Because Pandas changed from a version module to __version__ only

Also simplified, assuming __version__ is always there.

It relied directly on pandas string rather than using the tools.testing
function.

Numpy now follows its own documentation, but that's new!

Main differences: 'view' doesn't always return a ndarray and an array
with a single line or column is always both F and C contiguous

At least as far as tests are concerned

Now using LooseVersion for pandas, which seems to be working.

Now handles: < 0.14 (no inplace for Series)
             >= 0.14 < 0.17 (Series have inplace)
             >= 0.17 (using sort_values instead of sort)