Re-indexing causes coordinates to be dropped

[Hoeze]

Hi, I encounter a problem with the index being dropped when I rename a dimension and stack it afterwards:

MCVE Code Sample

ds = xr.Dataset({
    "test": xr.DataArray(
        [[[1,2],[3,4]], [[1,2],[3,4]]], 
        dims=("genes", "observations", "subtissues"), 
        coords={
            "observations": xr.DataArray(["x-1", "y-1"], dims=("observations",)),
            "individuals": xr.DataArray(["x", "y"], dims=("observations",)), 
            "genes": xr.DataArray(["a", "b"], dims=("genes",)), 
            "subtissues": xr.DataArray(["c", "d"], dims=("subtissues",)),
        }
    )
})

individuals is set here:

print(ds.rename_dims(observations="individuals"))

<xarray.Dataset>
Dimensions:       (genes: 2, individuals: 2, subtissues: 2)
Coordinates:
    observations  (individuals) <U3 'x-1' 'y-1'
  * individuals   (individuals) <U1 'x' 'y'
  * genes         (genes) <U1 'a' 'b'
  * subtissues    (subtissues) <U1 'c' 'd'
Data variables:
    test          (genes, individuals, subtissues) int64 1 2 3 4 1 2 3 4

Stacking caused individuals to disappear and be replaced with integers:

print(ds.rename_dims(observations="individuals").stack(observations=["individuals", "genes"]))

<xarray.Dataset>
Dimensions:       (observations: 4, subtissues: 2)
Coordinates:
  * observations  (observations) MultiIndex
  - individuals   (observations) int64 0 0 1 1
  - genes         (observations) object 'a' 'b' 'a' 'b'
  * subtissues    (subtissues) <U1 'c' 'd'
Data variables:
    test          (subtissues, observations) int64 1 1 3 3 2 2 4 4

Explicitly setting individuals keeps them correctly after stacking:

print(ds.rename_dims(observations="individuals").set_index({"individuals": "individuals"}).set_coords("individuals").stack(observations=["individuals", "genes"]))

<xarray.Dataset>
Dimensions:       (observations: 4, subtissues: 2)
Coordinates:
  * observations  (observations) MultiIndex
  - individuals   (observations) object 'x' 'x' 'y' 'y'
  - genes         (observations) object 'a' 'b' 'a' 'b'
  * subtissues    (subtissues) <U1 'c' 'd'
Data variables:
    test          (subtissues, observations) int64 1 1 3 3 2 2 4 4

Is this by intention?

Output of xr.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.7.4 (default, Aug 13 2019, 20:35:49)
[GCC 7.3.0]
python-bits: 64
OS: Linux
OS-release: 3.10.0-957.10.1.el7.x86_64
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8
libhdf5: 1.10.4
libnetcdf: 4.6.1

xarray: 0.14.0
pandas: 0.25.1
numpy: 1.17.2
scipy: 1.3.1
netCDF4: 1.4.2
pydap: None
h5netcdf: 0.7.4
h5py: 2.9.0
Nio: None
zarr: 2.3.2
cftime: 1.0.3.4
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: 2.5.2
distributed: 2.5.2
matplotlib: 3.1.1
cartopy: None
seaborn: 0.9.0
numbagg: None
setuptools: 41.4.0
pip: 19.2.3
conda: None
pytest: 5.0.1
IPython: 7.8.0
sphinx: None

[keewis]

I don't know whether that is intended or not, but I think that the operation of replacing observations with individuals is better done with swap_dims:

>>> xr.__version__
'0.14.0'
>>> ds.rename_dims(observations="individuals").stack(observations=["individuals", "genes"])
<xarray.Dataset>
Dimensions:       (observations: 4, subtissues: 2)
Coordinates:
  * observations  (observations) MultiIndex
  - individuals   (observations) int64 0 0 1 1
  - genes         (observations) object 'a' 'b' 'a' 'b'
  * subtissues    (subtissues) <U1 'c' 'd'
Data variables:
    test          (subtissues, observations) int64 1 1 3 3 2 2 4 4
>>> ds.swap_dims({"observations": "individuals"}).stack(observations=["individuals", "genes"])
<xarray.Dataset>
Dimensions:       (observations: 4, subtissues: 2)
Coordinates:
  * observations  (observations) MultiIndex
  - individuals   (observations) object 'x' 'x' 'y' 'y'
  - genes         (observations) object 'a' 'b' 'a' 'b'
  * subtissues    (subtissues) <U1 'c' 'd'
Data variables:
    test          (subtissues, observations) int64 1 1 3 3 2 2 4 4

Indeed it seems that on 0.13.0 using rename_dims worked, so that would be a regression. However, I would argue that rename_dims should raise an error if a new dimension already exists (and maybe point to swap_dims).

[shoyer]

However, I would argue that rename_dims should raise an error if a new dimension already exists (and maybe point to swap_dims).

+1 this sounds like a good idea