fillna(None) inconsistent with fillna(1) or fillna(np.NaN)

[PhE]

Migrated from StackOverflow comment : http://stackoverflow.com/questions/12598520/set-index-on-multiple-columns-with-one-empty-column

    df = DataFrame([
    dict(a=1, p=0), 
    dict(a=2, m=10), 
    dict(a=3, m=11, p=20), 
    dict(a=4, m=12, p=21)
    ], columns=('a', 'm', 'p', 'x'))

. a m p x
0 1 NaN 0 NaN
1 2 10 NaN NaN
2 3 11 20 NaN
3 4 12 21 NaN

Applying fillna(None) turns one value to 0 !

df.fillna(None)

. a m p x
0 1 NaN 0 NaN
1 2 10 0 NaN
2 3 11 20 NaN
3 4 12 21 NaN

Applying fillna(1) works as expected :

df.fillna(1)

. a m p x
0 1 1 0 1
1 2 10 1 1
2 3 11 20 1
3 4 12 21 1

Applying fillna(np.NaN) works as expected :

df.fillna(np.NaN)

. a m p x
0 1 NaN 0 NaN
1 2 10 NaN NaN
2 3 11 20 NaN
3 4 12 21 NaN

Could be related to #1971

[changhiskhan]

Note that fillna(None) is equivalent to fillna(), which means the value parameter is unused. Instead, it uses the method parameter which is by default forward fill. Which is why the NaN after the 0 in column p is filled.

[hayd]

Note: you can do this using the where method:

In [10]: df.where(pd.notnull(df), None)
Out[10]: 
   a     m     p     x
0  1  None     0  None
1  2    10  None  None
2  3    11    20  None
3  4    12    21  None

[thadd3us-abcellera]

Note: you can do this using the where method:

In [10]: df.where(pd.notnull(df), None)
Out[10]: 
   a     m     p     x
0  1  None     0  None
1  2    10  None  None
2  3    11    20  None
3  4    12    21  None

The df.where method no longer works at Pandas>=1.3. Instead, use:

df.replace({np.nan: None})