issue_comments

I agree that we should have a "middle" or "centered" option for xarray.diff. On Mon, Sep 3, 2018 at 9:14 AM T. Chor notifications@github.com wrote:

I'm not sure where we stand on this issue, but I think since numpy.gradient already exists, it makes more sense to wrap that function for the sake of simplicity instead of making xr.diff differ from the original premise of np.diff.

On the same topic, it bothers me that xr.diff only accepts "upper" and "lower" arguments for the label. The most obvious (and useful) value in my opinion would be "middle", which would correspond to a centered finite differences. Is there any special reason why that option isn't there?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/pydata/xarray/issues/1332#issuecomment-418154705, or mute the thread https://github.com/notifications/unsubscribe-auth/ABKS1pWdcdpNtLK18J8v6z8AimuBryT_ks5uXVV3gaJpZM4Mq3x- .

I'm not sure where we stand on this issue, but I think since numpy.gradient already exists, it makes more sense to wrap that function for the sake of simplicity instead of making xr.diff differ from the original premise of np.diff.

On the same topic, it bothers me that xr.diff only accepts "upper" and "lower" arguments for the label. The most obvious (and useful) value in my opinion would be "middle", which would correspond to a centered finite differences. Is there any special reason why that option isn't there?

@rabernat, do you think that the proposed keyword additions should be included in xarray or not?

If you are going to have diff return a same-length array, then the keywords should allow the user to specify the boundary condition, e.g. extend, reflect, periodic, blank, etc. I agree this would be useful to have in xarray.

I also agree we should copy numpy wherever possible.

As I mentioned in #1288, I think basic functionality like integrate and gradient is totally within appropriate scope for xarray.

I recall now that people have requested similar functionality for numpy.diff: https://github.com/numpy/numpy/issues/8132. It would be nice to resolve this upstream in NumPy first (e.g., with https://github.com/numpy/numpy/pull/8206), and then simply copy the API design in xarray.

I'm not sure we want to wrap np.gradient. It seems like other approaches like @rabernat 's xgcm would be more appropriate as a superset of xarray.

Certainly grid-aware differencing and integral operators are preferred when the grid information is known and available, but I'm not sure that therefore a more naive version akin to np.gradient would not be useful. It's quite likely that there are xarray users (e.g. in non climate/weather/ocean-related fields) wherein a 'c' grid is meaningless to them, yet they still would appreciate being able to easily compute derivatives via xarray operations.

But then we're back to the valid questions raised before re: what is the appropriate scope of xarray functionality, c.f. https://github.com/pydata/xarray/issues/1288#issuecomment-283062107 and subsequent in that thread

@rabernat, do you think that the proposed keyword additions should be included in xarray or not? I personally would like to see them in xarray but don't know if it is just me or not. If you think they should be in xarray, are you ok with the api above?

It is just super inconvenient to do array resizing following the diff of a time vector to get timesteps, but maybe this use case is too niche to be useful for the community.

Maybe for the xarray community, but not the xgcm community ;) We definitely want round-trip-compatibly diff and cumsum operations (see xgcm/xgcm#49, which is very close to implemented)

@shoyer, I'm not sure we want to wrap np.gradient. It seems like other approaches like @rabernat 's xgcm would be more appropriate as a superset of xarray.

Fundamentally, I want something that is like an inverse of cumsum and the proposed change could be used in that context. It is just super inconvenient to do array resizing following the diff of a time vector to get timesteps, but maybe this use case is too niche to be useful for the community.

FYI @pwolfram, if you are interested in more "grid-aware" finite differencing, you might have a look at xgcm. We are working on it again and have implemented some basic difference and interpolation operators: http://xgcm.readthedocs.io/en/latest/grids.html#grid-objects

Would it fill the same need to add a wrapped version of np.gradient, which already works similarly to diff but preserves input shape?