Pandas Error Repro

README.md

environment.yml

channels:
- conda-forge
dependencies:
- pandas=1.2.1
- dask=2021.01.1
- pytest

test_groupby_errors.py

import pandas as pd
import dask.dataframe as ddf

test_frame = pandas.DataFrame.from_dict(
    {
        "a": ["a", "b"],
         "b": ["foo", "foo"]
    }
).astype(
    dtype={
        "a": pandas.CategoricalDtype(["a", "b", "c"]),
        "b": pandas.CategoricalDtype(["foo", "bar", "bat", "bazz"])
    }
)

d_test_frame = ddf.from_pandas(test_frame, npartitions=1)

def test_pandas_observed():
    test_frame.groupby("a", observed=True)["b"].value_counts()
  
def test_pandas_default():
    # Errors out, potentially related to 
    # https://github.com/pandas-dev/pandas/issues/36698
    test_frame.groupby("a", observed=False)["b"].value_counts()

def test_dask_observed():
    d_test_frame.groupby("a", observed=True)["b"].value_counts()

def test_dask_observed():
    d_test_frame.groupby("a", observed=False)["b"].value_counts()