Commit Graph

14 Commits

Author SHA1 Message Date
Dan Kortschak
91a06ac64c graph/path: reduce allocations in path reconstructions
│   old.bench   │               new.bench               │
                                              │    sec/op     │    sec/op      vs base                │
ShortestAlts/AllTo_100×2|0.5(17)-8              12.380µ ±  9%   8.577µ ±   6%  -30.72% (p=0.000 n=10)
ShortestAlts/AllTo_1000×2|0.5(51)-8              74.87µ ± 17%   47.66µ ±   1%  -36.35% (p=0.000 n=10)
ShortestAlts/AllTo_100×4|0.25(53)-8              25.50µ ±  3%   15.35µ ±   3%  -39.82% (p=0.000 n=10)
ShortestAlts/AllTo_1000×4|0.25(574)-8            556.5µ ±  5%   279.2µ ±   4%  -49.83% (p=0.000 n=10)
ShortestAlts/AllTo_100×16|0.1(76)-8              35.79µ ±  9%   21.71µ ±   3%  -39.35% (p=0.000 n=10)
ShortestAlts/AllTo_1000×16|0.1(822)-8            851.8µ ±  6%   400.5µ ±   6%  -52.98% (p=0.000 n=10)
AllShortest/AllBetween_100×2|0.5(17)-8          12.482µ ±  6%   8.573µ ±   3%  -31.32% (p=0.000 n=10)
AllShortest/AllBetween_1000×2|0.5(51)-8          80.74µ ±  2%   52.90µ ±   4%  -34.48% (p=0.000 n=10)
AllShortest/AllBetween_100×4|0.25(53)-8          22.54µ ±  4%   16.17µ ±   4%  -28.26% (p=0.000 n=10)
AllShortest/AllBetween_1000×4|0.25(574)-8        543.8µ ±  3%   326.7µ ±   5%  -39.93% (p=0.000 n=10)
AllShortest/AllBetween_100×16|0.1(76)-8          31.21µ ±  1%   22.45µ ±   3%  -28.07% (p=0.000 n=10)
AllShortest/AllBetween_1000×16|0.1(822)-8        678.4µ ±  1%   402.5µ ±   3%  -40.67% (p=0.000 n=10)
ShortestAlts/AllToFunc_100×2|0.5(17)-8                          5.892µ ±   2%
ShortestAlts/AllToFunc_1000×2|0.5(51)-8                         40.81µ ±   2%
ShortestAlts/AllToFunc_100×4|0.25(53)-8                         9.178µ ±   2%
ShortestAlts/AllToFunc_1000×4|0.25(574)-8                       214.6µ ±   2%
ShortestAlts/AllToFunc_100×16|0.1(76)-8                         12.62µ ±   3%
ShortestAlts/AllToFunc_1000×16|0.1(822)-8                       310.7µ ±   3%
AllShortest/AllBetweenFunc_100×2|0.5(17)-8                      5.713µ ±   1%
AllShortest/AllBetweenFunc_1000×2|0.5(51)-8                     45.68µ ±   3%
AllShortest/AllBetweenFunc_100×4|0.25(53)-8                     9.775µ ±   1%
AllShortest/AllBetweenFunc_1000×4|0.25(574)-8                   245.2µ ±   5%
AllShortest/AllBetweenFunc_100×16|0.1(76)-8                     13.76µ ±   1%
AllShortest/AllBetweenFunc_1000×16|0.1(822)-8                   314.7µ ±   1%
geomean                                          82.87µ         43.03µ         -38.14%

                                              │   old.bench   │              new.bench               │
                                              │     B/op      │     B/op      vs base                │
ShortestAlts/AllTo_100×2|0.5(17)-8              14.148Ki ± 0%   9.727Ki ± 0%  -31.25% (p=0.000 n=10)
ShortestAlts/AllTo_1000×2|0.5(51)-8              189.8Ki ± 0%   127.5Ki ± 0%  -32.80% (p=0.000 n=10)
ShortestAlts/AllTo_100×4|0.25(53)-8              22.32Ki ± 0%   14.99Ki ± 0%  -32.83% (p=0.000 n=10)
ShortestAlts/AllTo_1000×4|0.25(574)-8           1272.9Ki ± 0%   681.9Ki ± 0%  -46.42% (p=0.000 n=10)
ShortestAlts/AllTo_100×16|0.1(76)-8              33.23Ki ± 0%   22.66Ki ± 0%  -31.79% (p=0.000 n=10)
ShortestAlts/AllTo_1000×16|0.1(822)-8           1799.9Ki ± 0%   953.2Ki ± 0%  -47.04% (p=0.000 n=10)
AllShortest/AllBetween_100×2|0.5(17)-8          13.039Ki ± 0%   8.617Ki ± 0%  -33.91% (p=0.000 n=10)
AllShortest/AllBetween_1000×2|0.5(51)-8          181.5Ki ± 0%   119.3Ki ± 0%  -34.29% (p=0.000 n=10)
AllShortest/AllBetween_100×4|0.25(53)-8          21.34Ki ± 0%   14.01Ki ± 0%  -34.35% (p=0.000 n=10)
AllShortest/AllBetween_1000×4|0.25(574)-8       1264.8Ki ± 0%   673.8Ki ± 0%  -46.72% (p=0.000 n=10)
AllShortest/AllBetween_100×16|0.1(76)-8          32.24Ki ± 0%   21.68Ki ± 0%  -32.76% (p=0.000 n=10)
AllShortest/AllBetween_1000×16|0.1(822)-8       1791.8Ki ± 0%   945.1Ki ± 0%  -47.25% (p=0.000 n=10)
ShortestAlts/AllToFunc_100×2|0.5(17)-8                          6.922Ki ± 0%
ShortestAlts/AllToFunc_1000×2|0.5(51)-8                         119.8Ki ± 0%
ShortestAlts/AllToFunc_100×4|0.25(53)-8                         9.531Ki ± 0%
ShortestAlts/AllToFunc_1000×4|0.25(574)-8                       611.1Ki ± 0%
ShortestAlts/AllToFunc_100×16|0.1(76)-8                         13.12Ki ± 0%
ShortestAlts/AllToFunc_1000×16|0.1(822)-8                       870.7Ki ± 0%
AllShortest/AllBetweenFunc_100×2|0.5(17)-8                      5.812Ki ± 0%
AllShortest/AllBetweenFunc_1000×2|0.5(51)-8                     111.5Ki ± 0%
AllShortest/AllBetweenFunc_100×4|0.25(53)-8                     8.547Ki ± 0%
AllShortest/AllBetweenFunc_1000×4|0.25(574)-8                   603.0Ki ± 0%
AllShortest/AllBetweenFunc_100×16|0.1(76)-8                     12.14Ki ± 0%
AllShortest/AllBetweenFunc_1000×16|0.1(822)-8                   862.6Ki ± 0%
geomean                                          126.5Ki        68.27Ki       -37.99%

                                              │  old.bench  │              new.bench              │
                                              │  allocs/op  │  allocs/op   vs base                │
ShortestAlts/AllTo_100×2|0.5(17)-8              141.00 ± 0%    94.00 ± 0%  -33.33% (p=0.000 n=10)
ShortestAlts/AllTo_1000×2|0.5(51)-8              420.0 ± 0%    269.0 ± 0%  -35.95% (p=0.000 n=10)
ShortestAlts/AllTo_100×4|0.25(53)-8              279.0 ± 0%    174.0 ± 0%  -37.63% (p=0.000 n=10)
ShortestAlts/AllTo_1000×4|0.25(574)-8           2.888k ± 0%   1.741k ± 0%  -39.72% (p=0.000 n=10)
ShortestAlts/AllTo_100×16|0.1(76)-8              395.0 ± 0%    244.0 ± 0%  -38.23% (p=0.000 n=10)
ShortestAlts/AllTo_1000×16|0.1(822)-8           4.128k ± 0%   2.485k ± 0%  -39.80% (p=0.000 n=10)
AllShortest/AllBetween_100×2|0.5(17)-8          136.00 ± 0%    89.00 ± 0%  -34.56% (p=0.000 n=10)
AllShortest/AllBetween_1000×2|0.5(51)-8          415.0 ± 0%    264.0 ± 0%  -36.39% (p=0.000 n=10)
AllShortest/AllBetween_100×4|0.25(53)-8          275.0 ± 0%    170.0 ± 0%  -38.18% (p=0.000 n=10)
AllShortest/AllBetween_1000×4|0.25(574)-8       2.884k ± 0%   1.737k ± 0%  -39.77% (p=0.000 n=10)
AllShortest/AllBetween_100×16|0.1(76)-8          391.0 ± 0%    240.0 ± 0%  -38.62% (p=0.000 n=10)
AllShortest/AllBetween_1000×16|0.1(822)-8       4.124k ± 0%   2.481k ± 0%  -39.84% (p=0.000 n=10)
ShortestAlts/AllToFunc_100×2|0.5(17)-8                         71.00 ± 0%
ShortestAlts/AllToFunc_1000×2|0.5(51)-8                        211.0 ± 0%
ShortestAlts/AllToFunc_100×4|0.25(53)-8                        114.0 ± 0%
ShortestAlts/AllToFunc_1000×4|0.25(574)-8                     1.156k ± 0%
ShortestAlts/AllToFunc_100×16|0.1(76)-8                        160.0 ± 0%
ShortestAlts/AllToFunc_1000×16|0.1(822)-8                     1.652k ± 0%
AllShortest/AllBetweenFunc_100×2|0.5(17)-8                     66.00 ± 0%
AllShortest/AllBetweenFunc_1000×2|0.5(51)-8                    206.0 ± 0%
AllShortest/AllBetweenFunc_100×4|0.25(53)-8                    110.0 ± 0%
AllShortest/AllBetweenFunc_1000×4|0.25(574)-8                 1.152k ± 0%
AllShortest/AllBetweenFunc_100×16|0.1(76)-8                    156.0 ± 0%
AllShortest/AllBetweenFunc_1000×16|0.1(822)-8                 1.648k ± 0%
geomean                                          649.3         336.5       -37.70%
2023-04-24 20:31:56 +09:30
Dan Kortschak
8f6a7707f4 graph/internal/ordered: replace slice sort types with sort.Slice calls 2021-10-14 07:01:58 +10:30
Dan Kortschak
3d5d9d4a35 graph/path: allow incremental Bellman-Ford search
The impact on the current implementation is negligible.

name                          old time/op    new time/op    delta
BellmanFordFrom/500_tenth-8     4.14ms ± 1%    4.11ms ± 1%  -0.73%  (p=0.001 n=10+9)
BellmanFordFrom/1000_tenth-8    16.3ms ± 1%    16.1ms ± 1%  -0.82%  (p=0.000 n=10+10)
BellmanFordFrom/2000_tenth-8    66.2ms ± 1%    60.9ms ± 0%  -8.08%  (p=0.000 n=10+9)
BellmanFordFrom/500_half-8      18.0ms ± 5%    16.9ms ± 1%  -6.28%  (p=0.000 n=10+9)
BellmanFordFrom/1000_half-8     68.5ms ± 2%    67.4ms ± 1%  -1.50%  (p=0.000 n=9+8)
BellmanFordFrom/2000_half-8      281ms ± 1%     276ms ± 1%  -1.45%  (p=0.000 n=9+10)
BellmanFordFrom/500_full-8      33.4ms ± 0%    33.0ms ± 1%  -1.29%  (p=0.000 n=9+9)
BellmanFordFrom/1000_full-8      137ms ± 1%     133ms ± 0%  -3.11%  (p=0.000 n=8+9)
BellmanFordFrom/2000_full-8      549ms ± 1%     546ms ± 1%    ~     (p=0.065 n=10+9)

name                          old alloc/op   new alloc/op   delta
BellmanFordFrom/500_tenth-8      142kB ± 0%     142kB ± 0%    ~     (p=0.110 n=10+9)
BellmanFordFrom/1000_tenth-8     284kB ± 0%     284kB ± 0%    ~     (p=0.382 n=10+10)
BellmanFordFrom/2000_tenth-8     624kB ± 0%     624kB ± 0%    ~     (p=0.956 n=10+10)
BellmanFordFrom/500_half-8       142kB ± 0%     142kB ± 0%    ~     (p=0.666 n=10+10)
BellmanFordFrom/1000_half-8      284kB ± 0%     284kB ± 0%    ~     (p=0.070 n=10+8)
BellmanFordFrom/2000_half-8      624kB ± 0%     624kB ± 0%    ~     (p=0.290 n=10+10)
BellmanFordFrom/500_full-8       142kB ± 0%     142kB ± 0%    ~     (p=0.234 n=9+10)
BellmanFordFrom/1000_full-8      284kB ± 0%     284kB ± 0%    ~     (p=0.051 n=10+9)
BellmanFordFrom/2000_full-8      624kB ± 0%     624kB ± 0%    ~     (p=0.790 n=9+10)

name                          old allocs/op  new allocs/op  delta
BellmanFordFrom/500_tenth-8      2.03k ± 0%     2.03k ± 0%    ~     (all equal)
BellmanFordFrom/1000_tenth-8     4.03k ± 0%     4.03k ± 0%    ~     (all equal)
BellmanFordFrom/2000_tenth-8     8.03k ± 0%     8.03k ± 0%    ~     (all equal)
BellmanFordFrom/500_half-8       2.03k ± 0%     2.03k ± 0%    ~     (p=0.294 n=10+8)
BellmanFordFrom/1000_half-8      4.03k ± 0%     4.03k ± 0%    ~     (all equal)
BellmanFordFrom/2000_half-8      8.03k ± 0%     8.03k ± 0%    ~     (all equal)
BellmanFordFrom/500_full-8       2.03k ± 0%     2.03k ± 0%    ~     (p=1.000 n=10+10)
BellmanFordFrom/1000_full-8      4.03k ± 0%     4.03k ± 0%    ~     (all equal)
BellmanFordFrom/2000_full-8      8.03k ± 0%     8.03k ± 0%    ~     (p=1.000 n=10+10)

The performance of the lazy implementation is a little worse and allocations
are a lot worse than the eager implementation. This reflects that the benchmark
is performed on a graph with a single connected component which gains no benefit
from the incremental approach but suffers the cost of repeated reallocation for
appends.

name                          new time/op    inc time/op    delta
BellmanFordFrom/500_tenth-8     4.11ms ± 1%    4.28ms ± 1%   +4.05%  (p=0.000 n=9+10)
BellmanFordFrom/1000_tenth-8    16.1ms ± 1%    16.0ms ± 6%     ~     (p=0.143 n=10+10)
BellmanFordFrom/2000_tenth-8    60.9ms ± 0%    62.8ms ± 0%   +3.13%  (p=0.000 n=9+10)
BellmanFordFrom/500_half-8      16.9ms ± 1%    17.5ms ± 1%   +3.97%  (p=0.000 n=9+10)
BellmanFordFrom/1000_half-8     67.4ms ± 1%    70.0ms ± 2%   +3.76%  (p=0.000 n=8+9)
BellmanFordFrom/2000_half-8      276ms ± 1%     285ms ± 0%   +2.91%  (p=0.000 n=10+10)
BellmanFordFrom/500_full-8      33.0ms ± 1%    35.5ms ± 5%   +7.40%  (p=0.000 n=9+10)
BellmanFordFrom/1000_full-8      133ms ± 0%     137ms ± 2%   +3.48%  (p=0.000 n=9+9)
BellmanFordFrom/2000_full-8      546ms ± 1%     592ms ± 6%   +8.42%  (p=0.000 n=9+10)

name                          new alloc/op   inc alloc/op   delta
BellmanFordFrom/500_tenth-8      142kB ± 0%     182kB ± 0%  +27.84%  (p=0.000 n=9+10)
BellmanFordFrom/1000_tenth-8     284kB ± 0%     363kB ± 0%  +27.94%  (p=0.000 n=10+10)
BellmanFordFrom/2000_tenth-8     624kB ± 0%     893kB ± 0%  +43.26%  (p=0.000 n=10+10)
BellmanFordFrom/500_half-8       142kB ± 0%     182kB ± 0%  +27.83%  (p=0.000 n=10+10)
BellmanFordFrom/1000_half-8      284kB ± 0%     363kB ± 0%  +27.90%  (p=0.000 n=8+10)
BellmanFordFrom/2000_half-8      624kB ± 0%     893kB ± 0%  +43.25%  (p=0.000 n=10+10)
BellmanFordFrom/500_full-8       142kB ± 0%     182kB ± 0%  +27.82%  (p=0.000 n=10+10)
BellmanFordFrom/1000_full-8      284kB ± 0%     363kB ± 0%  +27.89%  (p=0.000 n=9+10)
BellmanFordFrom/2000_full-8      624kB ± 0%     893kB ± 0%  +43.23%  (p=0.000 n=10+10)

name                          new allocs/op  inc allocs/op  delta
BellmanFordFrom/500_tenth-8      2.03k ± 0%     2.10k ± 0%   +3.50%  (p=0.000 n=8+8)
BellmanFordFrom/1000_tenth-8     4.03k ± 0%     4.12k ± 0%   +2.39%  (p=0.000 n=10+10)
BellmanFordFrom/2000_tenth-8     8.03k ± 0%     8.17k ± 0%   +1.77%  (p=0.000 n=10+10)
BellmanFordFrom/500_half-8       2.03k ± 0%     2.10k ± 0%   +3.49%  (p=0.000 n=8+10)
BellmanFordFrom/1000_half-8      4.03k ± 0%     4.12k ± 0%   +2.36%  (p=0.000 n=9+10)
BellmanFordFrom/2000_half-8      8.03k ± 0%     8.17k ± 0%   +1.77%  (p=0.000 n=8+10)
BellmanFordFrom/500_full-8       2.03k ± 0%     2.10k ± 0%   +3.50%  (p=0.000 n=10+10)
BellmanFordFrom/1000_full-8      4.03k ± 0%     4.12k ± 0%   +2.36%  (p=0.000 n=9+10)
BellmanFordFrom/2000_full-8      8.03k ± 0%     8.17k ± 0%   +1.76%  (p=0.000 n=10+10)
2021-01-07 06:43:24 +10:30
Dan Kortschak
9ae0689dae graph/path: add option to retain all paths from single-source path finding routines 2020-07-10 07:22:35 +09:30
Dan Kortschak
3bedd58622 graph/path: properly report negative cycles in Shortest.To
We still use the maxEdges condition to allow a better exploration of the graph
before terminating.
2020-07-07 06:20:59 +09:30
Dan Kortschak
8d4cfcdeb8 graph/path: fix error messages 2020-07-02 15:56:32 +09:30
Egon Elbre
b6ceadb9b7 graph/path: make tests parallel 2020-03-16 16:10:59 +02:00
Dan Kortschak
6b03bc22e1 graph: use int64 for node retrieval 2018-03-24 09:21:44 +10:30
kortschak
805531d142 all: change capitalization of gonum in license header 2017-11-02 06:54:08 +10:30
kortschak
7ba61f0ead graph/...: remove Weight method from Edge 2017-08-24 10:21:15 +09:30
kortschak
15ecc07d40 graph/path/...: update for int64 IDs 2017-07-02 08:38:42 +09:30
kortschak
7af70dd796 all: fix many go vet errors 2017-06-20 16:29:35 +09:30
Brendan Tracey
d33397aa65 all: change import paths 2017-05-23 00:03:03 -06:00
Brendan Tracey
059a6c43f9 graph: imported graph as a subtree 2017-05-23 00:02:59 -06:00