ser = pd.Series(pd.Categorical([1], categories=[1, 2, 3]))
print(ser)
# 0 1
# dtype: category
# Categories (3, int64): [1, 2, 3]
ser = ser.replace(1, None)
print(ser)
# 0 NaN
# dtype: category
# Categories (2, object): [2, 3]
One gets the same behavior if None is replaced by pd.NA, and similar behavior with np.nan but this is coerced to float. Since the categories can be integers and the Series still hold NA values, it seems to me this shouldn't coerce to object.
ser = pd.Series(pd.Categorical([None], categories=[1, 2, 3]))
print(ser)
# 0 NaN
# dtype: category
# Categories (3, int64): [1, 2, 3]
Probably unrelated: #46884
Comment From: lukemanley