WIP/TST: Added test of unstacking two levels with NaN.

[seth-p]

Added a test to test_unstack_nan_index() for unstacking two levels simultaneously when one of them contains a NaN.

Should delete the print() statements before merging...

This is the output I get:

======================================================================
FAIL: test_unstack_nan_index (pandas.tests.test_frame.TestDataFrame)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "c:\Users\seth\github\pandas2\build\lib.win-amd64-3.4\pandas\tests\test_frame.py", line 12439, in test_unstack_nan_index
    assert_series_equal(left, right)
  File "c:\Users\seth\github\pandas2\build\lib.win-amd64-3.4\pandas\util\testing.py", line 674, in assert_series_equal
    assert_almost_equal(left.values, right.values, check_less_precise)
  File "das\src\testing.pyx", line 58, in pandas._testing.assert_almost_equal (pandas\src\testing.c:2745)
  File "das\src\testing.pyx", line 93, in pandas._testing.assert_almost_equal (pandas\src\testing.c:1830)
  File "das\src\testing.pyx", line 135, in pandas._testing.assert_almost_equal (pandas\src\testing.c:2514)
nose.proxy.AssertionError: (very low values) expected 3.00000 but got 0.00000, with decimal 5
-------------------- >> begin captured stdout << ---------------------
dfs=
       C
A B     
a 0    0
  1    1
  2    2
  NaN  3
b 4    4
  5    5
  6    6
  7    7

left=
   A  B
C  a  1    0
      2    1
      4    2
      0    3
   b  6    4
      7    5
   a  0    6
      1    7
dtype: int32

right=
   A  B  
C  a  NaN    3
      0      0
      1      1
      2      2
   b  4      4
      5      5
      6      6
      7      7
dtype: int32

--------------------- >> end captured stdout << ----------------------

[seth-p]

cc: @behzadnouri

[seth-p]

@jreback, @behzadnouri: In view of @behzadnouri's observation #9023 (comment) that implementing DataFrame.unstack() in terms of (my modified) DataFrame.stack() slowed things down tremendously, I tried going the other way and implementing DataFrame.stack() in terms of DataFrame.unstack() (after making a few simple modifications to the latter to support dropna and sequentially). I noticed that then some of my stack() tests were failing, and isolated this case as an example.

[seth-p]

For comparison, this is the result in 0.15.2 in the absence of NaNs.

In [1]: import pandas as pd

In [2]: df = pd.DataFrame({'A': list('aaaabbbb'),'B':range(8), 'C':range(8)})

In [3]: dfs = df.set_index(['A', 'B'])

In [4]: dfs.unstack([0, 1])
Out[4]:
   A  B
C  a  0    0
      1    1
      2    2
      3    3
   b  4    4
      5    5
      6    6
      7    7
dtype: int32

[seth-p]

[Recommitted code with correct expected right.]

[jreback]

@seth-p pls submit this as a separate issue report

[seth-p]

Created #9497.