Crate petite_ad 

petite AD

A pure Rust automatic differentiation library supporting both single-variable and multi-variable functions with reverse-mode differentiation (backpropagation).

Features

• Single-variable autodiff - Chain operations like sin, cos, exp
• Multi-variable autodiff - Build computational graphs for multiple inputs
• Zero-copy backward pass - Efficient gradient computation through closure chains
• Convenient macros - Use mono_ops![] and multi_ops![] for concise notation

Examples

Single-variable function

use petite_ad::{MonoAD, mono_ops};

let ops = mono_ops![sin, cos, exp];
let (value, grad_fn) = MonoAD::compute_grad(&ops, 2.0);
println!("f(2.0) = {}", value);
println!("f'(2.0) = {}", grad_fn(1.0));

Multi-variable function

use petite_ad::{MultiAD, multi_ops};

let exprs = multi_ops![
    (inp, 0),    // x₁
    (inp, 1),    // x₂
    (add, 0, 1), // x₁ + x₂
    (sin, 0),    // sin(x₁)
    (mul, 2, 3), // sin(x₁) * (x₁ + x₂)
];

let (value, grad_fn) = MultiAD::compute_grad(&exprs, &[0.6, 1.4]).unwrap();
let gradients = grad_fn(1.0);
println!("f(0.6, 1.4) = {}", value);
println!("∇f = {:?}", gradients);

Modules

traits

Traits for implementing custom differentiable functions.

types

Type definitions for autodiff results and gradient functions.

Macros

mono_ops

Macro to convert function names to MonoAD enum at compile time. This avoids the function pointer comparison issue across library boundaries.

multi_ops

Macro to build multi-variable computation graphs with lowercase operation names. Converts lowercase identifiers to MultiAD enum variants.

Structs

GraphBuilder

Builder for constructing multi-variable computation graphs.

Enums

AutodiffError

Errors that can occur during automatic differentiation computations.

MonoAD

Single-variable automatic differentiation operations.

MultiAD

Multi-variable automatic differentiation operations.

Type Aliases

Result

Result type for automatic differentiation operations.