DataFrame.where does not work as expected with inplace=True

[rockg]

related #4667, #7647

According to the documentation, inplace should replace values appropriately with other, but this does not seem to work as expected. This worked in 0.12.

In [6]: df = pandas.DataFrame([{'A': 1, 'B': np.nan, 'C': 'Test'}, {'A': np.nan, 'B': 'Test', 'C': np.nan}])

In [7]: df
Out[7]: 
    A     B     C
0   1   NaN  Test
1 NaN  Test   NaN

In [8]: df1 = df.where(~pandas.isnull(df), None)

In [9]: df1
Out[9]: 
      A     B     C
0     1  None  Test
1  None  Test  None

In [10]: df.where(~pandas.isnull(df), None, inplace=True)

In [11]: df
Out[11]: 
    A     B     C
0   1   NaN  Test
1 NaN  Test   NaN

[jreback]

see #4667.

doing an .where with inplace=True is de-facto df[mask] = value

That does dtype inference on the scalar values and thus the result is never object when presented with None.

e.g.

In [1]: df = DataFrame({'series': Series(range(10))}).astype(float)

In [2]: df
Out[2]: 
   series
0       0
1       1
2       2
3       3
4       4
5       5
6       6
7       7
8       8
9       9

In [3]: df.dtypes
Out[3]: 
series    float64
dtype: object

In [4]: df[df>7] = None

In [5]: df
Out[5]: 
   series
0       0
1       1
2       2
3       3
4       4
5       5
6       6
7       7
8     NaN
9     NaN

I guess this is an inconsistency, not sure what to do with this. Fixing this breaks the more important case above.

[jreback]

similar to this as well: #7647

[jreback]

will mark as a bug/api for 0.15, though not sure what if anything can/should do

[rockg]

Perhaps at the very least the doc should be updated to reflect this issue. Namely, that inplace won't work with None as some might expect, but assignment works fine.

[jreback]

yep, could have a doc mention if you'd like. Pls submit a pull-request. FYI, inplace operatiosn are in IMHO not very useful (and not faster/more efficient in general).

[jreback]

@rockg actually put up #7657 which makes this raise a TypeError (to be consistent with an expression like df[mask] = None).