Evgeni Burovski

In the end of the day, I'm 99/100 certain that anyhow it boils down to a variation of ``` if ndim == 1: # use a 1D interpolator else: ......

Given that the original PR, https://github.com/scipy/scipy/pull/13432, was about linear and nearest scattered ND interpolators, and that the griddata code is literally (https://github.com/scipy/scipy/blob/main/scipy/interpolate/_ndgriddata.py#L240) ``` if ndim == 1 and method in...

Makes sense! I'm with you on both counts; so the status here is then "triaged, waiting for a contributor": if somebody wants this enough to send a PR, we'll review.

> We should really check for deadlinks when building the doc. IIRC we tried running linkcheck, it did smoke out a few problems, but turned out to be too noisy...

Better not add an extra layer of pytest skips. Ideally, remove scipy.misc from the list of submodules to doctest, add scipy.datasets (the latter is a good move regardless of the...

> We can add a private derivative method I'd suggest to copy-paste it to scipy stats as a private function now and use it.

As notebooks are discussed, linking a stale issue https://github.com/scipy/scipy/issues/5233

At a minimum, this seems to need an explicit atol/rtol.

Two quick notes: - this is a clean-room reimplementation of the GCV (generalized cross-validation) spline smoothing algorithm. The canonical implementation, gcvspl.f, is packaged in e.g. Matlab/Octave, but is license-incompatible, see...

@dschmitz89 Looks related indeed. I'd expect this to be lower level, and certainly much more narrow in functionality. Will need to check the references in https://pygam.readthedocs.io/en/latest/notebooks/tour_of_pygam.html#Generalized-Additive-Models,-in-general; will take a look...