Get BlackBoxOptimizationBenchmarking.jl updated to Optimization.jl's interface and run on SciMLBenchmarks

[ChrisRackauckas]

@Vaibhavdixit02 can you take a look over the next week into updating/upgrading https://github.com/jonathanBieler/BlackBoxOptimizationBenchmarking.jl to Optimization.jl's interface? It would nice to have this easily re-runnable over all of Optimization.jl's methods given that it has been a pretty useful benchmark for us in the past. It will also show the completeness of Optimization.jl's interface to have something comprehensive like this.

CC @jonathanBieler

[EdwardBerman]

Hi all, I'm excited to work on this problem as part of the SciML small grants program. Is here an appropriate to share my progress, as well as on the official Julia slack? Thanks!

[ChrisRackauckas]

Please use a WIP PR to share the progress.

[EdwardBerman]

Sounds good, I submitted a WIP PR here: #936 . Thanks!

[jonathanBieler]

Hi @EdwardBerman, tell me if you need help :)

[parteekcoder]

Hi @ChrisRackauckas, can you please confirm is this issue still active in SciML projects, or if anyone is working on this?

[Vaibhavdixit02]

This is open, but the scope has increased, we want to wrap all the classes in https://numbbo.github.io/coco/testsuites the current BlackBoxOptimizationBenchmarking.jl only does the bbob suite from there

[parteekcoder]

is anyone working on this, I would like to start working on it. Can you please tell me the procedure for the application? @Vaibhavdixit02

[ChrisRackauckas]

The instructions are here: https://sciml.ai/small_grants/#declaring_for_a_project

[jonathanBieler]

FYI, I released a new version based around Optimization.jl interface. I ran a small benchmark with some of the algo in Optimization.jl but doing it systematically is quite time consuming (I ran into issues with some algo, like crash, too much printing, weird performance, ...).

https://github.com/jonathanBieler/BlackBoxOptimizationBenchmarking.jl

[ChrisRackauckas]

Holy shit that's amazing thanks!

[jonathanBieler]

Some issues I had :

• Didn't find a way to set the number of objective function evaluations (instead of number iterations), so I'm wrapping my function in a call counter to get that, but this give me strange results with e.g. OptimizationMetaheuristics.CGSA. Not sure if it's an issue with my code or Metaheuristics. In low dimension the difference between the two isn't too big though.
• Some OptimizationEvolutionary's algo terminate very early and I didn't find a way to tweak the convergence parameters. Arguably that's an issue with Evolutionary's defaults.
• I didn't find a way to turn off verbosity for NOMADOpt (it slows things down).
• Some of OptimizationPRIMA method just crashed.

[ChrisRackauckas]

Can we get those into issues?