Sunday, March 10, 2019

Rust for linear algebra and neural networks

From time to time I do a little bit of fooling around with algorithms and AI, I suppose it's a break from what I do in work, which is more about building business applications. I don't have a real goal in mind, I don't think I'm going to stumble about an algorithm for general intelligence or something, but I like to understand how things work at a low level.

So I've writing a bit of Rust code to implement some linear algebra (some matrix and vector operations) and use that to build some small neural networks. You can find the current state of that code here. Of course, if you really want to use production ready code for linear algebra you should check and if neural networks and AI in Rust are what you're after please check Note that rusty-machine is still in a state of flux since it's using its own linear algebra library but there's talk about using nalgebra from rustsim hopefully that will bring together all these initiatives in one unified Rust platform for numeric operations and AI.

I have to say I'm still impressed with Rust. I like that you can do a lot of things with a functional approach, while still being able to update variable values or do other imperative operations when it makes sense. It does feel a lot like "pragmatic Haskell" at times. So far I haven't struggled with the borrow checker, but the code I've written so far is fairly simple and self-contained.

As always, I'm still wondering if there are some projects that it would make sense for me to contribute too. I'm certainly not a AI or math expert, but if you have a Rust project that could do with some contributions, let me know!

Happy Rust Hacking!

No comments: