multi-dimensional category_edges for contingency, reliability, etc``

[aaronspring]

so far with xhistogram, we can only create 1d category_edges. however, sometimes we would want to make category_edges multi-dimensional (dependent on lon and lat) when e.g. calculating wrt. climatological terciles.

inspired by #275 and used in #277

I found a way to bin datasets with multi-dim edges resulting in the same results as xhistogram.

e=[.1,.2,.4,.5,.7,.9]
edges = xr.DataArray(e, dims='edge')
(ds<edges).sum('member').diff('edge') == xs.core.utils.histogram(ds,bins=[np.array(e)],dim=['member']).rename({'var1_bin':'edge'})

sounds reasonable @dougiesquire ?

[dougiesquire]

Seems reasonable to me. And this adds really nice functionality - nice work!

Does this mean we should phase out our dependency on xhistogram? Alternatively, we could suggest that this type of functionality is added to xhistogram?

[aaronspring]

add to xhistogram could be nice. Although it seems they are mostly focusing on 1d edges. We could raise an issue there

[dougiesquire]

Also, just became aware of this:
https://boost-histogram.readthedocs.io/en/latest/notebooks/xarray.html
Haven't tried to use it yet, but apparently plays nicely with xarray, and faster than xhistogram. Not sure whether there is support for multidimensional bins

[dougiesquire]

Also, just became aware of this:
https://boost-histogram.readthedocs.io/en/latest/notebooks/xarray.html
Haven't tried to use it yet, but apparently plays nicely with xarray, and faster than xhistogram. Not sure whether there is support for multidimensional bins

On further inspection, I think not...

[aaronspring]

Use pydata/xarray#5400