🐛 Bug Report
'rolling_window' on a cube will fail it it has lazy aux-coords that parallel the "rolled" dimension.
Because this line and this one apply a type of indexing that doesn't work with Dask
-- but, heavy hint it probably works to just replace (0, -1) with [0, -1]
(see 'Additional Context' below)
How To Reproduce
dimco = DimCoord(np.arange(12), standard_name="time", units="s")
auxco = AuxCoord(da.arange(12), long_name="x")
cube = Cube(
np.arange(12),
dim_coords_and_dims=[(dimco, 0)],
aux_coords_and_dims=[(auxco, 0)],
)
attempt = cube.rolling_window(coord="time", window=3, aggregator=MEAN)
Result is, ultimately...
Traceback (most recent call last):
. . .
File "/home/users/patrick.peglar/git/iris/lib/iris/cube.py", line 4965, in rolling_window
new_bounds = new_bounds[:, (0, -1)]
~~~~~~~~~~^^^^^^^^^^^^
File "/data/apps/sss/environments/preproduction-os47/lib/python3.12/site-packages/dask/array/core.py", line 2045, in __getitem__
index2 = normalize_index(index, self.shape)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/data/apps/sss/environments/preproduction-os47/lib/python3.12/site-packages/dask/array/slicing.py", line 848, in normalize_index
check_index(axis, i, d)
File "/data/apps/sss/environments/preproduction-os47/lib/python3.12/site-packages/dask/array/slicing.py", line 919, in check_index
elif ind >= dimension or ind < -dimension:
^^^^^^^^^^^^^^^^
TypeError: '>=' not supported between instances of 'tuple' and 'int'
Expected behaviour
This operation should be succeed
Additional context
Click to expand this section...
Dask docs here
say "lists/arrays" , but does not include "tuples"
However, given careful reading of Numpy docs here,
it would seem that when applied to a given multidimensional index among others, a tuple or list or 1-D array of integers, should all be valid + produce the same result,
e.g. arr[:, (0, -1)], arr[:, [0, -1]] or arr[:, np.array([0, -1])] -- should all be the same
🐛 Bug Report
'rolling_window' on a cube will fail it it has lazy aux-coords that parallel the "rolled" dimension.
Because this line and this one apply a type of indexing that doesn't work with Dask
-- but, heavy hint it probably works to just replace
(0, -1)with[0, -1](see 'Additional Context' below)
How To Reproduce
Result is, ultimately...
Expected behaviour
This operation should be succeed
Additional context
Click to expand this section...
Dask docs here
say "lists/arrays" , but does not include "tuples"
However, given careful reading of Numpy docs here,
it would seem that when applied to a given multidimensional index among others, a tuple or list or 1-D array of integers, should all be valid + produce the same result,
e.g.
arr[:, (0, -1)],arr[:, [0, -1]]orarr[:, np.array([0, -1])]-- should all be the same