Multi-index with categorical values

[mancellin]

Building a dataset from pandas with a multi-index with categorical values:

import pandas as pd

cat = pd.CategoricalDtype(categories=['foo', 'bar', 'baz'])
i1 = pd.Series(['foo', 'bar'], dtype=cat)
i2 = pd.Series(['bar', 'bar'], dtype=cat)

df = pd.DataFrame({'i1': i1, 'i2': i2, 'values': [1, 2]})
ds = df.set_index(['i1', 'i2']).to_xarray()

print(ds)

Expected output:

<xarray.Dataset>
Dimensions:  (i1: 2, i2: 1)
Coordinates:
  * i1       (i1) object 'foo' 'bar'
  * i2       (i2) object 'bar'
Data variables:
    values   (i1, i2) int64 1 2

Actual output:

<xarray.Dataset>
Dimensions:  (i1: 3, i2: 3)
Coordinates:
  * i1       (i1) object 'foo' 'bar' 'baz'
  * i2       (i2) object 'foo' 'bar' 'baz'
Data variables:
    values   (i1, i2) float64 nan 1.0 nan nan 2.0 nan nan nan nan

It is not wrong, but it is inconsistent with the non-categorical case (which gives the expected output above) and the single-index case (no filling with NaNs for single index).

Output of xr.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 3.8.0 (default, Nov 6 2019, 21:49:08) [GCC 7.3.0] python-bits: 64 OS: Linux OS-release: 4.19.91-1-MANJARO machine: x86_64 processor: byteorder: little LC_ALL: None LANG: fr_FR.UTF-8 LOCALE: fr_FR.UTF-8 libhdf5: None libnetcdf: None

xarray: 0.14.1
pandas: 0.25.3
numpy: 1.17.4
scipy: None
netCDF4: None
pydap: None
h5netcdf: None
h5py: None
Nio: None
zarr: None
cftime: None
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: None
distributed: None
matplotlib: None
cartopy: None
seaborn: None
numbagg: None
setuptools: 44.0.0.post20200106
pip: 19.3.1
conda: None
pytest: None
IPython: None
sphinx: None

[fujiisoup]

Thanks for reporting again.
OK. It looks there are several places to be fixed.

Please add comments here if you find another not-working case.

[fujiisoup]

xref: #3670

[mancellin]

Thank you for your work on this! I haven't found any other issue at the moment, I guess we can close this issue.

[mancellin]

Actually, after updating to version 0.15 I've found another issue in the same context.
More precisely, when using a multi-index with a non-categorical repeated coordinate and a categorical coordinate:

import pandas as pd

i1 = pd.Series([0, 0])
cat = pd.CategoricalDtype(categories=['foo', 'bar', 'baz'])
i2 = pd.Series(['foo', 'bar'], dtype=cat)

df = pd.DataFrame({'i1': i1, 'i2': i2, 'values': [1, 2]})
ds = df.set_index(['i1', 'i2']).to_xarray()

print(ds)

raises the following error

Traceback (most recent call last):
  File "/home/matthieu/test.py", line 8, in <module>
    ds = df.set_index(['i1', 'i2']).to_xarray()
  File "/opt/anaconda3/lib/python3.7/site-packages/pandas/core/generic.py", line 2867, in to_xarray
    return xarray.Dataset.from_dataframe(self)
  File "/opt/anaconda3/lib/python3.7/site-packages/xarray/core/dataset.py", line 4555, in from_dataframe
    idx = remove_unused_levels_categories(dataframe.index)
  File "/opt/anaconda3/lib/python3.7/site-packages/xarray/core/indexes.py", line 26, in remove_unused_levels_
categories
    index = pd.MultiIndex.from_arrays(levels, names=index.names)
  File "/opt/anaconda3/lib/python3.7/site-packages/pandas/core/indexes/multi.py", line 425, in from_arrays
    raise ValueError("all arrays must be same length")
ValueError: all arrays must be same length

but works fine when i2 is not categorical.

[mancellin]

@fujiisoup Since you implemented remove_unused_levels_categories, do you have any clue how to fix this?

[fujiisoup]

@mancellin
Sorry for my no response.
Yes, there may be some possible workarounds, but nowadays I have less spare time...
Do you have the interest to send a PR?

[mancellin]

Yes, I'll give it a try.