Native Differential Programming Support for Rust

[mcy]

Yes, using Mul for something like the Hadamard product (or the Kronecker product!) is really silly, but I think bikeshedding the surface syntax of something built on the intrinsic in question feels very irrelevant. We should probably grow the relevant attr or core::arch item or whatever before getting bogged down in non-blocking details.

[toc]

I don’t suppose Rust has ever considered adding e.g. the @ multiplication operator from python. I’ve seen haskell-esque infix backtick operators (a `mod` b) discussed.

[Tom-Phinney]

I’m actually an appropriate person to ask this of, since in my early days of Rust I explored a similar question. Read the reference, then answer the question: What percentage of the alphanumeric multiply operator method calls in a large corpus of code would be replaced by use of a glyph operator?

I’ll let others respond about Rust’s history of use of the @ sigil, and point out where in Rust’s grammar it still occurs.

[Centril]

@Tom-Phinney

I’ll let others respond about Rust’s history of use of the @ sigil, and point out where in Rust’s grammar it still occurs.

Pattern matching, e.g. let x@[_, _] = [1, 2]. This is a special case of the "intersection pattern" in F#.

[Tom-Phinney]

@Tom-Phinney

I’ll let others respond about Rust’s history of use of the @ sigil

Guess I'll do it myself. Per the Wikipedia entry about Rust,

Starting in Rust 0.9 and ending in Rust 0.11, Rust had two built-in pointer types, ~ and @ , simplifying the core memory model. It reimplemented those pointer types in the standard library as Box and (the now removed) Gc .

[Schultzer]

Hi @karanchahal,

There is few project already out there, just to name a few https://crates.io/crates/autograd

potan/dual.rs

novacrazy/dual_num

I would properly argue that this should start by living in a crate. the only reason I could see this move into Rust std/core is if there is going to some llvm intrinsic.

There is still a lot of research going on so it’s properly for the best to be in a crate till things have settle.

[nestordemeure]

There is still a lot of research going on so it’s properly for the best to be in a crate till things have settle.

Exactly! Automatic differentiation is still a research subject and has only very targeted applications : it is the opposite of something we would want to freeze in a language's core implementation.

Plus we are lucky to have a macro system that should be powerful enough to emulate what other languages do with native support.

[toc]

@Tom-Phinney

What percentage of the alphanumeric multiply operator method calls in a large corpus of code would be replaced by use of a glyph operator?

Counting notebook code... enough that python adding the @ was quite nice. Not counting notebooks very very little. Admittedly I don't really use a notebook workflow for Rust.