[Bug]: `gempyor.statistics::llik()` only returns `xr.DataArrays` when RMSE, Abs.Err. output should be a scalar

[emprzy]

Label

gempyor, bug

Priority Label

medium priority

Describe the bug/issue

The gempyor.statistics::llik() function matches a config-given objective function to the appropriate class in gempyor.objective_functions(), calculates the error metric value(s), then casts the output as an xr.DataArray. In the cases of RMSE and Absolute Error, the error metric calculation is a single value, not an array.

llik() wants to return an xr.DataArray with the same shape as gt_data, but RMSE and Absolute Error error metrics generate single number scalars, causing failures down the line. Presently, if you try to configure rmse or absolute_error into a config, a NotImplementedError is thrown, but we need to either choose to not use RMSE/Absolute Error or find a way to have modular return shape with llik() (and other dependent functions).

To Reproduce

I have temporarily disabled rmse and absolute_error as likelihood distribution options to prevent unexpected failures for users, but:
See lines ~235-257 of statistics.py , as well as commented return values for objective_functions::AbsoluteError.error_metric_calculation(...) (~258) and objective_functions::RMSE.error_metric_calculation(...) (~279).

Environment, if relevant

No response

[TimothyWillard]

Possible duplicate of #477?

[emprzy]

Possible duplicate of #477?

Ooo that's a good catch. Perhaps we should keep this one though since it is more up to date with recent objective_functions.py updates?