Determining minimum number of vehicles about euro-neurips-2022 HOT 10 CLOSED

Ran over 10 seeds, qualification phase:

• 164474 with nbVeh computed by bin packing heuristic
• 164469 with nbVeh = nbClients

No difference. :-) I'll make a PR later to change this everywhere.

from euro-neurips-2022.

How bad is it to set nbVeh = nbClients (or dispatched clients) in each epoch? Does that (meaningfully) affect solution quality?

from euro-neurips-2022.

I tried this, but this still led to infeasible solutions in some epoch/seed combinations. Since you succesfully submitted the dynamic solver, did this fix work for you? @N-Wouda

I realize that this is stupid of course because a single-route-per-client solution should be feasible! Weird.

from euro-neurips-2022.

I think that using nbVeh=nbClients is fine (or divided by 2 or 3). The performance loss is negligible because the instances are small enough and hgs solver skips all empty routes anyway.

I don't understand why I had so many issues with this when testing. Now that everything seems to work I'll close this issue.

from euro-neurips-2022.

I'm re-opening this issue for now because I think we should make sure that we use nbVeh=nbClients everywhere. And of course make sure to test that that does not impact performance too much.

from euro-neurips-2022.

The motivation is that we should make sure we can definitely solve the instances of the final round. But we won't know 'bin packing + 30' will be sufficient there, and cannot find out before it's too late.

from euro-neurips-2022.

I'm currently running on the cluster (10 seeds, qualification time limits):

• nbVeh computed using bin packing heuristic
• nbVeh = nbClients

from euro-neurips-2022.

Does setting nbVeh = nbClients hurt performance a lot?

from euro-neurips-2022.

I'll know more in 1.5 hours-ish!

from euro-neurips-2022.

That's a relief!

from euro-neurips-2022.