SMAC3 Bayesian Optimisation Integration [🆕 Search Method] 🥉

[simonprovost]

Hi @PGijsbers,

As proposed in #202, here is finally my final contribution probably for this year, the implementation of a brand-new Bayesian Optimisation search method based on SMAC3 compatible with GAMA under ConfigSpace search space setup as currently proposed in #210 🎉 🥇

While the PR follows the feature proposal in #202, it also includes an update to search algorithms to reduce pipeline duplication as per #211 🥈

Given that this PR is a suite from #210 and #211, here's a link that focuses exclusively on the most recent commits for this update: Link to PR updates 🥉

Thank you, and I look forward to hearing from you!

What contributions have been made

• ✅ Improving GAMA's Search Methods by adding a new one based on Bayesian Optimisation via SMAC3
• ✅ Testing passes smoothly
• ✅ Informative commits

❌ DISCLAIMER ❌

Kindly refrain from merging this PR prior to ConfigSpace and Search Methods Improvements. I will need to perform a rebase from #211 before this PR can be merged!

[simonprovost]

@PGijsbers A #Waiting tag would be greatly appreciated here.

[PGijsbers]

Thank you! I will put this PR in the queue as you suggest :)

[simonprovost]

TODO:

1. Add random_state to SMAC's seed. @simonprovost Refer to your hidden PR.
2. Add output_directory to SMAC's output_directory. @simonprovost Refer to your hidden PR.

[simonprovost]

Update 15 jan. 2024 – NOT urgent at all.

@PGijsbers – Given that the BO's implementation is utilising the asynchronous evaluator, I reckon we should call it async_BO instead of bayesian_optimisation, the same as async_ea. Your thoughts ?

[simonprovost]

BRIEF UDPATE:

I intend to start working back on all this from the end of this month onwards. I have an important deadline for the 20th and had no time to look into those but will definitely afterward.

Cheers for your patience!

Simon