How to apply "mode" function to coarsen object

[chfite]

I was wondering if it is possible to apply "mode" to a coarsened array object, instead of mean, min, or max, etc. I know that numpy does not have a mode function, and so I was wanting to use the scipy.stats.mode function and apply it. I've looked at xr.apply_ufunc() but have had little luck getting it to work taking the mode of the coarsened object.

For example, lets say I have a dataarray with dims (lat, lon) and shape (415, 725) and I want to coarsen it every 5 grid cells. I was going to do datarray.coarsen(lat=5, lon=5) but am not sure the best way to get the mode of my values that are "grouped" in each larger gridcell, which is what I need.

[dcherian]

coarsen very recently gained a .reduce method which will be included in an upcoming release very soon. You could test it out on the master branch in the mean time.

[spencerkclark]

Indeed, this is a good application for the forthcoming .reduce method. That said, even with .reduce this is somewhat of a tricky (but solvable) problem, because scipy.stats.mode does not have the same interface as traditional NumPy reduction functions like mean or sum. The main challenge is wrapping it in such a way that you can do a mode reduction along multiple axes at once by passing a tuple of axes as an argument (currently it can only take a single axis). NumPy's private _ureduce function is a good illustration for how to do this; I would be curious to know if there is a public equivalent of that in NumPy or some other library.