issue #5791, dims & cords inference from xarray

[michaelraczycki]

added dim inference from xarray, deprecation warning and unittest for the new feature

What is this PR about?
Addresses changes requested in Issue #5791

• Inferring dims from named pandas indexes
• Inferring dims from DataArrays
• Inferring coords from DataArrays
• A unit test
• Backwards compatible with deprecation warning rename of the export_index_as_coords to infer_dims_and_coords.

New features

• It's now possible to infer dims & coords from xarray variables passed to pm.Data

Documentation

• inline documentation added to data.py

[michaelraczycki]

Do you have any tips on how to make the build pass? It seems really not descriptive for someone who runs it first time, and in the documentation I can't find anything to pin the reason for it failing. I've run the pre-commit checks and it seems fine:

[codecov]

Codecov Report

Merging #6514 (4038a7d) into main (28bac77) will decrease coverage by 12.20%.
The diff coverage is 76.47%.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #6514       +/-   ##
===========================================
- Coverage   94.74%   82.54%   -12.20%     
===========================================
  Files         146      147        +1     
  Lines       27807    27951      +144     
===========================================
- Hits        26346    23073     -3273     
- Misses       1461     4878     +3417     

Impacted Files	Coverage Δ
pymc/data.py	90.17% <76.47%> (-1.60%)	⬇️
pymc/tests/gp/test_gp.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/gp/test_mean.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/gp/test_util.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/smc/test_smc.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/step_methods/test_slicer.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/step_methods/hmc/test_nuts.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/step_methods/test_compound.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/step_methods/test_metropolis.py	0.00% <0.00%> (-100.00%)	⬇️
pymc/tests/gp/test_cov.py	0.00% <0.00%> (-99.82%)	⬇️
... and 31 more

[twiecki]

        split_key = key.split(".", 1)
        if len(split_key) != 2:
>           raise KeyError(key)
E           KeyError: 'dim'

../../../miniconda3/envs/pymc-test/lib/python3.9/site-packages/xarray/core/dataset.py:185: KeyError

[michaelraczycki]

oh, good catch. It seems like I put the quotes around dim in line 241 by mistake

[michaelraczycki]

@twiecki , it seems like my expectations on how the coords values should look like aren't good, could you check if I'm going in the right direction? I'm not sure if my initial assumption is correct and there's a bug I need to find, or the assumption is wrong and I should change it

[twiecki]

CC @michaelosthege

[michaelosthege]

Best run mypy locally by

pip install types-cachetools "typing-extensions>=3.7.4" "mypy==0.990"
python scripts/run_mypy.py --verbose

[michaelraczycki]

Can I change the type hinting within the Data.coords? It expects to have the keys with type Sequence, but unfortunately ndarray from numpy acts 1:1 like a sequence, but it doesn't inherit from Sequence, therefore it's not of a Sequence type.
If you really want to keep the sequence type check, I suggest using coords: Optional[Dict[str, Union(Sequence, np.ndarray)]], that and adjusting the type hinting for return value for determine_coords locally fixed mypy complaining

[twiecki]

@michaelosthege Can we merge this?

[twiecki]

Thanks @michaelraczycki