from_frame should handle a dataframe with a multi-index in columns

[alixdamman]

[gdementen]

arj & his team need this (or at least a work-around) quickly

[gdementen]

Here is a test case:

index = pd.MultiIndex.from_product((['a0', 'a1'], ['b0', 'b1']), names=('a', 'b'))
columns = pd.MultiIndex.from_product((['c0', 'c1'], ['d0', 'd1']), names=('c', 'd'))
df = pd.DataFrame(np.arange(16).reshape((4, 4)), index=index, columns=columns)
# c      c0      c1    
# d      d0  d1  d0  d1
# a  b                 
# a0 b0   0   1   2   3
#    b1   4   5   6   7
# a1 b0   8   9  10  11
#    b1  12  13  14  15
arr = asarray(df)
# currently fails because we get a 3d array with names ['a', 'b', None] instead
assert arr.axes.names == ['a', 'b', 'c', 'd']

[gdementen]

We also need to support duplicated axis names, which the workaround I just tried using split_axes chokes on: E.g:

# works for the unambiguous axis name
arr.rename(-1, 'c|d').split_axes(-1, sep='|', regex='([^,]+),(.*)')

index = pd.MultiIndex.from_product((['a0', 'a1'], ['b0', 'b1']), names=('a', 'a'))
columns = pd.MultiIndex.from_product((['c0', 'c1'], ['d0', 'd1']), names=('a', 'a'))

or even duplicated axes and labels (which do not seem to cause any extra problem:

index = pd.MultiIndex.from_product((['a0', 'a1'], ['a0', 'a1']), names=('a', 'a'))
columns = pd.MultiIndex.from_product((['a0', 'a1'], ['a0', 'a1']), names=('a', 'a'))
df = pd.DataFrame(np.arange(16).reshape((4, 4)), index=index, columns=columns)

[gdementen]

I should recover the test cases from #1132