Keep it simple, just one call to solve every model.

*sigh* We really need to teach this new crop the term "no free lunch". Again.

I personally disagree with "no free lunch"; (for the uninitiated, "no free lunch" refer to the fact for any deterministic algorithm, there exist a problem that will force the algorithm to go through the entire solution space to find the optimal solution, with every single other possible algorithm beating it (https://en.wikipedia.org/wiki/No_free_lunch_theorem)). For many planning problems, finding a good enough solution is sufficient, and there are many optimization algorithms that work for a wide variety of problems and provide a good enough solution in reasonable time. Different algorithms are better for different problems (ex: Metaheuristic (ex: Late Acceptance) Solvers beats MIP Solvers on vehicle routing, whereas MIP Solvers beat Metaheuristic Solvers on Employee Scheduling and Bin Packing. But both Metaheuristic and MIP Solvers provider good enough solutions for both vehicle routing and bin packing.

▲

uoaei 5 hours ago | parent [-]

No free lunch theorem has nothing to say about approximate solutions, so I'm really not sure what you're going on about.

OR-tools is almost exclusively linear programming which according to its strict assumptions converges more or less trivially, assuming a correctly composed program.

Which means if you're paying for it "as a service" you all but deserve to lose that money.

> Different algorithms are better for different problems

So... why does your rhetorical style have such oppositional tone if you're just going to reaffirm the no free lunch theorem?

	▲	cchianel 3 hours ago \| parent [-]
		Look at it this way: I am arguing against "No Free Lunch theorem says an optimization algorithm cannot solve all problems because for some problems it performs worse than other algorithms"; I am arguing approximate solutions are good enough, and in practice a wide variety of optimization algorithms find good enough solutions despite being worse than others algorithms for the problem class. Moreover, some algorithms/solvers can be configured, which fundamentally change the direction the solving takes (for example, a custom phase that uses your domain knowledge of the particular problem to get a good enough initial solution to be improved upon) (Side note: I am NOT affiliated with this post/project; from the website I don't really see a value add for it, especially since the site is lacking so many details).