Auto chunk (#4064)

* Added chunks='auto' option in dataset.py

* FIX: correct dask array handling in _calc_idxminmax (#3922)

* FIX: correct dask array handling in _calc_idxminmax

* FIX: remove unneeded import, reformat via black

* fix idxmax, idxmin with dask arrays

* FIX: use array[dim].data in `_calc_idxminmax` as per @keewis suggestion, attach dim name to result

* ADD: add dask tests to `idxmin`/`idxmax` dataarray tests

* FIX: add back fixture line removed by accident

* ADD: complete dask handling in `idxmin`/`idxmax` tests in test_dataarray, xfail dask tests for dtype dateime64 (M)

* ADD: add "support dask handling for idxmin/idxmax" in whats-new.rst

* MIN: reintroduce changes added by #3953

* MIN: change if-clause to use `and` instead of `&` as per review-comment

* MIN: change if-clause to use `and` instead of `&` as per review-comment

* WIP: remove dask handling entirely for debugging purposes

* Test for dask computes

* WIP: re-add dask handling (map_blocks-approach), add `with raise_if_dask_computes()` context to idxmin-tests

* Use dask indexing instead of map_blocks.

* Better chunk choice.

* Return -1 for _nan_argminmax_object if all NaNs along dim

* Revert "Return -1 for _nan_argminmax_object if all NaNs along dim"

This reverts commit 58901b9.

* Raise error for object arrays

* No error for object arrays. Instead expect 1 compute in tests.

Co-authored-by: dcherian <[email protected]>

* fix the failing flake8 CI (#4057)

* rename d and l to dim and length

* Fixed typo in rasterio docs (#4063)

* Added chunks='auto' option in dataset.py

Added changes to whats-new.rst

* Added chunks='auto' option in dataset.py

Added changes to whats-new.rst

* Error fix, catch chunks=None

* Minor reformatting + flake8 changes

* Added isinstance(chunks, (Number, str)) in dataset.py, passing

* format changes

* added auto-chunk test for dataarrays

* Assert chunk sizes equal in auto-chunk test

Co-authored-by: Kai Mühlbauer <[email protected]>
Co-authored-by: dcherian <[email protected]>
Co-authored-by: keewis <[email protected]>
Co-authored-by: clausmichele <[email protected]>
Co-authored-by: Keewis <[email protected]>

doc/whats-new.rst

@@ -36,6 +36,10 @@ Breaking changes
 
 New Features
 ~~~~~~~~~~~~
+
+- ``chunks='auto'`` is now supported in the ``chunks`` argument of
+  :py:meth:`Dataset.chunk`. (:issue:`4055`)
+  By `Andrew Williams <https://github.com/AndrewWilliams3142>`_ 
 - Added :py:func:`xarray.cov` and :py:func:`xarray.corr` (:issue:`3784`, :pull:`3550`, :pull:`4089`).
   By `Andrew Williams <https://github.com/AndrewWilliams3142>`_ and `Robin Beer <https://github.com/r-beer>`_.
 - Added :py:meth:`DataArray.polyfit` and :py:func:`xarray.polyval` for fitting polynomials. (:issue:`3349`)

xarray/core/dataset.py

@@ -1707,7 +1707,10 @@ def chunks(self) -> Mapping[Hashable, Tuple[int, ...]]:
     def chunk(
         self,
         chunks: Union[
-            None, Number, Mapping[Hashable, Union[None, Number, Tuple[Number, ...]]]
+            None,
+            Number,
+            str,
+            Mapping[Hashable, Union[None, Number, str, Tuple[Number, ...]]],
         ] = None,
         name_prefix: str = "xarray-",
         token: str = None,
@@ -1725,7 +1728,7 @@ def chunk(
 
         Parameters
         ----------
-        chunks : int or mapping, optional
+        chunks : int, 'auto' or mapping, optional
             Chunk sizes along each dimension, e.g., ``5`` or
             ``{'x': 5, 'y': 5}``.
         name_prefix : str, optional
@@ -1742,7 +1745,7 @@ def chunk(
         """
         from dask.base import tokenize
 
-        if isinstance(chunks, Number):
+        if isinstance(chunks, (Number, str)):
             chunks = dict.fromkeys(self.dims, chunks)
 
         if chunks is not None:

xarray/tests/test_dask.py

@@ -1035,6 +1035,14 @@ def test_unify_chunks_shallow_copy(obj, transform):
     assert_identical(obj, unified) and obj is not obj.unify_chunks()
 
 
+@pytest.mark.parametrize("obj", [make_da()])
+def test_auto_chunk_da(obj):
+    actual = obj.chunk("auto").data
+    expected = obj.data.rechunk("auto")
+    np.testing.assert_array_equal(actual, expected)
+    assert actual.chunks == expected.chunks
+
+
 def test_map_blocks_error(map_da, map_ds):
     def bad_func(darray):
         return (darray * darray.x + 5 * darray.y)[:1, :1]