Evgeni Burovski
Evgeni Burovski
Is this about `distance_upper_bound`? This is indeed new in SciPy 1.12: https://github.com/scipy/scipy/pull/19483
I guess I'd need to explain a bit the ordering here. This is indeed opinionated :-) and is very much up to a discussion. **First phase**: once these are all...
One more big scipy dependency: RBFInterpolator depends on KDTree. cf gh-7190
I think this would be very welcome @ideasrule! Not sure on the performance target, this is something we'll need to figure out indeed. That said, porting the scipy version to...
The current spsolve implementation uses a python loop over the 2nd axis of the r.h.s.: https://github.com/cupy/cupy/blob/main/cupyx/scipy/sparse/linalg/_solve.py#L518 Possible ways to improve: - push the loop to the C/Cython level. - in...
THis is basically what the `_qr_solve` (the third option above) does under the hood: https://github.com/cupy/cupy/blob/main/cupyx/scipy/interpolate/_bspline2.py#L688
Why closing as completed? make_interp_spline is still using a python loop, only make_lsq_spline invokes _qr_solve ATM
At least for FITPACK, I think the only sane approach could be to rewrite the Fortran implementations first as a python prototype to break the monolith of the original Dierckx...
lgpl licensing is incompatible, I'm afraid
> What's the closest function in CuPy to the LinearNDInterpolator for N-dimensions? Would it be the RBFInterpolator using the 'linear' kernel? From what is implemented at the moment, yes, RBF...