Rolling Std along axis

[babameme]

Hi team,
I have a DataArray like this:

X = xr.DataArray(np.arange(48).reshape(12, 4), coords = {'X': np.arange(12), 'Y': np.arange(4)}, dims = ['X', 'Y'])

And i want to calculate rolling Standard Deviation along 'X' axis.
If i use:

X.rolling(X = 3).std()

Along each 'Y' axis, i will receive one result, but that is not what i want.
The result i expect is a DataArray with dims = ['X'], mean that each rolling phase, i want to have all 3x4 data points to calculate its Standard Deviation

Regards

[max-sixty]

Thanks for writing out the example @babameme .

There may be a creative way to do this, but there isn't a multi-d rolling aggregation in xarray. The best I've come up with is this workaround: (Edit: this is wrong!! See @dcherian 's answer below)

In [21]: X.stack(z=['X','Y']).rolling(z=3*4).std()[::4]
Out[21]:
<xarray.DataArray (z: 12)>
array([       nan,        nan,        nan, 3.45205253, 3.45205253,
       3.45205253, 3.45205253, 3.45205253, 3.45205253, 3.45205253,
       3.45205253, 3.45205253])
Coordinates:
  * z        (z) MultiIndex
  - X        (z) int64 0 1 2 3 4 5 6 7 8 9 10 11
  - Y        (z) int64 0 0 0 0 0 0 0 0 0 0 0 0

[dcherian]

But we do support nD rolling!

import xarray as xr
import numpy as np

ds = xr.DataArray(
    np.arange(48).reshape(12, 4),
    coords={"X": np.arange(12), "Y": np.arange(4)},
    dims=["X", "Y"],
)

ds.rolling(X=3, Y=4).std()

Here's the result as a ndarray

array([[       nan,        nan,        nan,        nan],
       [       nan,        nan,        nan,        nan],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253],
       [       nan,        nan,        nan, 3.45205253]])

[max-sixty]

Bah amazing! Sorry. I even reviewed this! I will reflect on whether I'm answering these too quickly given my apparent worsening memory...

[dcherian]

or our API surface is too big ;)