experimental-design / bofire

Experimental design and (multi-objective) bayesian optimization.
https://experimental-design.github.io/bofire/
BSD 3-Clause "New" or "Revised" License
222 stars 23 forks source link

add output_scaler to mlp #305

Closed simonsung06 closed 11 months ago

simonsung06 commented 1 year ago

Added an output_scaler option for the MLP Ensemble surrogate. This is to give extra flexibility where standardizing data often improves fitting. I have set ScalerEnum.STANDARDIZE as the default output_scaler because this is common practice for regression problems. Currently only standardization and identity scalers are usable. Normalization is not implemented since there is no output transform for this natively on BoTorch. If needed one day, maybe we could implement it (but i believe standardization should be sufficient for us).

jduerholt commented 11 months ago

@simonsung06: I think the tests are failing due to the rdkit issue that you already fixed. I just merged it into main. Just merge it into your PR and the tests should work again!

simonsung06 commented 11 months ago

Thanks for the suggestion. The changes have been made.

Regarding adding it for GP's too should be an easy addition. Would you like me to do this in another PR?

jduerholt commented 11 months ago

Regarding adding it for GP's too should be an easy addition. Would you like me to do this in another PR?

No need to hurry, if you find time, it would be nice to have it in a seperate PR.