Opt out of auto creating index variables #8711

TomNicholas · 2024-02-05T22:04:36Z

Tries fixing #8704 by cherry-picking from #8124 as @benbovy suggested in #8704 (comment)

Closes Currently no way to create a Coordinates object without indexes for 1D variables #8704
Tests added
User visible changes (including notable bug fixes) are documented in whats-new.rst
~~New functions/methods are listed in api.rst~~

TomNicholas · 2024-02-05T22:09:48Z

This currently doesn't work - when I run test_coordinates.py I get some failures:

FAILED xarray/tests/test_coordinates.py::TestCoordinates::test_init_default_index - AssertionError: {'x': <class 'xarray.core.variable.Variable'>}
FAILED xarray/tests/test_coordinates.py::TestCoordinates::test_getitem - AssertionError: {'x': <class 'xarray.core.variable.Variable'>}
FAILED xarray/tests/test_coordinates.py::TestCoordinates::test_assign - AssertionError: {'x': <class 'xarray.core.variable.Variable'>, 'y': <class 'xarray.core.variable.IndexVariable'>}
FAILED xarray/tests/test_coordinates.py::TestCoordinates::test_align - ValueError: cannot reindex or align along dimension 'x' because of conflicting dimension sizes: {2, 3}

The first couple of errors seem to be because the 1D variable doesn't have a corresponding index created by default.

I noticed that the local variable default_indexes is being defined within Coordinates.__init__ as an empty dict, which is never changed before being used to update indexes. This seems like an indication of some missing logic to populate default_indexes?

https://github.com/pydata/xarray/pull/8711/files#diff-c73f63e529ab4065cca1aa23d93c91a3b060f314e2e498b708b90fef87141a88R286

* ``Coordinates.__init__`` create default indexes ... for any input dimension coordinate, if ``indexes=None``. Also, if another ``Coordinates`` object is passed, extract its indexes and raise if ``indexes`` is not None (no align/merge supported here). * add docstring examples * fix doctests * fix tests * update what's new

benbovy · 2024-02-06T12:36:48Z

Sorry the two commits picked from #8124 where done before #8107 and messed up things a bit. I also picked and added the last merge commit from #8124, which should hopefully make things better (coordinate tests are all running fine locally at least).

after unintentionally reverted a valid previous change.

benbovy · 2024-02-06T12:49:10Z

About the auto_convert option added to as_variable() here: this is a temporary workaround before we make a deeper and better refactoring of non-index vs. index coordinate variables. That is not super nice since as_variable() is public API, but maybe that's fine if it is clearly stated in the docstrings that auto_convert is for internal use only (and we likely be removed in the future)?

benbovy · 2024-02-06T13:10:37Z

xarray/core/variable.py

+    if auto_convert:
+        if name is not None and name in obj.dims and obj.ndim == 1:
+            # automatically convert the Variable into an Index
+            emit_user_level_warning(


Maybe we should remove this warning since we're not exactly sure yet what the future behavior will be?

Or at least let's make sure this warning won't be emitted in user code unless explicitly calling as_variable(..., auto_convert=True), i.e., in xarray internals all new variables should be converted into index variables explicitly (if needed).

Okay, so I tried to ensure that this warning won't be emitted in user code simply by replacing the warning with an exception and running all the tests to see where it raised. The only places it raised were:

In core.dataarray._check_data_shape, which is a place where creating an index is unnecessary, so I fixed it in 6bbcc8a,

In the TestVariable.test_as_variable test itself, which seems like it actually should raise the warning?

Is that sufficient?

Yes that may be enough for now.

TomNicholas · 2024-02-06T17:40:08Z

Thanks @benbovy !

That is not super nice since as_variable() is public API, but maybe that's fine if it is clearly stated in the docstrings that auto_convert is for internal use only (and we likely be removed in the future)?

I mean I don't think that's a huge deal, but if you definitely wanted to hide it from users couldn't you make a private _as_variable(..., auto_convert) and use that for internal calls and have the public as_variable call the private one?

dcherian · 2024-02-06T17:43:21Z

Why is as_variable public API?

benbovy · 2024-02-06T17:47:31Z

why is as_variable public API?

More context in #1303

benbovy · 2024-02-07T08:27:56Z

I mean I don't think that's a huge deal

Me neither (I wanted to see what others think).

@TomNicholas have you tried running your notebook mentioned in #8704 (comment) with this PR? Does it fix the errors raised?

TomNicholas · 2024-02-07T14:29:36Z

@TomNicholas have you tried running your notebook mentioned in #8704 (comment) with this PR? Does it fix the errors raised?

Yes I did, and I think it does! I still can't do a full xr.concat on Dataset objects because of a different bug (see #8714), but with this PR gets to the same point as v2023.08.0 does at least. 😀

TomNicholas · 2024-03-19T16:20:19Z

@benbovy I would really like to get this merged as I'm using it in a new package (see zarr-developers/VirtualiZarr#42), but I'm not really sure what the steps to make this PR ready for merging are. Does this need any new tests?

benbovy

@TomNicholas I think that this PR is almost good to go in.

We just need to ensure that unnecessary warnings won't get emitted all over the place (see #8711 (comment)).

Also a test for the skipped creation of index variables would be welcome (see my comment below).

benbovy · 2024-03-20T15:14:44Z

xarray/core/coordinates.py

+                var = as_variable(data, name=name, auto_convert=False)
                if var.dims == (name,) and indexes is None:


This place is (I think) the only place where a newly created "dimension coordinate" might not always be an IndexVariable (i.e., when an empty dictionary or any other value than None is passed as the indexes argument). It would be nice to add a test for this specific case.

The other refactored as_variable places in this PR always create an IndexVariable for a dimension coordinate so I think it is covered by the existing tests (invariant checks).

I've read this comment many times over, and I don't understand why this test isn't already testing the case you're asking about (i.e. passing indexes={} to the Coordinates constructor, hence preventing an IndexVariable being automatically created).

We might want to add a check in that specific test to make sure that the created "x" variable is not coerced to an IndexVariable (before this PR it was still the case even though no xr.indexes.PandasIndex was created from it).

Okay I added a check both explicitly and for if that warning is raised.

dcherian · 2024-03-22T21:16:59Z

Can we catch these warnings if they are expected?


xarray/tests/test_variable.py::TestVariable::test_as_variable
  /home/runner/work/xarray/xarray/xarray/tests/test_variable.py:1219: FutureWarning: variable 'x' with name matching its dimension will not be automatically converted into an `IndexVariable` object in the future.
    actual = as_variable(data, name="x")

xarray/tests/test_variable.py::TestVariable::test_as_variable
  /home/runner/work/xarray/xarray/xarray/tests/test_variable.py:1237: FutureWarning: variable 'time' with name matching its dimension will not be automatically converted into an `IndexVariable` object in the future.
    assert as_variable(dt, "time").dtype.kind == "M"

xarray/tests/test_variable.py::TestVariable::test_as_variable
  /home/runner/work/xarray/xarray/xarray/tests/test_variable.py:1239: FutureWarning: variable 'time' with name matching its dimension will not be automatically converted into an `IndexVariable` object in the future.
    assert as_variable(td, "time").dtype.kind == "m"

TomNicholas · 2024-03-22T22:04:31Z

This should be done now!

Can we catch these warnings if they are expected?

Done

benbovy

LGTM, Thanks @TomNicholas!

* main: (26 commits) [pre-commit.ci] pre-commit autoupdate (pydata#8900) Bump the actions group with 1 update (pydata#8896) New empty whatsnew entry (pydata#8899) Update reference to 'Weighted quantile estimators' (pydata#8898) 2024.03.0: Add whats-new (pydata#8891) Add typing to test_groupby.py (pydata#8890) Avoid in-place multiplication of a large value to an array with small integer dtype (pydata#8867) Check for aligned chunks when writing to existing variables (pydata#8459) Add dt.date to plottable types (pydata#8873) Optimize writes to existing Zarr stores. (pydata#8875) Allow multidimensional variable with same name as dim when constructing dataset via coords (pydata#8886) Don't allow overwriting indexes with region writes (pydata#8877) Migrate datatree.py module into xarray.core. (pydata#8789) warn and return bytes undecoded in case of UnicodeDecodeError in h5netcdf-backend (pydata#8874) groupby: Dispatch quantile to flox. (pydata#8720) Opt out of auto creating index variables (pydata#8711) Update docs on view / copies (pydata#8744) Handle .oindex and .vindex for the PandasMultiIndexingAdapter and PandasIndexingAdapter (pydata#8869) numpy 2.0 copy-keyword and trapz vs trapezoid (pydata#8865) upstream-dev CI: Fix interp and cumtrapz (pydata#8861) ...

benbovy added 2 commits February 5, 2024 16:45

as_variable: deprecate converting to IndexVariable

07b9856

fix multi-index edge case

be99673

TomNicholas mentioned this pull request Feb 5, 2024

Currently no way to create a Coordinates object without indexes for 1D variables #8704

Closed

5 tasks

fix deprecation warning

a3116f7

after unintentionally reverted a valid previous change.

benbovy reviewed Feb 6, 2024

View reviewed changes

Merge branch 'main' into opt-out-auto-create-index-variables

0f70805

TomNicholas mentioned this pull request Mar 18, 2024

Test fsspec roundtrip zarr-developers/VirtualiZarr#42

Merged

dcherian requested a review from benbovy March 19, 2024 17:04

benbovy reviewed Mar 20, 2024

View reviewed changes

TomNicholas added 2 commits March 22, 2024 16:13

avoid unnecessary auto-creation of index to avoid userwarning

6bbcc8a

Merge branch 'main' into opt-out-auto-create-index-variables

6bb7149

TomNicholas requested a review from benbovy March 22, 2024 20:39

TomNicholas added 3 commits March 22, 2024 17:32

catch expected FutureWarnings in test_as_variable

3581a44

check for coercion to IndexVariable

5649a18

whatsnew

66a6fd7

This was referenced Mar 25, 2024

Concatenation automatically creates indexes where none existed #8871

Open

Test concat of dimension coordinate not backed by an index zarr-developers/VirtualiZarr#44

Merged

TomNicholas mentioned this pull request Mar 25, 2024

Required changes in xarray to avoid creating indexes zarr-developers/VirtualiZarr#14

Closed

benbovy approved these changes Mar 26, 2024

View reviewed changes

Merge branch 'main' into opt-out-auto-create-index-variables

55c0b19

benbovy merged commit aaf3b7e into pydata:main Mar 26, 2024
29 checks passed

TomNicholas deleted the opt-out-auto-create-index-variables branch March 26, 2024 13:55

TomNicholas mentioned this pull request Mar 29, 2024

Avoid auto creation of indexes in concat #8872

Merged

4 tasks

CagtayFabry mentioned this pull request Apr 4, 2024

use xarray.Coordinates explicitly BAMWelDX/weldx#923

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Opt out of auto creating index variables #8711

Opt out of auto creating index variables #8711

TomNicholas commented Feb 5, 2024 •

edited

Loading

TomNicholas commented Feb 5, 2024

benbovy commented Feb 6, 2024

benbovy commented Feb 6, 2024

benbovy Feb 6, 2024 •

edited

Loading

TomNicholas Mar 22, 2024

benbovy Mar 22, 2024

TomNicholas commented Feb 6, 2024

dcherian commented Feb 6, 2024

benbovy commented Feb 6, 2024

benbovy commented Feb 7, 2024

TomNicholas commented Feb 7, 2024

TomNicholas commented Mar 19, 2024

benbovy left a comment •

edited

Loading

benbovy Mar 20, 2024

TomNicholas Mar 22, 2024

benbovy Mar 22, 2024

TomNicholas Mar 22, 2024

dcherian commented Mar 22, 2024

TomNicholas commented Mar 22, 2024

benbovy left a comment

		var = as_variable(data, name=name, auto_convert=False)
		if var.dims == (name,) and indexes is None:

Opt out of auto creating index variables #8711

Opt out of auto creating index variables #8711

Conversation

TomNicholas commented Feb 5, 2024 • edited Loading

TomNicholas commented Feb 5, 2024

benbovy commented Feb 6, 2024

benbovy commented Feb 6, 2024

benbovy Feb 6, 2024 • edited Loading

Choose a reason for hiding this comment

TomNicholas Mar 22, 2024

Choose a reason for hiding this comment

benbovy Mar 22, 2024

Choose a reason for hiding this comment

TomNicholas commented Feb 6, 2024

dcherian commented Feb 6, 2024

benbovy commented Feb 6, 2024

benbovy commented Feb 7, 2024

TomNicholas commented Feb 7, 2024

TomNicholas commented Mar 19, 2024

benbovy left a comment • edited Loading

Choose a reason for hiding this comment

benbovy Mar 20, 2024

Choose a reason for hiding this comment

TomNicholas Mar 22, 2024

Choose a reason for hiding this comment

benbovy Mar 22, 2024

Choose a reason for hiding this comment

TomNicholas Mar 22, 2024

Choose a reason for hiding this comment

dcherian commented Mar 22, 2024

TomNicholas commented Mar 22, 2024

benbovy left a comment

Choose a reason for hiding this comment

TomNicholas commented Feb 5, 2024 •

edited

Loading

benbovy Feb 6, 2024 •

edited

Loading

benbovy left a comment •

edited

Loading