18 releases

0.9.1 Apr 7, 2019
0.8.0 Jun 25, 2018
0.7.0 Mar 27, 2018
0.6.3 Dec 2, 2017
0.6.1 Nov 25, 2017

#7 in Machine learning

Download history 7/week @ 2019-01-22 10/week @ 2019-01-29 3/week @ 2019-02-05 5/week @ 2019-02-12 9/week @ 2019-02-19 42/week @ 2019-02-26 11/week @ 2019-03-05 7/week @ 2019-03-12 26/week @ 2019-03-19 87/week @ 2019-03-26 43/week @ 2019-04-02 19/week @ 2019-04-09 7/week @ 2019-04-16 32/week @ 2019-04-23 31/week @ 2019-04-30

89 downloads per month

Custom license

295KB
7.5K SLoC

autograd

Build Status

Provides differentiable operations and tensors.

Features

  • Lazy, side-effect-free tensors. autograd::Tensor<T> itself doesn't have its value basically. It realizes graphs that are immutable and eagerly executable at any timing, that is, it supports both run-by-define and define-by-run naturally in the context of neural networks.

  • Reverse-mode automatic differentiation. There are a lot of built-in operations that support higher-order derivatives, and you can define your own ops with ndarrays easily.

  • Pure Rust. The graph execution engine is implemented in pure Rust, so it's compilable to WebAssembly.

Installation

[dependencies]
autograd = { version = "0.9.1", features = ["mkl"] }

mkl feature is recommended to speedup gemm operations.

Examples

Here we are computing partial derivatives of z = 2x^2 + 3y + 1.

extern crate autograd as ag;

let ref x = ag::placeholder(&[]);
let ref y = ag::placeholder(&[]);
let ref z = 2.*x*x + 3.*y + 1.;

// dz/dy
let gy = &ag::grad(&[z], &[y])[0];
println!("{:?}", gy.eval(&[]));   // => Some(3.)

// dz/dx (requires to fill the placeholder `x`)
let gx = &ag::grad(&[z], &[x])[0];
println!("{:?}", gx.eval(&[(x, &ag::ndarray::arr0(2.).into_dyn())]));  // => Some(8.)

// ddz/dx (differentiates `z` again)
let ggx = &ag::grad(&[gx], &[x])[0];
println!("{:?}", ggx.eval(&[]));  // => Some(4.)

Another example: softmax regression for MNIST digits classification with Adam.

// This achieves 0.918 test accuracy after 3 epochs, 0.14 sec/epoch on 2.7GHz Intel Core i5


let ref w = ag::variable(ag::ndarray_ext::glorot_uniform::<f32>(&[28*28, 10]));
let ref b = ag::variable(ag::ndarray_ext::zeros::<f32>(&[1, 10]));
let ref x = ag::placeholder(&[-1, 28*28]);
let ref y = ag::placeholder(&[-1]);
let ref z = ag::matmul(x, w) + b;
let ref loss = ag::reduce_mean(&ag::sparse_softmax_cross_entropy(z, y), &[0, 1], false);
let ref params = [w, b];
let ref grads = ag::grad(&[loss], params);
let ref predictions = ag::argmax(z, -1, true);
let ref accuracy = ag::reduce_mean(&ag::equal(predictions, y), &[0], false);
let ref adam = ag::gradient_descent_ops::Adam::default();
let mut stateful_params = ag::gradient_descent_ops::Adam::vars_with_states(params);
let ref update_ops = adam.compute_updates(&stateful_params, grads);

// -- dataset --
let ((x_train, y_train), (x_test, y_test)) = dataset::load();

// -- training loop --
for epoch in 0..max_epoch {
    ...
    ag::eval(update_ops, &[(x, &x_batch), (y, &y_batch)]);
}

For more, see documentation or examples

Dependencies

~3.5MB
~55K SLoC