mgrad 0.1.0

A minimal automatic differentiation library. Just for fun...

All code are in a single file (src/mgrad.rs) to easily copy-paste into other projects.
Please refer to examples/a.rs for usage:

use mgrad::nn;

fn main() {
    // y = ln(x^2 * (sin(x) + 1))
    let x = nn::variable(1);
    let y = x.sin() + nn::constant(1);
    let y = x.pow(2) * y;
    let y = y.ln();
    y.backward(1);

    // dy/dx should be ~ 2.29341
    println!("The gradient of y=ln(x^2 * (sin(x) + 1)) at x=1 is: {:?}", x.grad);
}

Note that backward() is called with immutable declared variables to update their gradients.
This is because some unsafe code is used to avoid RefCell overhead.
Theoretically this is unsafe, but as long as we are not doing parallel computations on the same graph, it should behave correctly.

Run with cargo doc --open to see the documentation.