7

I am looking for examples to understand what kind of benefits a model-and-run framework solver offers that a more programmatic solver doesn't do or doesn't do as well.

What distinguishes a model-and-run framework?

Context: I am trying to understand what OptaPlanner is missing to be used as a model-and-run solver. I presume it's more than just Jupyter notebook integration (which is already supported for OptaPy)?

Geoffrey De Smet
  • 4,851
  • 10
  • 34
  • 1
    Interesting. Reading model-and-run, i always link that to black-box solver-performance given my model: thanks presolve/heuristics/x -> MIP, thanks CDCL/VSIDS/x -> SAT, less thanks -> CP (to be taken with a grain of salt). You seem to target the toolchain more? For what it's worth: I do think the CP-community discussed this topic a lot (combining our views?); maybe it serves as a start: Puget, "Constraint programming next challenge: Simplicity of use." 2004 | Freuder "Progress towards the holy grail." 2018 and many others.Maybe also publications related to Local-Solver or Minizinc – sascha Sep 28 '22 at 13:53
  • Thank you. "Configuration By Exception" goes pretty far for such black box solving, in my experience. It's probably about the toolchain indeed. The fact that OptaPlanner's quickstart starts with setting up a build file (pom.xml) is telling: it is comfortable for programmers, but not for data scientists. – Geoffrey De Smet Sep 28 '22 at 14:18
  • 1
    I think there could also be a matter of "genericity". How easy is it to solve any problem with OptaPlanner? Take any OR paper, after the description of the problem, the authors always start by writing the problem as a MILP, MINLP or CP. These languages gives some universal way to describe many problems. And actually, even CP had a hard time to be considered a valid way to describe a problem if I remember well – fontanf Sep 28 '22 at 19:46
  • Regarding benefits, this "universal language approach" only requires to learn a single way to formulate a problem to be able to formulate most of the other ones, be they routing, scheduling, time-tabling, packing or anything else. – fontanf Sep 28 '22 at 19:51
  • How many people can read this "universal language" of math equations? I've been reading it for 16 years - implemented various OR competitions based on such input alone - yet find it difficult as a communication method outside this community. – Geoffrey De Smet Oct 03 '22 at 07:17

0 Answers0