CP-SAT - Feasible hint not being used and solver doesn't find a solution

[Nara Torres Moreira]

Hi there,

I have a model that I'm solving with CP-SAT and I pass a complete and feasible initial solution to the solver. However, the time limit (10 min) is reached and the solver finishes with status UNKOWN, which I have been interpreting so far as "no solution found". I noticed though that in the CpSolverResponse summary the objective value is correspondent to the value of my hint.

Please find attached the proto file and the log. I tested it multiple times and every time I had the same behaviour, so maybe you can reproduce.

Is there any way to force the solver to give some priority to the hints? Or any other suggestion?

Thanks,

Nara

[Laurent Perron]

Here is the output on the main branch

...

Preloading model.

[Symmetry] Graph for symmetry has 316577 nodes and 1068431 arcs.

[Symmetry] Symmetry computation done. time: 0.152075 dtime: 0.193508

[Symmetry] #generators: 3, average support size: 8084

#Bound  11.63s best:inf   next:[0,718]    initial_domain

#Model  11.80s var:68920/68920 constraints:111880/111880

Starting Search at 11.80s with 8 workers.

6 full subsolvers: [default_lp, fixed, no_lp, max_lp, core, reduced_costs]

Interleaved subsolvers: [feasibility_pump, rnd_var_lns_default, rnd_cst_lns_default, graph_var_lns_default, graph_cst_lns_default, rins_lns_default, rens_lns_default]

#1      16.47s best:346   next:[0,345]    no_lp [hint] fixed_bools:0/69265

#2      16.52s best:45    next:[0,44]     no_lp fixed_bools:0/69265

#3      22.74s best:44    next:[0,43]     rnd_var_lns_default(d=0.71 s=12 t=0.10 p=1.00)

I assume this was a performance problem and the solver never finished the root node in 600s.

Laurent Perron | Operations Research | lpe...@google.com | (33) 1 42 68 53 00

--
You received this message because you are subscribed to the Google Groups "or-tools-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to or-tools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/408ab3b9-614e-44b8-a24c-71cde0214d06n%40googlegroups.com.

[Nara Torres Moreira]

Thanks Laurent. Are there any plans to release a new version soon? I see that 9.3 was released back in mid March.

Also, unless there are more than one issue, I don't think this is due to performance, because one of the runs I did for this same model (using 6 workers) had the opposite behaviour: it finished with 5 min also with status UNKOWN where time limit was 10 min. Maybe this was already fixed in the main branch though, who knows...

[Laurent Perron]

We have been trying to release 9.4 since mid June. We want to integrate arm64 builds (Linux and Mac M1). It cannot be done after the fact for java and .NET. 

We are getting there. note there is an open issue on GitHub that tracks it. 

To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/40173d67-a5a3-419b-82cd-18caa5600792n%40googlegroups.com.