do anybody know about some (speed) comparison of Java CSP systems? I was searching web and I have found only minizinc challenge, which has only a few participants (and JaCoP is the only Java implementation) and only a very few articles which state in one sentence that they think that JaCoP is faster than choco (which is also mine finding (I was testing it on graph-coloring)).
I am not aware of anything systematic. The comparisons are not always easy as it is always the case that you can find problems for which your solver is the fastest in town.
In JaCoP, we focus on efficient and strong propagation of global constraints. If you use the right constraints and the right model you can be orders of magnitiude better than the competition. We also focus on being efficient for large problems often at the cost of being slow on small problems.
Unfortunately, those type of comparisons are better left to users not solver authors. Even then the user may forget to turn off one feature in one of the solvers and comparison is not fair. I still remember somebody complaining about JaCoP versus other solver, while they did not realize that they forced JaCoP to remember millions of solutions while the other solver was just counting the solutions and not storing them. Little thing but caused the conclusions of the user study to be completely wrong.
Log in to post a comment.