Population based optimization

[facebook]

Population based optimization

[eytan]

Hi @riyadparvez, we don't have any immediate plans. We do have a tutorial notebook on using CMA-ES to optimize acquisition functions in BoTorch, but haven't had any use-cases for zeroth-order optimizers within Ax itself. Our goal has been to accelerate experimentation in cases where experiments are costly to run—so our focus has been around algorithms that perform well with less than a few hundred iterations.

What use-case did you have in mind for population-based optimization with Ax?

That being said, it should be fairly straightforward to add a model bridge for Nevergrad.
If you do end up putting something together, we'd be happy to review any PRs and add it to Ax!

[eytan]

Sounds good. Happy to review any PRs!

…

On Tue, Feb 25, 2020 at 11:09 PM Teytaud ***@***.***> wrote: Adding a bridge to Nevergrad is I think a good idea. The cost is moderate. The idea has been around since 2018 (originally I was proposing to do it by myself). It has not been done for bad reasons. It can extend the range of applications of AX. Nevergrad can include parallelism, noisy optimization, high scale optimization, and multiobjective optimization. Even for Nevergrad, given the number of contributors, if we identify problems that we can not correctly handle, that will help us a lot to improve the platform. I see no drawback for doing it. For example, this would have been done earlier CMA would have been added automatically, as well as quite a lot of algorithms. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#163?email_source=notifications&email_token=AAAW34KFYMCAO66ZUFULLDLREYIRHA5CNFSM4IS776S2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEM7CAAA#issuecomment-591273984>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAW34JC64G6HHTHGAP4VBTREYIRHANCNFSM4IS776SQ> .

[lena-kashtelyan]

We will now be tracking wishlist items / feature requests in a master issue for improved visibility: #566. Of course please feel free to still open new feature requests issues; we'll take care of thinking them through and adding them to the master issue.

For this specific issue, there is currently ongoing work that may address it; we will make sure to comment here if/when the prospects become clearer there.