search

pytorch / botorch

Bayesian optimization in PyTorch
https://botorch.org/
MIT License
3.09k stars 400 forks source link

[Feature Request] allow both get_ic_generator and batch_initial_conditions to be used in the same call. #2609

Open CompRhys opened 8 hours ago

CompRhys commented 8 hours ago

🚀 Feature Request

When targeting a campaign I might want to include some points explicitly and then fill up the rest of the initial candidates using the strategies in botorch. This behaviour would be triggered by having raw_samples > len(batch_initial_conditions)

Motivation

Makes it much easier to do https://github.com/facebook/Ax/pull/2938 as then all I have to worry about to get the desired behaviour for that to get merged is dropping the points from the given X_pareto front that do not satisfy the constraints

Pitch

this should be fairly minimal change in the logic to check that raw_samples is not None and then just using raw_samples-len(batch_initial_conditions) as the number of requested samples to the ic generator.

Are you willing to open a pull request? (See CONTRIBUTING)

yes

Additional context

saitcakmak commented 2 hours ago

Hi @CompRhys! That sounds reasonable to me. Happy to review a PR for this.