Skip to content

Latest commit

 

History

History
53 lines (51 loc) · 3.19 KB

commandLine.md

File metadata and controls

53 lines (51 loc) · 3.19 KB
Raw

Required

  • graph-type=market
    • This says the input will be a matrix-market (COO) format graph
  • graph-file="filename.mtx"
    • What is the graph you want to load? (default is STDIN)

Optional

  • quiet (T/F) (default = false)
  • device (int) (default = 0)
    • Set the CUDA device number
  • bfs (T/F) (default = true)
    • Run max clique algorithm on complete set of sublists all at once
    • Can choose either BFS, windowing, or run both to compare
  • windowing (T/F) (default = false)
    • Run max clique algorithm on one window of sublists at a time - basically a broad depth-first search
    • Can choose either BFS, windowing, or run both to compare.
  • heuristic (string) options: none, greedy, multi_greedy (default = "multi_greedy")
    • Specify which heuristic to use to establist an initial lower bound for pruning the search
  • frac_seeds (float) (default=1.0)
    • Specify the fraction (from 0.0 to 1.0) of vertices in the graph to use as seeds in the multi-run greedy heuristic - the number of seeds is the number of runs of the heuristic
  • pruning (string) options: none, simple (default = "simple")
    • Specify methods of pruning out non-viable candidate cliques
  • kcore (T/F) (default=false)
    • Whether to compute core numbers for vertices to use in pruning
  • timing (std::string) options: none, all, total, preproc, loop, windowed, bfs (default = "all")
    • Which parts of the algorithm to output the timing for
  • num_cliques (T/F) (default = false)
    • Print the number of candidate cliques at each iteration, revealing the distribution of clique sizes
  • window_size (unsigned int) (default = 8192)
    • Size of the subset of clique lists that will be solved depth-first for the implementation using windowing
  • sort_sublists (T/F) (default = true)
    • Sort the vertices (by degree or core number) before finding 2-cliques, such that the highly-connected vertices' candidate cliques will be searched first in windowed version
  • expected_max (unsigned int) (default = 0)
    • Expected maximum clique size. Used for validation.
  • overall_perf (T/F) (default = true)
    • If true, don't collect metrics that will interfere with runtime results.
  • lowerbound (unsigned int) (default = 0)
    • Known lower bound on the size of the maximum clique
    • Can also be used for testing performance/pruning with improved heuristic
  • orientation (string) options: index, degree (default = "degree")
    • Specify which edge to keep out of each undirected pair
    • "index" keeps the edge (u,v) where u < v
    • "degree" keeps edge (u,v) in which deg(u) < deg(v)
  • order_candidates (T/F) (default = true)
    • If true, sort the candidates within sublists (destination vertices) from low to high degree.
    • Never sort candidates if orientation is by index (to simplify implementation).

Pre-set gunrock parameters

  • undirected=true
    • Require the graph to be undirected to make sure we don't miss any cliques
    • Creates an edge (dest, src) for each (src, dest) in the input graph if it doesn't already exist
  • sort-csr=true
    • Sorts the edges within each src vertex's list (e.g., if you have (1,3) and (1,6), without sorting, it may be that the 6 will be stored first in vertex 1's segment without this flag)