issue_comments

In order to maintain a list of currently relevant issues, we mark issues as stale after a period of inactivity

If this issue remains relevant, please comment here or remove the stale label; otherwise it will be marked as closed automatically

Ok, thank you the information. I first worked with the API Clearly documented in the link you provided.

I noticed the attrs get lost after orthogonal interpolation (see the first/second plots of arr in mybinder, I might open a new issue for that

Thanks for your deeper analysis.

It seems everything's well with xarray.

Happy to hear that.

I first thought i'll get a 1D array which is not the case (this is often the behavior I want).

Our interp is working orthogonally by default, so passing two arrays sized 10,000 will result in interpolation of 100,000,000 values. In order to get a 1D array, you can pass two dataarrays with the same dimension, python new_tension = xr.DataArray(new_tension, dims='new_dim') new_resistance = xr.DataArray(new_resistance, dims='new_dim') arr.interp(tension=new_tension, resistance=new_resistance) which gives a 1d array with the new dimension new_dim. See here for the details.

I want to keep this issue open, as the performance can be increased for such a case.

In the above example, python arr.interp(tension=new_tension, resistance=new_resistance) and python arr.interp(tension=new_tension).interp(resistance=new_resistance) gives the same result (for 'linear' and 'nearest' methods), but the latter runs much faster. This difference looks similar to the difference between our orthogonal indexing and vectorized indexing. We may need orthogonal interpolation path, which would significantly increase the performance in some cases.

Thanks for your comment

had a deeper look at DataArray.interp and it computes a new DataArray with new coords given by the dict. I first thought i'll get a 1D array which is not the case (this is often the behavior I want). I then compared first 10_000 interpolations on sdf against 100_000_000 in xarray! It explains the gap.

Updated my comparison with the same behavior with scipy and sdf in a jupyter notebook on mybinder

It seems everything's well with xarray.

Extrapolation (linear first) would be a good feature too, I put an example at the end of the notebook about sdf interpolation/extrapolation possibilites (work for nd-arrays of dim 32 as with numpy)

@gwin-zegal , thank you for using our new feature and reporting the issue. I confirmed the poor performance of interp.

I will look inside later, whether problem is on our code or upstream (scipy.interpolate).

A possible workaround for your code is to change python arr.interp({'tension': new_tension, 'resistance': new_resistance}) to python arr.interp({tension': new_tension}).interp('resistance': new_resistance}) but it does not solve all the problems.