Codecov Report

All modified and coverable lines are covered by tests ✅

Hey @r3kste!

I discussed the problem / inconsistency with Janos. We realized that instead of adhering to the output of run_explorations, we'd rather change the output type! That is, we believe it is more fitting and in line with the rest of the codebase if run_explorations returns a dictionary of list[float]'s.

Can you rename the PR and make the required changes for this?

Thanks for the quick feedback.

That is, we believe it is more fitting and in line with the rest of the codebase if run_explorations returns a dictionary of list[float]'s.

The "sorted_sample" key in the output dictionary of run_explorations is currently a 2D numpy array. So I don't think we can change the return type to dictionary of list[float].

If we are going ahead with "sorted_values" being a list[float], We could change return type of run_explorations to dict[str, Any] and keep "sorted_sample" as a NDArray.

The exploration result has fixed fields, so it should not be a dict at all but a dataclass (say InternalExplorationResult) so we get type checking for attribute access.

The reason why I would have preferred a list is that this is in principle a history and for histories we usually use lists of floats or lists of arrays as this is the most intuitive way to store histories. However, I don't have a strong opinion here. So you can choose if InternalExplorationResult has list or array attributes but we definitely don't want to use a dict here.