issues: 182667672

This data as json

id node_id number title user state locked assignee milestone comments created_at updated_at closed_at author_association active_lock_reason draft pull_request body reactions performed_via_github_app state_reason repo type
182667672 MDU6SXNzdWUxODI2Njc2NzI= 1046 center=True for xarray.DataArray.rolling() 5572303 open 0     8 2016-10-13T00:37:25Z 2024-04-04T21:06:57Z   CONTRIBUTOR      

The logic behind setting center=True confuses me. Say window size = 3. The default behavior (center=False) sets the window to go from i-2 to i, so I would've expected center=True to set the window from i-1 to i+1. But that's not what I see.

For example, this is what data looks like:

```

data = xr.DataArray(np.arange(27).reshape(3, 3, 3), coords=[('x', ['a', 'b', 'c']), ('y', [-2, 0, 2]), ('z', [0, 1 ,2])])

data xarray.DataArray (x: 3, y: 3, z: 3), array([[[ 0, 1, 2], [ 3, 4, 5], [ 6, 7, 8]],

   [[ 9, 10, 11],
    [12, 13, 14],
    [15, 16, 17]],

   [[18, 19, 20],
    [21, 22, 23],
    [24, 25, 26]]])

Coordinates: * x (x) |S1 'a' 'b' 'c' * y (y) int64 -2 0 2 * z (z) int64 0 1 2 ```

Now, if I set y-window size = 3, center = False, min # of entries = 1, I get

```

r = data.rolling(y=3, center=False, min_periods=1) r.mean() <xarray.DataArray (x: 3, y: 3, z: 3)> array([[[ 0. , 1. , 2. ], [ 1.5, 2.5, 3.5], [ 3. , 4. , 5. ]],

   [[  9. ,  10. ,  11. ],
    [ 10.5,  11.5,  12.5],
    [ 12. ,  13. ,  14. ]],

   [[ 18. ,  19. ,  20. ],
    [ 19.5,  20.5,  21.5],
    [ 21. ,  22. ,  23. ]]])

Coordinates: * x (x) |S1 'a' 'b' 'c' * y (y) int64 -2 0 2 * z (z) int64 0 1 2 ```

Which essentially gives me a "trailing window" of size 3, meaning the window goes from i-2 to i. This is not explained in the doc but can be understood empirically.

On the other hand, setting center = True gives

```

r = data.rolling(y=3, center=True, min_periods=1) r.mean() <xarray.DataArray (x: 3, y: 3, z: 3)> array([[[ 1.5, 2.5, 3.5], [ 3. , 4. , 5. ], [ nan, nan, nan]],

   [[ 10.5,  11.5,  12.5],
    [ 12. ,  13. ,  14. ],
    [  nan,   nan,   nan]],

   [[ 19.5,  20.5,  21.5],
    [ 21. ,  22. ,  23. ],
    [  nan,   nan,   nan]]])

Coordinates: * x (x) |S1 'a' 'b' 'c' * y (y) int64 -2 0 2 * z (z) int64 0 1 2 ```

In other words, it just pushes every cell up the y-dim by 1, using nan to represent things coming off the edge of the universe. If you look at _center_result() of xarray/core/rolling.py, that's exactly what it does with .shift().

I would've expected center=True to change the window to go from i-1 to i+1. In which case, with min_periods=1, would not render any nan value in r.mean().

Could someone explain the logical flow to me?

Much obliged,

Chun

 
{
    "url": "https://api.github.com/repos/pydata/xarray/issues/1046/reactions",
    "total_count": 0,
    "+1": 0,
    "-1": 0,
    "laugh": 0,
    "hooray": 0,
    "confused": 0,
    "heart": 0,
    "rocket": 0,
    "eyes": 0
}
     13221727 issue

Links from other tables