CP-SAT vs Kalis and CP Optimizer?

[Stuart Rogers]

Are there any benchmarks comparing the performance of CP-SAT against the commercial CP solvers Kalis and CP Optimizer (in CPLEX Optimization Studio)?

Kalis: https://www.artelys.com/solvers/kalis/

CP Optimizer: https://www.ibm.com/analytics/cplex-cp-optimizer

[Laurent Perron]

On pure CP problems:

There are 2 CP challenges: minizinc and xcsp3

CP-SAT participates in the minizinc challenge every year and has won all gold medals in the track it participated in the last years.

None of the other 2 solvers have ever participated in any challenge.

I have not tested Kalis, but knowing it is not SAT/LCG based, and seeing how traditional CP solvers perform against CP-SAT at the minizinc challenge, I can safely say it is not competitive.

CPO is a very good solver. I actually wrote version 1.0. The pure CP part is very good, but non competitive.

On Scheduling problems:

There are a few papers around there. Unfortunately, they never compare to CP-SAT with 8 threads or more. And the performance gap between 1-4 workers and 8+ workers is huge.

So they are not very informative. 

My experiments show that on academic problems (PSPLIB, JSSP), we are better than CPO. This is thanks to recent work on scheduling LNS and scheduling cuts.

There are cases where I expect CPO to be better:

  - large scale scheduling (1k tasks and more). They have dedicated heuristics we do not have

  - scheduling with setup/transition times and costs. We are better at solving them, but the model is huge compared to CPO, and presolve is much slower. So there is a category of problems where CPO proves optimality in 20s, while CP-SAT presolves the problem in 1 minute, then proves optimality in 5s.

I hope this helps.

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/de2b8c69-149b-4481-ad47-734ea3d98fe1n%40googlegroups.com.

[Stuart Rogers]

Thanks for all the information. How does CP-SAT compare against SCIP, which supposedly has a FlatZinc interface (https://www.minizinc.org/software.html)? For some reason, SCIP is not included in the MiniZinc Challenge: https://www.minizinc.org/challenge.html 

SCIP: https://www.scipopt.org/

[Laurent Perron]

On CP problems, CP-sat beats gurobi, so SCIP...

On linear integer problems, CP-sat beats scip, is not far from cplex, and sometimes win against gurobi, but not often. 

To view this discussion on the web visit https://groups.google.com/d/msgid/or-tools-discuss/d9c3e296-1062-4146-9051-c6d14a5877aen%40googlegroups.com.