Linear Algebra Library

Looking forward to the Differentiable protocol, I thought I'd give it a try and implement a generic Kalman Filter. There is some similar effort for Swift Fusion at GeorgiaTech, but they seem to implement some own Linear Algebra Library.

Also S4TF has their Tensor types that, as I understand, can do Linear Algebra.

And then there is the recent post in this forum.

My question now: What should I choose? Ideally, I would make it generic enough so that any LinAlg implementation can be used, but I am not sure if there are any Protocols that are describing Matrices etc (Like in Swift-Numerics).

Or is there any candidate for a de-facto standard library? Like Eigen in C++?

Thanks for any opinions here!

1 Like

I don't think there is a standard yet for a linear algebra library, but as far as I know my LANumerics library is the only one that fully supports Swift-Numerics including complex numbers.

When developing the library I basically just needed a flexible implementation of singular value decomposition, and Accelerate had that, so I just built my library around it. But apart from that specific use case, I would also like to be able to run on Metal, and make that as transparent as possible for the end user.

You could try basing your work on LANumerics, and if something is not general enough for your purposes, we could fix that.

Terms of Service

Privacy Policy

Cookie Policy