[Python] Why does pd.isnull behave differently on DataFrame vs. single element?

I'm noticing an inconsistency in how pd.isnull behaves.

Given that pd.isnull('nan') returns False, but pd.isnull(float('nan')) returns True, I would have expected that applying pd.isnull to a DataFrame column previously converted to the 'Float64' dtype would return True for any value that was originally the string 'nan'. But that's not what I'm seeing.

Here's an example:

import pandas as pd
import numpy as np

df = pd.DataFrame([['1', pd.NA, 'nan'], ['re', 'ui1', np.nan]], columns=['a', 'b', 'c'])

pd.isnull(df['c']) # Returns [False, True] ==> expected
pd.isnull(df.loc[0, 'c']) # Returns False ==> expected
pd.isnull(df[['c']].astype('Float64')) # Returns [False, True] ==> unexpected?
pd.isnull(df[['c']].astype('Float64').values) # Returns [True, True] ==> expected?
pd.isnull(df[['c']].astype('Float64').loc[0, 'c']) # Returns True ==> expected


Why does pd.isnull exhibit different behavior when applied to an entire column or dataset compared to a single element? Am I missing something?

python: 3.9.12 ; pandas: 2.2.3 ; numpy:1.26.4

Continue reading...