Assignment Recommendation API - grouped planning entities

[EverettZ]

Hey all! Enterprise customer here.

My scheduling domain relates to machine tasking. Specifically, coordinated machine tasking. A while back we modeled the problem so that our planning entity is an individual task, with an ID attached to it that correlates it to other tasks. Roughly 1-6 tasks can share an ID. But the vast majority of tasks are in groups of 3.

I'd love to take advantage of the Assignment Recommendation API without having to change my model to be "coordinated/grouped". However, I cant see a way to utilize the API to pass more than one planning entity at a time.

Am I missing something, or is this a feature gap that can be solved? I'm about nearing time to figure out an implementation for this, but am struggling with how to accomplish this without some degree of brute-force.

[triceo]

Hello @EverettZ!

You are correct that the Recommendation API does not support more than one entity. It is not a feature gap though, it is by design - if it did support more than one entity, it would turn back to a complex issue, requiring the full power of the solver, and enough time to run it.

The goal of the Recommendation API is to give answers super-quickly, to problems where we do not run the risk of combinatorial explosion. We do not need to use metaheuristics here, when a simple exhaustive search will do. Once multiple entities are involved, the correct approach is to run the solver - and the solver can not give you a range of assignments to choose from. It only works with the best solution.

Summary: it's either one or the other. Recommendations for simple problems, full solver for bigger tasks.

Side note: we can of course continue our conversation here, but I would be remiss if I did not mention that you can also use your Timefold customer status to contact our Enterprise support. :-)

[EverettZ]

Understood. If my planning entity didn't need to be grouped, this would work as is. I'll need to write my own exhaustive search (still using score calc), or change the model.

[TomCools]

Would Pinning most/all of the other entities and doing a run work for you? That way, only unpinned entities will be scheduled and that should happen pretty fast. I do not think there is a need for writing your own exhaustive search.