Inconsistent treatment of None with .where() and inplace

[Wilfred]

Calling .where() treats None differently depending on the value of inplace.

In [1]: import pandas

In [2]: s1 = pandas.Series(['a', 'b', 'c'])

In [3]: s1.where(s1 != 'a', None)
Out[3]: 
0    None
1       b
2       c
dtype: object

In [4]: s2 = pandas.Series(['a', 'b', 'c'])

In [5]: s2.where(s1 != 'a', None, inplace=True)

In [6]: s2
Out[6]: 
0    NaN # would have expected None here!
1      b
2      c
dtype: object

This causes surprising behaviour when setting values in a series to None.

In [1]: import pandas

In [2]: s3 = pandas.Series(['a', 'b', 'c'])

In [3]: s3[0] = None

In [4]: s3[s3 == 'b'] = None

In [5]: s3
Out[5]: 
0    None
1     NaN # Expected None here too!
2       c
dtype: object

[jreback]

None is a real value, and you are using this on an object dtyped series. On other dtypes this works fine. Not sure that this is possible to deal with. How do you know that I don't want to store None in s3[0] = None (as apposed to NaN) above?

related is this: #7657 (this has to do with a mixed-dtype frame)

[Wilfred]

I've edited to clarify that I was expecting None in my Series. I do want to store None.

[jreback]

why do you want to actually store None? NaN is the missing value indicator

[Wilfred]

It's useful to store None when we're putting database data in pandas Series objects.

[jreback]

to_sql does this conversion for you.

[Wilfred]

We're not using a SQL database I'm afraid.

I've opened pull request #7941 to show one possible solution.

[jreback]

This is actually consistent now. With both series/frame having the same behavior, e.g. they presereve. None when inplace=False and convert when inplace=True. Further complicating this is a straight setitem s[0] = None works now, with no conversion at all.

So unless a comprehensive soln exists, can't do anything about this. I'll leave this issue open and if you want to address pls do so.