Closed magnushiie closed 1 year ago
Wow, we did miss that. Thank you Started #29
One-To-One
bench we suggest generate graph vertices based on n=2^k
- You can take a look already.Other works in that PR would be:
One-To-Many
it depends. May be 2 benchmarks: one is for different graph sizes; second - number of target nodesWell, we've just updated PR. Summary:
BenchmarkShortestPath
- generates graph with n=2^k
, computes its contraction hierarchies, takes two random vertices and computes shortest path between them.BenchmarkStaticCaseShortestPath
- just single run on predefined graph (187k vertices) with computing of contraction hierarchiesBenchmarkShortestPathOneToMany
(and old derivative ...OldWay...
) - generates graph with n=2^k
, computes its contraction hierarchies, takes one random vertex as source and five random vertices as targets and computes shortest pathsBenchmarkTargetNodesShortestPathOneToMany
(and old derivative ...OldWay...
) - takes predefined graph (187k vertices), computes its contraction hierarchies, takes predefined vertex as source, takes random number (1 up to 2^k
) of random vertices, computes shortest pathsBenchmarkStaticCaseShortestPathOneToMany
(and old derivative ...OldWay...
) - just single run (1 predefined source - 5 predefined targets) on predefined graph (187k vertices) with computing of contraction hierarchiesI believe bug
has been fixed, but for sure now we can improve tests (e.g. better random graph generation)
Describe the bug The benchmarks (like BenchmarkShortestPath) execute 12 iterations with
k = 0..11
, calculate ann = 2^k
, but then don't use either of them in the benchmark except for the title.To Reproduce Run the benchmarks and read source code.
Expected behavior Either
n
ork
should be used in the benchmark core (not sure what these could be forBenchmarkShortestPath
, forBenchmarkShortestPathOneToMany
it could be the number of target nodes).