#30 implements an adapter to use Ax for Bayesian Optimization in CADET-Process. The aim of that PR is a MVP, allowing us to explore further options.

This issue here collects some further ideas to improve BO in CADET-Process.

• Enable specification of Hyperparameters of the optimizer (moved to Enable specification of Hyperparameters for Ax #96)
• Implement linear equality constraints (see also Add method to transform linear equality constraints to variable dependencies #97)
• Implement early stopping and batch trials for Ax (moved to Implement early stopping and batch trials for Ax #98)
• Thompson sampling for trivial parallelization. Maybe parallelize the minimization of the acquisition fct if this is possible in ax; required (Implement early stopping and batch trials for Ax #98)
• Add Noise / Errors:
  • Standard Error of Measurement (SEM): ideally, SEM can be specified also in the Objectives and Nonlinear Constraints.
  • Explore if adding artificial noise to the result in CADETProcessRunner.run helps the fitting. I think it is common in BO

Other ideas (low priority)

• Explore "fidelity" / "multi-fidelity" concepts in Ax
• Staging can formalize constraints checking, etc. before trials are actually run (similar to RepairIndividual in pymoo).

  class CADETProcessRunner(Runner):
      ...
      def staging_required(self) -> bool:
          # return True  # if staging should be a required step
          return False
      ...

  Staging: For new trials, there are different staging steps (e.g. to check rejection, feasibility, ...). How can we leverage this?