BUG: applymap on empty DataFrame returns Series

[ojdo]

I'm unsure about a corner case I encountered:

>>> pd.DataFrame([]).applymap(round)
Series([], dtype: float64)

full example

In [1]: df = pandas.DataFrame({ 'x' : [], 'y' : [], 'z' : []})

In [2]: df
Out[2]: 
Empty DataFrame
Columns: [x, y, z]
Index: []

In [3]: df.applymap(lambda x: x)
Out[3]: 
x   NaN
y   NaN
z   NaN
dtype: float64

In [4]: df = pandas.DataFrame({ 'x' : [1], 'y' : [1], 'z' : [1]})

In [5]: df.applymap(lambda x: x)
Out[5]: 
   x  y  z
0  1  1  1

I would have expected to get the empty DataFrame back.

[jreback]

the test for reduction is on an empty Series (e.g. round(Series([])))

so maybe not that sophisticated.

pull-request welcome to fix (though maybe tricky).

[stringfellow]

This is an API change from 0.12 and is not documented!

0.12:

In [1]: import pandas as pd

In [2]: df = pd.DataFrame()

In [3]: df
/Users/spike/.virtualenvs/pd_0_12/lib/python2.7/site-packages/pandas/core/config.py:570: DeprecationWarning: height has been deprecated.

  warnings.warn(d.msg, DeprecationWarning)

Out[3]:
Empty DataFrame
Columns: []
Index: []

In [4]: df.applymap(lambda x: 1)
Out[4]:
Empty DataFrame
Columns: []
Index: []

0.15:

In [1]: import pandas as pd

In [2]: df = pd.DataFrame()

In [3]: df
Out[3]:
Empty DataFrame
Columns: []
Index: []

In [4]: df.applymap(lambda x: 1)
Out[4]: Series([], dtype: float64)

[jreback]

@stringfellow well I suspect it actually changed in 0.13. Since their wasn't a test, a regression like this is very hard to detect. As I said above a pull-request to restore this behavior would be welcome.

I just checked this with 0.18.0 and I still get a Series back. Is this supposed to be fixed in the near future?

[jreback]

this issue is still open, pull-requests are welcome