Bug: `_get_weights_and_dims` doesn't work properly about esmlab

[andersy005]

It appears that the issue is in_apply_nan_mask(). This function assumes that all dimensions should be used:

esmlab/esmlab/statistics.py

Lines 51 to 53 in 479fa70

	weights = xr.DataArray(np.ones(dims_shape), dims=op_over_dims)
	weights = _apply_nan_mask(weights, x, y)

esmlab/esmlab/statistics.py

Lines 21 to 32 in 479fa70

	def _apply_nan_mask(weights, x, y=None):
	# If y is specified, make sure x and y have same shape
	if y is not None and isinstance(y, xr.DataArray):
	assert x.shape == y.shape
	valid = x.notnull() & y.notnull()
	else:
	valid = x.notnull()
	# Apply nan mask
	return weights.where(valid)

As a consequence, the shape of the returned weights is always equal to the input array's shape, even when the user wants to apply the operation along a particular dimension.

To fix this, we will need to come up with a way of doing valid = x.notnull() while taking into consideration dimensions supplied by the user.

[andersy005]

Fixed in #97