Param mapping

[hellkite500]

This should allow a calibration parameter to set an alias key, and any params across any models sharing the same alias name will be treated as the same parameter for the purpose of permutation generation.

Note, needs #180 to merge first.

[aaraney]

This should allow a calibration parameter to set an alias key, and any params across any models sharing the same alias name will be treated as the same parameter for the purpose of permutation generation.

Just to clarify (haven't looked at the code yet), the default behavior is if two modules have the same calibration parameter name (no aliases involved) there is effectively a single parameter in the parameter space that is calibrated, so on the next iteration, both modules will get the same value for that parameter?

Related, but kind of aside, what happens (in code at HEAD) if module param combination A.a and B.a are both defined to have different bounds and default values which, if any, is given precedence?

[aaraney]

In looking at the tests, what are your thoughts on validating that:

• init, min, and max are not nan (on the fence about this one)
• init is in the bounds of min and max
• min >= max?

These seem like sane invariants that we should uphold.

[aaraney]

Aside, we should add a warning here if a Parameter mapping has been configured for a module that is not present.

ngen-cal/python/ngen_cal/src/ngen/cal/ngen.py

Line 64 in b7482de

dfs.append(_params_as_df(params, m.params.model_name))

[aaraney]

At first glance, this overlaps with multi_model_shared_params s.t. I dont think we need it? Can we just get rid of this fixture?

[aaraney]

Minor improvement that without could result in a misleading error.

Suggested change

	@root_validator
	@root_validator(skip_on_failure=True)