Notebook for trust regions #777
Conversation
There was a problem hiding this comment.
I am wondering if it would make sense to have a first section showing TREGO. Or ideally, TREGO, then batch-TR with Thompson sampling, and finally TurBO?
So that this notebook would give an overview of the TR features we have in trieste and would also act as an integration test?
khurram-ghani
force-pushed
the
khurram/multi_tr_notebook
branch
from
97b954c
to
d0b85d4
Compare
Yes, makes sense. I'll add that in the notebook using the existing versions of TREGO and TurBO. We can then migrate to the batch implementations later in the respective PRs. |
khurram-ghani
changed the title
| # | ||
| # Finally, we show how to run Bayesian optimization with the `TurBO` algorithm. This is a | ||
| # trust region algorithm that uses local models and datasets to approximate the objective function | ||
| # within one trust region. |
There was a problem hiding this comment.
not sure what the right wording is, but perhaps "within their respective trust regions"? otherwise it could sound like a single TR algorithm? ignore if you think this is sufficiently clear as is
There was a problem hiding this comment.
I did have slightly different wording before that implied multiple regions, but explicitly changed it to one region. The current TurBO implementation only supports one region. When that is replaced with new classes in a future PR, I can update the wording.
Related issue(s)/PRs: #773
Summary
Added example notebook for three trust region algorithms.
Fully backwards compatible: yes
PR checklist