Use axis = pd.DataFrame()._get_axis_number(axis) instead.

What would this do here? I'm checking if q is a list or not

numeric_only specifies whether or not to filter the non-numeric columns. Does not need error checking.

If numeric_only is true, then pandas returns value errors for mixed dtype dataframes

I'm not able to reproduce, seems that pandas drops non-numeric columns when numeric_only is True.

In [3]: df
Out[3]: 
   col1 col2
0     1    a
1     2    b
2     3    c

In [4]: df.dtypes
Out[4]: 
col1     int64
col2    object
dtype: object

In [5]: df.quantile(numeric_only=True)
Out[5]: 
col1    2.0
Name: 0.5, dtype: float64

This is true also in the axis=1 case.

In [6]: df.quantile(numeric_only=True, axis=1)
Out[6]: 
0    1.0
1    2.0
2    3.0
Name: 0.5, dtype: float64

Whoops, when numeric_only is FALSE, it returns TYPEerrors.

Try on

df = pd.DataFrame({"A": [1, 2, 
            "B": [2., 3., 4.],
            "C": pd.date_range('20130101', periods=3),
            "D": ['foo', 'bar', 'baz']})
df.quantile(.5, axis=1, numeric_only=False)

This returns a TypeError

This is true, although the check here does not handle this case. The error I see is TypeError: Cannot compare type 'Timestamp' with type 'float'.

Use the pandas _check_percentile method here.

Updated

When would this ValueError be thrown?

This exception is thrown when there are only non-numeric columns in this partition
Added comment

I see, this should probably be handled somewhere to ensure concordance.

Describe also handles it like this, so if we do something about it we should make sure to modify it there as well

_arithmetic_helper should return a Series.

This is true, although the check here does not handle this case. The error I see is TypeError: Cannot compare type 'Timestamp' with type 'float'.

As discussed above, TypeErrors are thrown here. Perhaps you can add some logic at the beginning of the function to ensure all the dtypes are comparable.

The issue seems to arise not from non-numeric data, but from non-comparable types.

The lines should be consolidated.

Wanted to keep all the datetime imports on their own line, should I still do this?

If you do this, you need to remove test_series.

Should I just move the entire file?

No, it just needs to be removed from the .travis.yml file

Don't we want to run those tests? Or no?

While Series is unimplemented, we are importing it directly from Pandas. Skipping these tests for now as they expect NotImplementedErrors to be thrown.

I know that this error is concordant with Pandas, but it feels very out of place in this context. Perhaps we should change it to something more descriptive. Honestly, this case seems like a bug with Pandas, it should handle the case where all columns are dropped.

Looping in @devin-petersohn here.

I agree -- I originally had a more descriptive error but scrapped it in favor of using the pandas one. Something that references mixed dtypes would be better

As before use axis = pd.DataFrame()._get_axis_number(axis) here

Seems that np.datetime64 type columns are supported.

In [8]: x
Out[8]: 
                 col1
0 2018-05-04 20:50:29
1 2018-05-04 20:50:29
2 2018-05-04 20:50:29

In [9]: x.dtypes
Out[9]: 
col1    datetime64[ns]
dtype: object

In [10]: x.quantile(q=0.5, axis=1, numeric_only=True)
Out[10]: 
0   NaN
1   NaN
2   NaN
Name: 0.5, dtype: float64

Made a fix to handle all datetime objects

Why is this function defined twice?

One returns a Series the other returns a DataFrame