KnowledgeGradient: Tracking issue

[lena-kashtelyan]

Merging three sub-issues as these are likely best addressed at-once:

1. [Done] Support inequality constraints for knowledge gradient as long as they are not inter-point #1639
2. [Bug] Multiple fidelity parameters results in "ValueError: Argument to not_none was None" during GPKG #1211
3. [Bug] evaluate_acquisition_function with qKnowledgeGradient get TypeError: evaluate() missing 1 required positional argument: 'bounds' #1719

Please add your issues with KG here instead of as separate issues. It is not often used in Ax currently, so we recommend trying other generation strategies in the meantime if you are blocked.

[sgbaird]

I want to mention that I have a small, standalone example of using GPKG with AxClient: https://github.com/sgbaird/honegumi/blob/main/scripts/refreshers/continuous_multi_fidelity.py [permalink].

[esantorella]

Just confirming that this hasn't been forgotten, but hasn't been a top priority.

[esantorella]

The first and third issues are fixed. As for the second, multi-fidelity optimization is not supported and is likely the root of the issue. So I will consider the KG issues resolved.