Collapse mean points

[duncanwp]

Follow up on #3008 - Adding support for partial collapse of multi-dimensional coordinates and now taking the mean of the points from the bounds

(I can rebase this once/if #3028 is merged).

[pelson]

I've merged #3028 if you'd like to rebase.
I support this change, but I do recognise that one could consider this to be a breaking/major change.

Relevant threads:

• Partial collapse multi-dim aux coords #3008 (review)
• Partial collapse multi-dim aux coords #3008 (comment)

Specifically, @pp-mo commented that he considers this to be a breaking change: #3008 (comment)

[duncanwp]

OK, rebased now.

What's the process with breaking changes? Could we squeeze a warning into 2.1 that this will change in 2.2?

[corinnebosley]

Hi @duncanwp, I'm afraid a warning won't let us make a breaking change in a minor version (aside from the fact that users get so many warnings they all get ignored anyway, it's just plain against the rules).

What I can do is put a label on this so that we are ready to pop it in the next major release and we don't forget about it.

It will need rebasing (again, sorry...), but it might be worth saving that until we're ready to merge it.

[rcomer]

There was a question at #3008 about whether multiple collapses are commutative. It looks like by averaging item here, which is the bounds of the input coord where available, we do lose commutativity. E.g.

mycoord = iris.coords.AuxCoord([[1, 2], [10, 11], [100, 101]], long_name='foo')
print(mycoord.collapsed())
print(mycoord.collapsed(0).collapsed())
print(mycoord.collapsed(1).collapsed())

gives:

AuxCoord(array([37]), bounds=array([[  1, 101]]), standard_name=None, units=Unit('1'), long_name='foo')
AuxCoord(array([51]), bounds=array([[  1, 101]]), standard_name=None, units=Unit('1'), long_name='foo')
AuxCoord(array([37]), bounds=array([[  1, 101]]), standard_name=None, units=Unit('1'), long_name='foo')

Is this the intended behaviour, or should the new points be averaged from the input points?

[rcomer]

Just some thoughts that occurred to me while thinking about this change. Feel free to answer/argue/ignore me as you see fit. I appreciate I've stumbled in somewhat late in the day!

• If we change the treatment of points in collapsed, should we also change it in aggregated_by for consistency (I don't know much about rolling_window, but maybe that too)?

• What happens if a weighted aggregator is passed? Should the points averaging use a weighted mean?

[bjlittle]

@duncanwp We're not going to able to get this in for iris 3.0 :-(

Our plan is to add this a into iris v3.x but as an opt-in future, which will then become the default behaviour for iris 4.0

[bjlittle]

@duncanwp Sadly this PR has been stalled for almost 2 and a half years now.

Rather than let it linger still longer, I'm going to close this PR, which is not something that I really want to do. Personally, I'd rather see it merged and its benefit utilised by the community.

However, if you have the time and motivation, then please re-open this PR and we'll take it from there. Hopefully we can get it across the line. Thanks 👍