create test_weights.py

[raybellwaves]

Taken from #221 (comment) (thanks @dougiesquire)

This needs to be addressed before #221.

In our testing suite we have a function https://github.com/xarray-contrib/xskillscore/blob/master/xskillscore/tests/test_deterministic.py#L72

Is there a way this could be moved upstream to conftest.py?

While verbose we could create fixtures for these (5) e.g.

adjust_weights_time():
weights.isel(lon=0, lat=0)

adjust_weights_lat():
weights.isel(lon=0, time=0)

https://matthewrocklin.com/blog/work/2019/06/23/avoid-indirection

Also i'm not sure if we are testing the weight implementation anywhere. As we found out in #248 test_determinstic.py mostly tests the ufunc part. There are some separate test files for functionality e.g. test_skipna_functionality. We should create one for weights.

[aaronspring]

I like the weights function as it easily integrates in loops over dims or pytest parametrize.

My PR tested weights for some commits https://github.com/xarray-contrib/xskillscore/pull/231/files

[aaronspring]

I don’t understand Matthews post that we should more verbose functions here. I rather like the idea that weights is abstracted away here as the weights depend on dims. I see much code duplication in the new suggestion which will be harder to maintain.

+1 for a weights testing file refactoring.

[raybellwaves]

I know it's a simple function. But part of me is thinking shouldn't that function be tested so I know it's doing what I expect. It appears in a couple of test files https://github.com/xarray-contrib/xskillscore/blob/master/xskillscore/tests/test_deterministic.py#L72 https://github.com/xarray-contrib/xskillscore/blob/master/xskillscore/tests/test_accessor_deterministic.py#L62 hence moving it upstream would help duplication (regardless of verbose fixture vs. function solution)

Thanks for the info in your commit. Lots of useful stuff there e.g. https://xarray.pydata.org/en/stable/generated/xarray.DataArray.weighted.html. I believe this is the file which has the weights testing https://raw.githubusercontent.com/aaronspring/xskillscore/b61185bab28f5a6d4f2aad8211e9e1881a4bb03c/asv_bench/benchmarks/xr_vs_xs.py

Some (haven't checked all) scikit-learn metrics have a weight parameter
https://scikit-learn.org/stable/modules/generated/sklearn.metrics.mean_squared_error.html#sklearn.metrics.mean_squared_error

We could create a test_weighted_metric_results_accurate.py akin to test_weighted_metric_results_accurate.py

[aaronspring]

Can be closed?

[raybellwaves]

I need to finish the other (non scikit-learn metrics)

[raybellwaves]

I need to add scipy stats then can close