Closed torressa closed 2 years ago
Tell a lie, #101 not sure if fixed. Do we have to check upper TW for departure times? I don't we can as it is not enforced at the subproblem level.
Great job @torressa ! Thanks for all the work.
There is a test regarding issue #99 with non elementary routes that still fails. Not sure whats going on, needs more investigation.
Thanks man! #99 is on cspy side, the next release works!
@Kuifje02 Hey Romain! I think this is ready for a review now. I'm using black (the formatter) so hopefully won't be looking at much useless diff after this. It would be cool if you could test a little bit too. The unit test for issue 99 is the only one that doesn't pass (but this works fine with the new version of cspy). Same for some of the examples (failing with cspy-v101, working with the new version). I think this may be a source of new issues (missing nodes and such), so I've included an option for people to turn this off
elementary
(and removed theexact
option).I've been running mostly regression tests wrt cspy-v101 and vrpy-v051 (previous releases). Results on the Solomon instances (![image](https://user-images.githubusercontent.com/23246013/163586160-e112bc53-b178-4ece-afd6-a3999f1d3f56.png)
benchmarks/tests/test_cvrptw_solomon_range.py
) and it looks pretty good (times per subproblem averaged over 10 runs) (x-axis should start at 7-69 as it's the number of nodes):The grey (
fwd_new
) and orange lines (both_new
) are the new unreleased version of cspy (which I'm planning on releasing soon) with this new non-elementary route stuff. Performance is more stable (fwd_new
being slightly better most of the time). I've also included the previous release of vrpy with cspy (vrpy_051_cspy_101
), which is flailing about all over the place (also giving me some memory issues but I think that's cause of pytest).I think this also contains fixes #101 and related.