Skip to content

Optimizers

A deck’s Optimize tab needs an optimizer backend installed in that deck’s Python. The app offers three, each at a version it was tested with:

BackendPackageNotes
Axax-platformMeta’s Bayesian optimization platform, on BoTorch
BayBEbaybeMerck’s Bayesian back end; [chem] adds substance parameters (SMILES)
NIMOnimoNIMS orchestration; takes discrete candidate lists

Open the deck’s Settings → Optimizers, tick what you want, and the app installs it with the deck’s other packages in one go: a version conflict is refused before anything changes, and the deck file is updated only once the install succeeded. Taking one out leaves it installed, since decks share one Python environment. Every install uses the CPU build of PyTorch, so adding an optimizer never swaps PyTorch for another build.

A platform installed from the Hub asks the same question on its way in, so a new deck is ready for the Optimize page without a trip to Settings.

An optimizer that is installed but missing from the Optimize tab’s dropdown failed to load; the deck’s Log tab says why at startup ([optimizer] baybe is installed but could not be loaded: …). How the three behave differently is in How the optimizers behave.