Optimization Framework

Learn
#1

Are there any good optimization frameworks that:

  • have predefined search functions but also allows custom functions
  • limit search “work” time and compute used
  • status updates (latest found results)
  • handles discontinuous results spaces
  • data boundaries (for when the grading function is expensive and the seed can be disqualified because it will produce an out of bound result)
1 Like
#2

Yes, but but but.

What kind of functions to be optimized? Are they known analytically or only from samples? Is the gradient available? Are the parameters continuous or discrete?

Also, if you bound compute a priori it’s hard to tell how much useful progress the optimizer will make within the budget.

2 Likes
#3

Links Please! Great questions to consider when examining a framework.

#4

For example dataHaskell : Current Environment

Could share what sort of optimization problems you have in mind?

#5

2D packing of polygons with internal voids on multiple planes bounded by varying polygons

#6

sounds NP-hard. good luck!

1 Like
#7

Wouldn’t it be great if the easy stuff paid

#8

It is, even packing simple polygons in a square container is NP-hard. Depending on the type of operations you allow it may even be ER-complete. See #13474 - Framework for $\exists \mathbb{R}$-Completeness of Two-Dimensional Packing Problems

By the way, this years CG:SHOP optimization challenge was about packing problems, it may give some inspiration on tactics to try :).

1 Like