Issues with numeric_only for DataFrame.std()

[mortada]

The docstring shows a numeric_only option for DataFrame.std() but it does not seem to actually be implemented. I'm happy to take a crack at fixing it but I'm not sure whether it's the doc or the implementation that needs fixing.

To see this consider a mixed-type DataFrame where I'm setting one entry to be a str of '100' while all other entries are float. For std() It does not matter whether numeric_only is True or False, but for max() it clearly makes a difference.

In [1]: import pandas as pd
In [2]: import numpy as np
In [3]: df = pd.DataFrame(np.random.randn(5, 2), columns=['foo', 'bar'])
In [4]: df.ix[0, 'foo'] = '100'

In [5]: df
Out[5]:
         foo       bar
0        100 -1.958036
1   0.221049  0.309971
2   1.200093 -0.103244
3  -2.475388 -2.279483
4  0.1623936 -1.185682

In [6]: df.std(numeric_only=True)
Out[6]:
foo    44.841828
bar     1.129182
dtype: float64

In [7]: df.std(numeric_only=False)
Out[7]:
foo    44.841828
bar     1.129182
dtype: float64

In [8]: df.max(numeric_only=False)
Out[8]:
foo    100.000000
bar      0.309971
dtype: float64

In [9]: df.max(numeric_only=True)
Out[9]:
bar    0.309971
dtype: float64

[jreback]

numeric_only matters if axis=1 otherwise the dtype controls what happens
your first column is an object dtype
max and min will work on an object dtype (though not sure what std should do here - you would have to trace it and see)

[mortada]

I see. But even for axis=1 it seems to be inconsistent. You can see below that numeric_only has an effect on max(), but for std() it makes no difference

In [17]: df.max(axis=1, numeric_only=True)
Out[17]:
0   -1.958036
1    0.309971
2   -0.103244
3   -2.279483
4   -1.185682
dtype: float64

In [18]: df.max(axis=1, numeric_only=False)
Out[18]:
0    100.000000
1      0.309971
2      1.200093
3     -2.279483
4      0.162394
dtype: float64

In [19]: df.std(axis=1, numeric_only=True)
Out[19]:
0    72.095218
1     0.062878
2     0.921598
3     0.138526
4     0.953233
dtype: float64

In [20]: df.std(axis=1, numeric_only=False)
Out[20]:
0    72.095218
1     0.062878
2     0.921598
3     0.138526
4     0.953233
dtype: float64

[jreback]

hmm could be a buggie

[mortada]

@jreback I made a PR for this #9209