baybe.simulation¶
Functionality to “simulate” Bayesian DOE given a lookup mechanism.
The term “simulation” can have two slightly different interpretations, depending on the applied context:
It can refer to “backtesting” a particular DOE strategy on a fixed (finite) dataset. In this context, “simulation” means investigating what experimental trajectory we would have observed had we applied the recommender in a certain defined context and restricted the possible parameter configurations to those contained in the dataset.
It can refer to the simulation of an actual DOE loop (i.e. recommending experiments and retrieving the corresponding measurements) where the loop closure is realized in the form of a callable (black-box) function that can be queried during the optimization to provide target values.
Modules
Core simulation and backtesting functionality. |
|
Target lookup mechanisms. |
|
Batch simulation of multiple campaigns. |
|
Functionality for transfer learning backtesting. |