Evgeni Burovski
Evgeni Burovski
> I mean the functions I listed are currently exposed by the wrappers. I already said the rest looks fortran free. Not sure what I miscommunicated Superb, sorry for the...
> Our previous plan That plan was drafted before Ilhan joined the fun. `surfit` is done in this PR already, and it looks like `spgrid` is on Ilhan's chopping block....
> All that is currently missing are the list above Out of https://github.com/scipy/scipy/pull/24022#issuecomment-3584054524 I believe we have everything apart from `sproot` (which solves for zeros of a cubic parabola via...
Of `_fitpack_py`, `_fitpack_impl` and `_fitpack2` files: - `_fitpack2` implements `*UnivariateSpline` and `*BivariateSpline` families - `_fitpack_impl` implements the "functional" interface, `spl*` and `bispl*` functions. - `_fitpack_py` is a collections of trampolines...
> can't comment on the remaining class as I couldn't figure out what is importing which one yet which class is it?
`info = -4` from LAPACK's `dgeev` means that the [4th argument has an illegal value](https://www.netlib.org/lapack/explore-html/d4/d68/group__geev_ga7d8afe93d23c5862e238626905ee145e.html). The fourth argument is the matrix `a` itself. Printing the values shows it contains `nans`,...
`NdBSpline.derivative` is the recommended replacement for `RectBivariateSpline.partial_derivative`, esp after https://github.com/scipy/scipy/pull/23962 lands. @T90REAL how are you finding these corner cases, is this some usage or is it an automated tester or...
Okay, benchmarks seem to show that 1) batches of small matrices do benefit from looping in C, by a factor of ~4x. Shoehorning the banded solver into the same loop...
To balance convenience, performance and backwards compat (in main, `solve` includes "banded"), I'd do something like this: - `solve_banded(ab, ...)`: Here the input `ab` is `(..., ku + kl +...
> solve_banded is not batched yet. It is, https://github.com/scipy/scipy/blob/main/scipy/linalg/_basic.py#L831, since gh-22192. > and if it becomes that then kl, ku needs to be array if we allow for different band...