Any hope for sciml integration?

[jlperla]

The sciml metapackages give a lot of comfort to people because of promised maintenance, consistency of interface for using auto-diff, and CI testing resources.

However, they would be the first to say that interpolation packages are incomplete. The DataInterpolations.jl you benchmark against is only 1D, and there has only recently been a move towards https://github.com/SciML/DataInterpolationsND.jl which will support ND interpolation but also kernelabstractions for GPUs, etc.

Any chance you could join forces? If not, any chance that this could be a good reason to spearhead a sciml metapackage for interpolation problems (along the same lines as they have for optimization.jl, etc.)

[mgyoo86]

Thanks a lot for the interest — I really appreciate it.
We think highly of the SciML ecosystem and have benefited a lot from the tools and community around it.

First of all, I should clarify that I don’t represent the ownership of this repository. I did most of the development work, but this isn’t my personal repo — it belongs to General Atomics’ GitHub organization, ProjectTorreyPines.

I’m also not familiar enough with how the SciML organization is run (or the history behind particular integration paths like the Optimization.jl example you mentioned) to say what would or wouldn’t be possible. But I can share some background from our side.

General Atomics is a private company that works on nuclear fusion research, and ProjectTorreyPines is one of the company’s GitHub organizations focused on Julia packages. In particular, we’re developing FUSE.jl as an integrated modeling tool for nuclear fusion, supported by our own ecosystem of 50+ Julia packages. Most of these repos are public under the Apache 2.0 license, but many are domain-specific, so we maintain a separate FuseRegistry. A smaller subset — including this package — is registered in the General registry to make it easier for broader users to discover and use.

Given that context, my personal guess is that an org-level integration would be difficult in the near term (again: I’m not representing any ownership/policy). We really love Julia, actively promote Julia adoption in the fusion community, and we’ll keep supporting the Fuse ecosystem — including this package. This package is also still early-stage and evolving quickly, so keeping development more centralized for now helps us move fast and keep the API consistent.

That said, we’d absolutely welcome tighter collaboration and stronger coupling with SciML packages where it makes sense.
If you have specific integration points in mind (interfaces, shared abstractions, or any background/materials on how the SciML organization works), I’m happy to discuss and help where I can.

[jlperla]

I understand completely. @ChrisRackauckas were you thinking of having an interpolations meta-package at some point, or that it would include https://github.com/SciML/DataInterpolationsND.jl and https://github.com/SciML/DataInterpolations.jl as top-level projects without a meta package wrapper?

If you were planning a meta-package, then it seems like an overhead-free wrapper for this one might be possible at some point?