Dataset.reduce methods

[jhamman]

A first attempt at implementing Dataset reduction methods.
#131

[shoyer]

Let's make this OrderedDict() instead of an unordered dictionary, just so the result will be less surprising (that is, with variables in the same order as the original).

[shoyer]

Very nice start! Please also add a try/except TypeError block to skip variables where the reduction isn't well defined.

[shoyer]

Please add a test for dimension=[]:
self.assertDatasetEqual(data.mean(dimension=[]), data)

[shoyer]

This suggests another reason why it might make more sense to loop over variables (and handle coordinates explicitly) instead of only looping over noncoordinates: it's kind of weird to lose a dimension that wasn't summed over.

[shoyer]

You could simply make this:

bad_dims = [dim for dim in dims if dim not in self.coordinates]
if bad_dims:
    raise ValueError

[shoyer]

This getting pretty close but I would like more comprehensive tests to be confident that it is working properly:

• Verify an exception is raised if a bad dimension is supplied.
• Verify that arrays with non-numeric data types are skipped if doing an inappropriate reduction (e.g., mean of a string variable).

[jhamman]

@shoyer - re. the two tests you requested.

• See test_reduce_bad_dimension in test_dataset.py.
• see test_reduce_non_numeric in test_dataset.py.

[shoyer]

To make this slightly less dissonant, let's rename da to something like var which doesn't suggest this is DataArray variable.