Pivoting on multiple columns causes data loss

[ukch]

Seen on HEAD. This problem is new since 0.5.0.

In [3]: data
Out[3]: 
   foo   bar   baz   spam   data
0  foo1  bar1  baz1  spam2  20  
1  foo1  bar2  baz1  spam3  30  
2  foo2  bar2  baz1  spam2  40  
3  foo1  bar1  baz2  spam1  50  
4  foo3  bar1  baz2  spam1  60  

In [4]: pandas.pivot_table(data, values="data", rows=["foo", "bar"], cols=["baz", "spam"])
Out[4]: 
  baz      baz1   baz2 
  spam     spam1  spam1
foo  bar               
foo1 bar1  20     50   
     bar2  30     NaN  
foo2 bar2  40     NaN  
foo3 bar1  NaN    60   

As you can see, the ("baz1", "spam2"), ("baz2", "spam2"), ("baz1", "spam3") and ("baz2", "spam3") columns have disappeared and their contents have been aggregated into the remaining two columns.

[wesm]

Indeed looks something went haywire in groupby it looks like-- I'll sort this out today.

[wesm]

thanks a lot for catching this. fixed in the above commit. i'm going to overhaul some of the multi-key groupby code for 0.6.0 later this year