BUG: groupby.nth reorders categories

[rhshadrach]

df = pd.DataFrame(
    {"x": pd.Categorical([4, 4, 2, 2, 4], categories=[1, 2, 3, 4]), "y": range(5)}
)

gb_ait = df.groupby('x', observed=True, sort=False, as_index=True)
gb_aif = df.groupby('x', observed=True, sort=False, as_index=False)

result1 = gb_ait.nth(0).reset_index()
result2 = gb_aif.nth(0)

print(result1["x"])
print()
print(result2["x"])

# 0    4
# 1    2
# Name: x, dtype: category
# Categories (4, int64): [4, 2, 1, 3]
# 
# 0    4
# 2    2
# Name: x, dtype: category
# Categories (4, int64): [1, 2, 3, 4]

While the values are the same, the order of the categories are not. I'm not certain which one here is correct - is the expectation that sort=False orders the categories by appearance?

[rhshadrach]

Reducers seem to give order of appearance, transforms maintain the original order. But I've only tested a handful of reducers / transformers thus far.

Reducers

df = pd.DataFrame(
    {"x": pd.Categorical([4, 4, 2, 2, 4], categories=[1, 2, 4, 3]), "y": range(5)}
)

gb_ait = df.groupby('x', observed=True, sort=False, as_index=True)
gb_aif = df.groupby('x', observed=True, sort=False, as_index=False)

result1 = gb_ait.sum().reset_index()
result2 = gb_aif.sum()

print(result1["x"])
print()
print(result2["x"])

# 0    4
# 1    2
# Name: x, dtype: category
# Categories (4, int64): [4, 2, 1, 3]
# 
# 0    4
# 1    2
# Name: x, dtype: category
# Categories (4, int64): [4, 2, 1, 3]

Transformers

df = pd.DataFrame(
    {"x": pd.Categorical([4, 4, 2, 2, 4], categories=[1, 2, 4, 3]), "y": range(5)}
).set_index('x')

gb_ait = df.groupby('x', observed=True, sort=False, as_index=True)
gb_aif = df.groupby('x', observed=True, sort=False, as_index=False)

result1 = gb_ait.cumsum().reset_index()
result2 = gb_aif.cumsum().reset_index()

print(result1["x"])
print()
print(result2["x"])

# 0    4
# 1    4
# 2    2
# 3    2
# 4    4
# Name: x, dtype: category
# Categories (4, int64): [1, 2, 4, 3]
# 
# 0    4
# 1    4
# 2    2
# 3    2
# 4    4
# Name: x, dtype: category
# Categories (4, int64): [1, 2, 4, 3]

[rhshadrach]

Duplicate of #48749