botorch is loaded when import baybe

[rjavadi]

Since campaign uses recommenders and they depend on botorch, we still relay on torch. If we lazy load recommenders in campaign like this:

if TYPE_CHECKING:
    from baybe.recommenders.base import RecommenderProtocol
    from baybe.recommenders.meta.sequential import TwoPhaseMetaRecommender

we can compeletly get rid of torch, but the tests fail with this error:

baybe/campaign.py:57: in Campaign
    recommender: RecommenderProtocol = field(factory=TwoPhaseMetaRecommender)

Now we should look for a workaround to use TwoPhaseMetaRecommender in campaign.

Here's the import tree of importing baybe:

And here's the import tree of baybe after lazy loading recommenders in campaign:

[rjavadi]

Update: so far, I couldn't find a way to decouple or loosely couple Recommener and campaign. As long as we import recommender in baybe, we load torch as well.

[AdrianSosic]

Hi @rjavadi, thanks for the update. I also had a brief look and know how to get rid of the remaining imports. However, it requires one additional change to happen first, which you should not have to bother with. Let me take care of the remaining open changes here so that you can fully focus on the polars part. Does that sound good to you?

[rjavadi]

Hi Adrian, thanks for the heads up.
Yes, sounds good.

[Scienfitz]

closing as no longer relevant