Fix Y binning and revert TH2.numpy behavior

[nsmith-]

Fixes #31
Changes to return type of TH2.numpy should come with a minor version change

[jpivarski]

Thanks for the fix to the attribute name (surprising that the Axis class uses the letter "x" in its names), but the change to the output of the numpy property was deliberate. It's for similarity with the numpy.histogram2d function, which returns a flat, 3-item tuple. Do you disagree that this should be a goal? (I'm open to counter-arguments; the above is my reasoning.)

[lgray]

Either format is fine to me. So long as it's well advertised. :-)

[nsmith-]

@jpivarski to me, nested edges generalizes to n-D better, but since the method is called numpy it should probably behave like numpy. One can always get edges and values separately. What about whether to include overflow?

[jpivarski]

These methods map onto ROOT histograms, which only go up to 3D. However, even if generalization was an issue, I don't see how unpacking flat vs. nested is different for an unknown number of dimensions:

values, edges = hist.numpy()

Gives you edges as an array for 1D and a 2- or 3-tuple of arrays for 2D or 3D, regardless of whether the result was flat or nested. I assume this is how you'd want to unpack it. If you know the dimensionality, you'd say

values, xedges, yedges = hist.numpy()

for flat and

values, (xedges, yedges) = hist.numpy()

for nested. So it's only if you do know the number of dimensions that there's a difference.

It wouldn't be like Numpy to include overflow bins, but following the other conventions, that could be a hist.allnumpy() method.

[nsmith-]

Works for me, done