groupby_bins groups not correctly applied with built-in methods

[michaelaye]

What happened?

Setup

I want to calculate image statistics per chunk in one dimension. Let's assume a very small image for demonstration purposes:

a = xr.DataArray(np.arange(12).reshape(6,2), dims=('x', 'y'))
a

Trying to chunk this into three subimages, I use these bins into the x dimension:

x_bins = (0, 2, 4, 6)

I look at the groups this creates by default:

for iv, g in a.groupby_bins('x', x_bins):
    print(iv)
    print(g)

I don't understand the use-case for this grouping, as it's missing the beginning and is having uneven sized last group (Obviously a follow-error from not including the first row).

To force the even chunking of the image I need to call it with these parameters:

groups = a.groupby_bins('x', x_bins, include_lowest=True, right=False)
for iv, g in groups:
    print(iv)
    print(g)

Issue

But now, calculating the mean value of each group, I get different results when doing it by hand using the groups or doing it using the groups inherent method mean():

Indeed, I verified, that these results are what one gets, using the first version of applying the bins:

The same is true when I use the elliptical operator to receive the mean over the remaining dimensions (note, the 2nd cell here is using the groups variable as defined in the cell before, so should really return the same values, but it doesn't:

Application

I believe that groupby_bins is the most appropriate tool to do this in xarray. I wished that one could enforce the dask-chunks in dask arrays to survive and return stats from them, but haven't found a way to do that.

What did you expect to happen?

That the inherent stats methods of the groups method respect the interval use constraints from the groupby_bins call.

I also have verified that the same problem exists with groups.std().

Minimal Complete Verifiable Example

import xarray as xr
import numpy as np

a = xr.DataArray(np.arange(12).reshape(6,2), dims=('x', 'y'))

x_bins = (0, 2, 4, 6)

default_groups = a.groupby_bins('x', x_bins)
my_groups = a.groupby_bins('x', x_bins, include_lowest=True, right=False)

print("Weird grouping using default call:")
for iv, g in default_groups:
    print("Interval:",iv)
    print(g.data)
    print()
    
print("Evenly chunked using `my_groups`:")    
for iv, g in my_groups:
    print("Interval:", iv)
    print(g.data)
    print()

print("Calculating mean on my own using loop over groups:")
for iv, g in my_groups:
    print(g.mean('x').data)
    
print("Calculting same using my_groups.mean()")
print("No dim given:")
print(my_groups.mean().data.T)
print("using mean('x'):")
print(my_groups.mean('x').data.T)

print("These results come from the default groups!:")
for iv, g in default_groups:
    print(g.mean('x').data)
    
print("STD has the same issue")

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.

Relevant log output

No response

Anything else we need to know?

No response

Environment

INSTALLED VERSIONS

commit: None
python: 3.10.9 | packaged by conda-forge | (main, Feb 2 2023, 20:20:04) [GCC 11.3.0]
python-bits: 64
OS: Linux
OS-release: 6.0.12-76060006-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: en_US.UTF-8
LANG: en_US.UTF-8
LOCALE: ('en_US', 'UTF-8')
libhdf5: 1.12.2
libnetcdf: 4.9.1

xarray: 2023.2.0
pandas: 1.5.3
numpy: 1.23.5
scipy: 1.10.1
netCDF4: 1.6.3
pydap: None
h5netcdf: None
h5py: 3.8.0
Nio: None
zarr: None
cftime: 1.6.2
nc_time_axis: None
PseudoNetCDF: None
rasterio: 1.3.6
cfgrib: None
iris: None
bottleneck: 1.3.7
dask: 2023.3.0
distributed: 2023.3.0
matplotlib: 3.7.1
cartopy: 0.21.1
seaborn: 0.12.2
numbagg: None
fsspec: 2023.3.0
cupy: None
pint: None
sparse: None
flox: 0.6.8
numpy_groupies: 0.9.20
setuptools: 67.5.1
pip: 23.0.1
conda: installed
pytest: 7.1.3
mypy: None
IPython: 8.7.0
sphinx: None

[dcherian]

I think this is a bug with not passing cut_kwargs to flox.

Use xr.set_options(use_flox=False) for now. I think this is fixed in #7206

[dcherian]

I wished that one could enforce the dask-chunks in dask arrays to survive and return stats from them, but haven't found a way to do that

What do you mean?

[michaelaye]

That, while I can open an xarray as a dask array using chunks:

I cannot make "use" of these chunks to get statistics per chunk, right? It's just an efficiency question for the dask.compute() stage, but not an actual way to get statistics per chunk?