I have two group by operations.

One uses pandas:

dg_cmp = pd.DataFrame(data).groupby(header)

where data is a numpy.ndarray (120120,1500) and header is a pandas.core.series.Series:

<class 'pandas.core.series.Series'>
1            1
2            2
3            3
4            4
5            5
          ... 
120116    2138
120117    2139
120118    2140
120119    2141
120120    2142
Name: CDP, Length: 120120

The groups returned look like:

{1.0: array([1]), 2.0: array([2]), 3.0: array([  3, 121]), 4.0: array([  4, 122]), 

The other uses xarray.DataArrays:

dg_cmp = data.groupby(cdp_dataarray)

where data is

<xarray.DataArray 'data' (cdp: 120120, twt: 1500)> Size: 721MB
dask.array<open_dataset-data, shape=(120120, 1500), dtype=float32, chunksize=(51860, 647), chunktype=numpy.ndarray>
Coordinates:
  * cdp      (cdp) int64 961kB 0 1 2 3 4 ... 120115 120116 120117 120118 120119
  * twt      (twt) float64 12kB 0.0 4.0 8.0 ... 5.988e+03 5.992e+03 5.996e+03

and cdp_dataarray is:

<xarray.DataArray 'CDP' (trace: 120120)> Size: 240kB
array([   1,    2,    3, ..., 2140, 2141, 2142], dtype=uint16)
Coordinates:
  * trace    (trace) int64 961kB 0 1 2 3 4 ... 120116 120117 120118 120119

The DataArrayGroupBy groups returned are:

{1: [0], 2: [1], 3: [2, 120], 4: [3, 121], 5: [4, 122, 240], 

Which is not the same as the pandas groupby. Can anyone spot why this is and suggest how to get xarray groupby to match the pandas?