1、A Multi-Level Parallel Implementation of a Program for Finding Frequent Patterns in a Large Sparse Graph,Steve Reinhardt, Interactive Supercomputing George Karypis, Dept. of Computer Science, University of Minnesota,Outline,Problem definition Prior work Problem and Approach Results Issues and Concl
2、usions,Graph Datasets,Flexible and powerful representation Evidence extraction and link discovery (EELD) Social Networks/Web graphs Chemical compounds Protein structures Biological Pathways Object recognition and retrieval Multi-relational datasets,Finding Patterns in Graphs Many Dimensions,Structur
3、e of the graph dataset many small graphs graph transaction setting one large graph single-graph setting Type of patterns connected subgraphs induced subgraphs Nature of the algorithm Finds all patterns that satisfy the minimum support requirement Complete Finds some of the patterns Incomplete Nature
4、 of the patterns occurrence The pattern occurs exactly in the input graph Exact algorithms There is a sufficiently similar embedding of the pattern in the graph Inexact algorithms,MIS calculation for frequency exact approximate upper bound Algorithm vertical (depth-first) horizontal (breadth-first),
5、Single Graph Setting,Find all frequent subgraphs from a single sparse graph. Choice of frequency definition,vSIGRAM: Vertical Solution,Candidate generation by extension Add one more edge to a current embedding. Solve MIS on embeddings in the same equivalence class. No downward-closure-based pruning
6、Two important components Frequency-based pruning of extensions Treefication based on canonical labeling,vSIGRAM: Connection Table,Frequency-based pruning. Trying every possible extension is expensive and inefficient. A particular extension might have been tested before. Categorize extensions into eq
7、uivalent classes (in terms of isomorphism), and record if each class is frequent or not. If a class becomes infrequent, never try it in later exploration.,Parallelization,Two clear sources of parallelism in the algorithm Amount of parallelism from each source not known in advance The code is typical
8、 C code structs, pointers, frequent mallocs/frees of small areas, etc. nothing like the “Fortran”-like (dense linear algebra) examples shown for many parallel programming methods Parallel structures need to accommodate dynamic parallelism Dynamic specification of parallel work Dynamic allocation of
9、processors to work Chose OpenMP taskq/task constructs Proposed extensions to OpenMP standard Support parallel work being defined in multiple places in a program, but be placed on a single conceptual queue and executed accordingly 20 lines of code changes in 15,000 line program Electric Fence was ver
10、y useful in finding coding errors,Algorithmic Parallelism,vSiGraM (G, MIS_type, f) 1. F 2. F1 all frequent size-1 subgraphs in G 3. for each F1 in F1 do 4. M(F1) all embeddings of F1 5. for each F1 in F1 do / high-level parallelism 6. F F vSiGraM-Extend(F1, G, f) return FvSiGraM-Extend(Fk, G , f) 1.
11、 F 2. for each embedding m in M(Fk) do / low-level parallelism 3. Ck+1 C k+1 all (k+1)-subgraphs of G containing m 4. for each Ck+1 in Ck+1 do 5. if Fk is not the generating parent of Ck+1 then 6. continue 7. compute Ck+1.freq from M(Ck+1) 8. if Ck+1.freq f then 9. continue 10. F F vSiGraM-Extend(Ck
12、+1, G, f) 11.return F,Simple Taskq/Task Example,main() int val; #pragma intel omp taskqval = fib(12345); fib(int n) int partret2;if (n2) #pragma intel omp taskfor(i=n-2; in; i+) partretn-2-i = fib(i);return (partret0 + partret1); else return 1; ,High-Level Parallelism with taskq/task,/ At the bottom
13、 of expand_subgraph, after all child / subgraphs have been identified, start them all. #pragma intel omp taskqfor (ii=0; iict, lg, ls, o); / end-task,Low-Level Parallelism with taskq/task,#pragma omp parallel shared(nt, priv_es) #pragma omp masternt = omp_get_num_threads(); /#threads in parpriv_es =
14、 (ExtensionSet *)kmp_calloc(nt, sizeof(ExtensionSet *); #pragma omp barrier #pragma intel omp taskqfor (i = 0; i sg_vmap_size(sg); i+) #pragma intel omp task captureprivate(i)int th = omp_get_thread_num();if (priv_esth = NULL) priv_esth = exset_init(128);expand_map(sg, ct, ams, i, priv_esth, lg); /
15、end parallel section; next loop is serial reductionfor (i=0; i nt; i+) if (priv_esi != NULL) exset_merge(priv_esi,es);kmp_free(priv_es); ,Implementation due to Grant Haab and colleagues from Intel OpenMP library group,Experimental Results,SGI Altix 32 Itanium2 sockets (64 cores), 1.6GHz 64 GBytes (t
16、hough not memory limited) Linux No special dplace/cpuset configuration Minimum frequencies chosen to illuminate scaling behavior, not provide maximum performance,Dataset 1 - Chemical,Dataset 2 aviation,Performance of High-level Parallelism,When sufficient quantity of work (i.e., frequency threshold
17、is low enough) Good speed-ups to 16P Reasonable speed-ups to 30P Little or no benefit above 30P No insight into performance plateau,Poor Performance of Low-level Parallelism,Several possible effects ruled out Granularity of data allocation Barrier before master-only reduction Source: highly variable
18、 times for register_extension 100X slower in parallel than serial, but different instances from execution to execution Apparently due to highly variable run-times for malloc Not understood,Issues and Conclusions,OpenMP taskq/task were straightforward to use in this program and implemented the desire
19、d model Performance was good to a medium range of processor counts (best 26X on 30P) Difficult to gain insight into lack of performance High-level parallelism 30P and above Low-level parallelism,Backup,Datasets,Datasets,Generally, vSIGRAM is 2-5 times faster than hSIGRAM (with exact and upper bound
20、MIS) Largest pattern contained 13 edges.,Aviation Dataset,Credit Dataset,Generally, vSIGRAM is 2-5 times faster than hSIGRAM (with exact and upper bound MIS). Largest pattern contained 13 edges.,But, hSIGRAM can be more efficient especially with upper bound MIS (ub). Largest pattern contained 16 edg
21、es.,Citation Dataset,Contact Map Dataset,DTP Dataset,VLSI Dataset,Exact MIS never finished. Longest pattern contained 5 edges (constraint).,SUBDUE,D. J. Cook and L. B. Holder. J. Artificial Intelligence Research, vol. 1, 1994. Heuristic pattern discovery system based on MDL, written in C. Version 5.
22、0.6 With the default setting, finds 3 most interesting patterns. No overlaps are allowed.,Comparison with SUBDUE,Similar results with SEuS,Comparison With SEuS,S. Ghazizadeh and S. Chawathe. DS2002. Pattern discovery algorithm using the summary data structure. Allows overlaps when counting frequency
23、. Tends to produce more number of patterns, because the frequency of each patterns becomes generally higher. Written in JAVA From Credit Dataset, SEuS discovered 48 patterns for 50 seconds (the support threshold unknown). vSIGRAM (apprx) spent 20 seconds to find 11,696 patterns.,Summary,With approxi
24、mate and exact MIS, vSIGRAM is 2-5 times faster than hSIGRAM. With upper bound MIS, however, hSIGRAM can prune a larger number of infrequent patterns. The downward closure property plays the role. For some datasets, using exact MIS for frequency counting is just intractable. Compared to SUBDUE, SIGR
25、AM finds more and longer patterns in shorter amount of runtime.,Thank You!,Slightly longer version of this paper is also available as a technical report. SIGRAM executables will be available for download soon from http:/www.cs.umn.edu/karypis/pafi/,Complete Frequent Subgraph MiningExisting Work So F
26、ar,Input: A set of graphs (transactions) + support threshold Goal: Find all frequently occurring subgraphs in the input dataset. AGM (Inokuchi et al., 2000), vertex-based, may not be connected. FSG (Kuramochi et. al., 2001), edge-based, only connected subgraphs AcGM (Inokuchi et al., 2002), gSpan (Y
27、an & Han, 2002), FFSM (Huan et al., 2003), etc. follow FSGs problem definition. Frequency of each subgraph The number of supporting transactions. Does not matter how many embeddings are in each transaction.,Frequency Under Transaction Setting,Transaction 1,Transaction 2,Transaction 3,Convenient assu
28、mption No need to care multiple embeddings per transaction,Wait!,What happens if there is no notion of transactions in input datasets?Many real graph datasets are not in the transaction format. Network-related, VLSI design, etc. Graphs created from data with temporal nature (e.g., link discovery, in
29、trusion detection),What is the reasonable frequency definition?,Two reasonable choices: The frequency is determined by the total number of embeddings. Not downward closed. Too many patterns. Artificially high frequency of certain patterns. The frequency is determined by the number of edge-disjoint e
30、mbeddings (Vanetik et al, ICDM 2002). Downward closed. Since each occurrence utilizes different sets of edges, occurrence frequencies are bounded. Solved by finding the maximum independent set (MIS) of the embedding overlap graph.,Embedding Overlap and MIS,Edge-disjoint embeddings E1, E2, E3 E1, E2,
31、 E4 Create an overlap graph and solve MIS Vertex Embedding Edge Overlap,E1,E2,E3,E4,OK. Definition is Fine, but ,MIS-based frequency seems reasonable.Next question: How to develop mining algorithms for the single graph setting.,How to Handle Single Graph Setting?,Issue 1: Frequency counting Exact MI
32、S is often intractable.Issue 2: Choice of search scheme Horizontal (breadth-first) Vertical (depth-first),Issue 1: MIS-Based Frequency,We considered approximate (greedy) and upper bound MIS too. Approximate MIS may underestimate the frequency. Upper bound MIS may overestimate the frequency. MIS is N
33、P-complete and not be approximated. Practically simple greedy scheme works pretty well. Halldrsson and Radhakrishnan. Greed is good, 1997.,Approximate and Upper Bound MIS,Greedy MIS Successively remove lowest degree vertices,Issue 2: Search Scheme,Frequent subgraph mining Exploration in the lattice
34、of subgraphs Horizontal Level-wise Candidate generation and pruning Joining Downward closure property Frequency counting Vertical Traverse the lattice as if it were a tree.,Stop to Summarize for the moment,Type of MIS for frequency counting Approximate (greedy) Exact Upper bound Search scheme Horizo
35、ntal Vertical,hSIGRAM: Horizontal Method,Natural extension of FSG to the single graph setting. Candidate generation and pruning. Downward closure property Tighter pruning than vertical method Two-phase frequency counting All embeddings by subgraph isomorphism Anchor edge list intersection, instead o
36、f TID list intersection. Localize subgraph isomorphism MIS for the embeddings Approximate and upper bound MIS give subset and superset respectively.,TID List Recap,TID( ) TID( ) TID( ) TID( )= T1, T3 ,TID( ) = T1, T2, T3 ,Anchor Edges,Each subgraph must appear close enough together. Keep one edge fo
37、r each. Complete embeddings require too much memory. Localize subgraph isomorphism.,Treefication,Lattice of Subgraphs,Treefied Lattice,size k - 1,size k,size k + 1,: a node in the search space (i.e., a subgraph) Based on subgraph/supergraph relation Avoid visiting the same node in the lattice more than once.,