotesApp"); const TStr ToDo = Env.GetIfArgPrefixStr("-do:", "", "To do").GetLc(); if (Env.IsEndOfRun()) { printf("To do:\n"); printf(" MkDataset : Make memes dataset (extract quotes and save txt)\n"); printf(" ExtractSubset : Extract a subset of memes containing particular words\n"); printf(" MemesToQtBs : Load memes dataset and create quote base\n"); printf(" MkClustNet : Build cluster network from the quote base\n"); return; } #pragma region mkdataset // extract quotes and links and make them into a single file if (ToDo == "mkdataset") { const TStr InFNm = Env.GetIfArgPrefixStr("-i:", "files.txt", "Spinn3r input files (one file per line)"); const TStr OutFNm = Env.GetIfArgPrefixStr("-o:", "Spinn3r-dataset.txt", "Output file"); const int MinQtWrdLen = Env.GetIfArgPrefixInt("-w:", 3, "Minimum quote word length"); const TStr UrlFNm = Env.GetIfArgPrefixStr("-u:", "", "Seen url set (THashSet<TMd5Sig>) file name"); const bool UrlOnlyOnce = Env.GetIfArgPrefixBool("-q:", true, "Only keep unique Urls"); //// parse directly from Spinn3r TStr Spinn3rFNm; THashSet<TMd5Sig> SeenUrlSet; if (UrlOnlyOnce && ! UrlFNm.Empty()) { // keep track of already seen urls (so that there are no duplicate urls) TFIn FIn(UrlFNm); SeenUrlSet.Load(FIn); } FILE *F = fopen(OutFNm.CStr(), "wt"); TFIn FIn(InFNm); int Items=0; for (int f=0; FIn.GetNextLn(Spinn3rFNm); f++) { TQuoteExtractor QE(Spinn3rFNm.ToTrunc()); printf("Processing %02d: %s [%s]\n", f+1, Spinn3rFNm.CStr(), TExeTm::GetCurTm()); fflush(stdout); for (int item = 0; QE.Next(); item++) { const TMd5Sig PostMd5(QE.PostUrlStr); if (QE.QuoteV.Empty() && QE.LinkV.Empty()) { continue; } // no quotes, no links if (UrlOnlyOnce) { if (SeenUrlSet.IsKey(PostMd5)) { continue; } SeenUrlSet.AddKey(PostMd5); } fprintf(F, "P\t%s\n", QE.PostUrlStr.CStr()); //if (QE.PubTm > TSecTm(2008,8,30) || QE.PubTm < TSecTm(2008,7,25)) { printf("%s\n", QE.PubTm.GetStr().CStr()); } fprintf(F, "T\t%s\n", QE.PubTm.GetYmdTmStr().CStr()); for (int q = 0; q < QE.QuoteV.Len(); q++) { if (TStrUtil::CountWords(QE.QuoteV[q]) >= MinQtWrdLen) { fprintf(F, "Q\t%s\n", QE.QuoteV[q].CStr()); } } for (int l = 0; l < QE.LinkV.Len(); l++) { fprintf(F, "L\t%s\n", QE.LinkV[l].CStr()); } fprintf(F, "\n"); if (item>0 && item % Kilo(100) == 0) { QE.DumpStat(); QE.ExeTm.Tick(); } Items++; } printf("file done. Total %d all posts, %d all items\n", SeenUrlSet.Len(), Items); fflush(stdout); } printf("all done. Saving %d post urls\n", SeenUrlSet.Len()); fflush(stdout); if (! SeenUrlSet.Empty()) { TFOut FOut(OutFNm.GetFMid()+".SeenUrlSet"); SeenUrlSet.Save(FOut); } fclose(F); } #pragma endregion mkdataset #pragma region extractsubset // save posts with memes containing particular words else if (ToDo == "extractsubset") { const TStr InFNmWc = Env.GetIfArgPrefixStr("-i:", "memes_*.rar", "Input file prefix"); const bool IsInFNmWc = Env.GetIfArgPrefixBool("-w:", true, "Input is wildcard (else a file with list of input files)"); const TStr OutFNm = Env.GetIfArgPrefixStr("-o:", "memes-subset.txt", "Output memes file"); const TStr WordsFNm = Env.GetIfArgPrefixStr("-p:", "phrases-in.txt", "Phrases that memes have to contain"); TChAV CatchMemeV;// = TStr::GetV("great depression", "economic meltdown", "recession had bottomed out", "green shoots", "slow recovery", "gradual recovery"); printf("Loading %s\n", WordsFNm.CStr()); { TFIn FIn(WordsFNm); for (TStr Ln; FIn.GetNextLn(Ln); ) { printf(" %s\n", Ln.GetLc().CStr()); CatchMemeV.Add(Ln.GetLc()); } } printf("%d strings loaded\n", CatchMemeV.Len()); TFOut FOut(OutFNm); TMemesDataLoader Memes(InFNmWc, IsInFNmWc); for (int posts = 0, nsave=0; Memes.LoadNext(); posts++) { bool DoSave = false; for (int m = 0; m < Memes.MemeV.Len(); m++) { for (int i = 0; i < CatchMemeV.Len(); i++) { if (Memes.MemeV[m].SearchStr(CatchMemeV[i]) != -1) { DoSave=true; break; } } if (DoSave) { break; } } if (DoSave) { Memes.SaveTxt(FOut); nsave++; } if (posts % Mega(1) == 0) { printf("%dm posts, %d saved\n", posts/Mega(1), nsave); FOut.Flush(); } } } #pragma endregion extractsubset #pragma region memestoqtbs // load memes dataset (MkDataset) and create quote base else if (ToDo == "memestoqtbs") { const TStr InFNm = Env.GetIfArgPrefixStr("-i:", "201007_201107.txt", "Input Memes dataset files"); const TStr MediaUrlFNm = Env.GetIfArgPrefixStr("-u:", "news_media.txt", "Fule with news media urls"); const TStr Pref = Env.GetIfArgPrefixStr("-o:", "qt", "Output file name prefix"); const int MinWrdLen = Env.GetIfArgPrefixInt("-l:", 4, "Min quote word length"); const int MinMemeFq = Env.GetIfArgPrefixInt("-f:", 5, "Min meme frequency"); const TStr MinTmStr = Env.GetIfArgPrefixStr("-mint:", "20100714", "Min time of quotes, format = YYYYMMDD"); const TStr MaxTmStr = Env.GetIfArgPrefixStr("-maxt:", "20110728", "Max time of quotes, format = YYYYMMDD"); TSecTm MinTm(atoi(MinTmStr.GetSubStr(0,3).CStr()),atoi(MinTmStr.GetSubStr(4,5).CStr()),atoi(MinTmStr.GetSubStr(6,7).CStr())); TSecTm MaxTm(atoi(MaxTmStr.GetSubStr(0,3).CStr()),atoi(MaxTmStr.GetSubStr(4,5).CStr()),atoi(MaxTmStr.GetSubStr(6,7).CStr())); PQuoteBs QtBs = TQuoteBs::New(); int HashTableSize = 100; // 100 for each quarter, for one year data, use 400 int UrlSetSize = 4 * HashTableSize; QtBs->ConstructQtBsZarya(InFNm, Pref, MediaUrlFNm, MinTm, MaxTm, MinWrdLen, MinMemeFq, HashTableSize, UrlSetSize); } #pragma endregion memestoqtbs #pragma region mkclustnet // make cluster network else if (ToDo == "mkclustnet") { TStr InQtBsNm = Env.GetIfArgPrefixStr("-i:", "", "Input quote base file name"); TStr Pref = Env.GetIfArgPrefixStr("-o:", "qt", "Output network/updated QtBs filename"); TStr BlackListFNm = Env.GetIfArgPrefixStr("-b:", "quote_blacklist.txt", "Blacklist file name"); bool IsShglReady = Env.GetIfArgPrefixBool("-shglready:", false, "Indicate whether shingle hash table is ready"); bool IsNetReady = Env.GetIfArgPrefixBool("-netready:", false, "Indicate whether cluster net is ready"); double BktThresh = Env.GetIfArgPrefixFlt("-bktthresh:", 0.4, "Threshold for bad shingle bucket elimination"); double MxTmDelay = Env.GetIfArgPrefixFlt("-delaythresh:", 5, "Max time delay between two quotes in the same cluster"); double MxTmDev = Env.GetIfArgPrefixFlt("-devthresh:", 3, "Max time deviation for a quote to be specific rather than general"); double RefineThresh = Env.GetIfArgPrefixFlt("-refinethresh:", 0.2, "Threshold for merging quote cluster in refining process"); const int MinWrdLen = Env.GetIfArgPrefixInt("-minl:", 4, "Min quote word length"); const int MaxWrdLen = Env.GetIfArgPrefixInt("-maxl:", 200, "Max quote word length"); const int MinMemeFq = Env.GetIfArgPrefixInt("-mf:", 5, "Min meme frequency"); const int MinClustFq = Env.GetIfArgPrefixInt("-cf:", 50, "Min quote cluster frequency"); // Load quote base PQuoteBs QtBs; if (TZipIn::IsZipFNm(InQtBsNm)) { TZipIn ZipIn(InQtBsNm); QtBs = TQuoteBs::Load(ZipIn); } else { TFIn FIn(InQtBsNm); QtBs = TQuoteBs::Load(FIn); } // Cluster the quotes QtBs->ClusterQuotes(MinMemeFq, MinWrdLen, MaxWrdLen, BlackListFNm, Pref, IsShglReady, IsNetReady, BktThresh, MxTmDelay, MxTmDev, RefineThresh); // Dump the clusters bool SkipUrl = true, FlashDisp = true; QtBs->DumpQuoteClusters(MinWrdLen, MinClustFq, SkipUrl, FlashDisp, Pref); } #pragma endregion mkclustnet #pragma region memeclust else if (ToDo.SearchStr(TStr("memeclust")) >= 0) { const TStr InFNm = Env.GetIfArgPrefixStr("-i:", "201101.txt", "Input Memes dataset files"); const TStr MediaUrlFNm = Env.GetIfArgPrefixStr("-u:", "news_media.txt", "Fule with news media urls"); TStr Pref = Env.GetIfArgPrefixStr("-o:", "qt", "Output file name prefix"); const int MinWrdLen = Env.GetIfArgPrefixInt("-minl:", 4, "Min quote word length"); const int MaxWrdLen = Env.GetIfArgPrefixInt("-maxl:", 200, "Max quote word length"); const int MinMemeFq = Env.GetIfArgPrefixInt("-f:", 5, "Min meme frequency"); const int MinClustFq = Env.GetIfArgPrefixInt("-cf:", 50, "Min quote cluster frequency"); TStr BlackListFNm = Env.GetIfArgPrefixStr("-b:", "quote_blacklist.txt", "Blacklist file name"); bool IsQtBsReady = Env.GetIfArgPrefixBool("-qtbsready:", false, "Indicate whether quote base is ready and can be loaded readily"); bool IsShglReady = Env.GetIfArgPrefixBool("-shglready:", false, "Indicate whether shingle hash table is ready"); bool IsNetReady = Env.GetIfArgPrefixBool("-netready:", false, "Indicate whether cluster net is ready"); double BktThresh = Env.GetIfArgPrefixFlt("-bktthresh:", 0.4, "Threshold for bad shingle bucket elimination"); double MxTmDelay = Env.GetIfArgPrefixFlt("-delaythresh:", 5, "Max time delay between two quotes in the same cluster"); double MxTmDev = Env.GetIfArgPrefixFlt("-devthresh:", 3, "Max time deviation for a quote to be specific rather than general"); double RefineThresh = Env.GetIfArgPrefixFlt("-refinethresh:", 0.2, "Threshold for merging quote cluster in refining process"); const TStr MinTmStr = Env.GetIfArgPrefixStr("-mint:", "20010101", "Min time of quotes, format = YYYYMMDD"); const TStr MaxTmStr = Env.GetIfArgPrefixStr("-maxt:", "20200101", "Max time of quotes, format = YYYYMMDD"); TSecTm MinTm(atoi(MinTmStr.GetSubStr(0,3).CStr()),atoi(MinTmStr.GetSubStr(4,5).CStr()),atoi(MinTmStr.GetSubStr(6,7).CStr())); TSecTm MaxTm(atoi(MaxTmStr.GetSubStr(0,3).CStr()),atoi(MaxTmStr.GetSubStr(4,5).CStr()),atoi(MaxTmStr.GetSubStr(6,7).CStr())); // Construct the quote base from Zarya data PQuoteBs QtBs = TQuoteBs::New(); if (!IsQtBsReady) { int HashTableSize = 100; // 100 for each quarter, for one year data, use 400 if (ToDo == "memeclustzarya") { int UrlSetSize = 4 * HashTableSize; QtBs->ConstructQtBsZarya(InFNm, Pref, MediaUrlFNm, MinTm, MaxTm, MinWrdLen, MinMemeFq, HashTableSize, UrlSetSize); } else if (ToDo == "memeclustqtonly") { QtBs->ConstructQtBsQtOnly(InFNm, Pref, MediaUrlFNm, MinWrdLen, MinMemeFq, HashTableSize); } else if (ToDo == "memeclustqttime") { QtBs->ConstructQtBsQtTime(InFNm, Pref, MediaUrlFNm, MinWrdLen, MinMemeFq, HashTableSize); } else { printf("Please specify one of the three options for -do : memeclustzarya, memeclustqtonly, memeclustqttime!\n"); return; } } else { TStr InQtBsNm = TStr::Fmt("%s-w%dmfq%d.QtBs", Pref.CStr(), MinWrdLen, MinMemeFq); if (TZipIn::IsZipFNm(InQtBsNm)) { TZipIn ZipIn(InQtBsNm); QtBs = TQuoteBs::Load(ZipIn); } else { TFIn FIn(InQtBsNm); QtBs = TQuoteBs::Load(FIn); } } // Cluster the quotes QtBs->ClusterQuotes(MinMemeFq, MinWrdLen, MaxWrdLen, BlackListFNm, Pref, IsShglReady, IsNetReady, BktThresh, MxTmDelay, MxTmDev, RefineThresh); // Dump the clusters bool SkipUrl = true, FlashDisp = true; QtBs->DumpQuoteClusters(MinWrdLen, MinClustFq, SkipUrl, FlashDisp, Pref); } #pragma endregion memeclust }
/** * Used for benchmarking sorting by source algorithm. * Takes as input starting point of * a top cascade and outputs time taken for casacade detection. * Input : Source, Dest, Start, Duration * Output : Prints the time for cascade detection */ int main(int argc,char* argv[]) { TTableContext Context; Schema TimeS; TimeS.Add(TPair<TStr,TAttrType>("Source",atInt)); TimeS.Add(TPair<TStr,TAttrType>("Dest",atInt)); TimeS.Add(TPair<TStr,TAttrType>("Start",atInt)); TimeS.Add(TPair<TStr,TAttrType>("Duration",atInt)); PTable P1 = TTable::LoadSS(TimeS,"./../../../../datasets/temporal/yemen_call_201001.txt",&Context,' '); TIntV MapV; TStrV SortBy; SortBy.Add("Source"); P1->Order(SortBy); TIntV Source; // Sorted vec of start time P1->ReadIntCol("Source",Source); for (TRowIterator RI = P1->BegRI(); RI < P1->EndRI(); RI++) { MapV.Add(RI.GetRowIdx()); } // Attribute to Int mapping TInt SIdx = P1->GetColIdx("Source"); TInt DIdx = P1->GetColIdx("Dest"); TInt StIdx = P1->GetColIdx("Start"); TInt DuIdx = P1->GetColIdx("Duration"); int W = atoi(argv[1]); int len = 0; // Find the starting point int TSource = atoi(argv[2]); int TDest = atoi(argv[3]); int TStart = atoi(argv[4]); int TDur = atoi(argv[5]); TInt RIdx; for (TRowIterator RI = P1->BegRI(); RI < P1->EndRI(); RI++) { RIdx = RI.GetRowIdx(); int RSource = P1->GetIntValAtRowIdx(SIdx,RIdx).Val; int RDest = P1->GetIntValAtRowIdx(DIdx,RIdx).Val; int RStart = P1->GetIntValAtRowIdx(StIdx,RIdx).Val; int RDur = P1->GetIntValAtRowIdx(DuIdx,RIdx).Val; if (TSource == RSource && TDest == RDest && TStart == RStart && TDur == RDur) break; } // Start building the cascade from the start point clock_t st,et; st = clock(); for (int i = 0; i < 1; i++) { THashSet<TInt> VisitedH; TSnapQueue<TInt> EventQ; EventQ.Push(RIdx); VisitedH.AddKey(RIdx); while (!EventQ.Empty()) { TInt CIdx = EventQ.Top(); EventQ.Pop(); int CDest = P1->GetIntValAtRowIdx(DIdx,CIdx).Val; int CStart = P1->GetIntValAtRowIdx(StIdx,CIdx).Val; int CDur = P1->GetIntValAtRowIdx(DuIdx,CIdx).Val; // In line binary search int val = CDest; int lo = 0; int hi = Source.Len() - 1; int index = -1; while (hi >= lo) { int mid = lo + (hi - lo)/2; if (Source.GetVal(mid) > val) { hi = mid - 1;} else if (Source.GetVal(mid) < val) { lo = mid + 1;} else { index = mid; hi = mid - 1;} } // End of binary search int BIdx = index; for(int i = BIdx; i < Source.Len(); i++) { int PId = MapV.GetVal(i).Val; if (! VisitedH.IsKey(PId)) { int TSource = P1->GetIntValAtRowIdx(SIdx,PId).Val; int TStart = P1->GetIntValAtRowIdx(StIdx,PId).Val; if (TSource != CDest) { break; } if (TStart >= (CDur + CStart) && TStart - (CDur + CStart) <= W) { VisitedH.AddKey(PId); EventQ.Push(PId); } } } } len = VisitedH.Len(); } et = clock(); float diff = ((float) et - (float) st)/CLOCKS_PER_SEC; printf("Size %d,Time %f\n",len,diff); return 0; }