Deepak Cherian comments

Results 1084 comments of


                                            Deepak Cherian

`sum(..., skipna=True)` should dispatch to `sparse.nansum` for sparse arrays

I support us attempting to detect if `xp.nan*` exists, and cache the results of this lookup, because of the perf impact here; and because we choose `nan*` reductions by default...

`sum(..., skipna=True)` should dispatch to `sparse.nansum` for sparse arrays

Yes, `__array_function__` would be fine. Indeed I'm surprised we don't do that already? @peanutfun are you able to dig in here?

`sum(..., skipna=True)` should dispatch to `sparse.nansum` for sparse arrays

I guess we could try to dispatch `nansum` using `__array_function__`, and use `sum_where` as a fallback? > Combining datasets after applying separate masks via fillna or combine_first should be relatively...

Lazy concatenation of arrays

From @rabernat in #6588: Right now, if I want to concatenate multiple datasets (e.g. as in open_mfdataset), I have two options: - Eagerly load the data as numpy arrays ➡️...

`Dataset.interp()` silently dropping time-like data arrays

This should be an easy fix with a `.view(np.int64)` and undoing that at the end.

Support rechunking to a frequency.

Ready for review! The syntax is `ds.chunk(time=TimeResampler("YE"))`.

Support rechunking to a frequency.

This is pretty cool, and could use a review!

Fix 1931

OK I'll try and add a hypothesis test for this later.

Fix 1931

@jhamman I have reproduced the failure with a hypothesis test now (#1746 ). This branch makes that test always pass (it seems like) Should I point that PR at main...

follow upstream scipy interpolation improvements

Nice! > the goal is to support tensor product interpolation wherever scipy interpn supports it, and remove any no-longer-necessary dimension checks Yes, I believe so.