mini-ode 0.1.5

A minimalistic ODE solvers library built on top of PyTorch
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

use crate::optimizers::gradient_and_hessian;
use tch::{IndexOp, Tensor};

#[test]
fn test_gradient_and_hessian_case1() {
    let x = Tensor::from_slice(&[1.0f32, 2.0f32]);
    let (grad, hessian) = gradient_and_hessian(&|y| y.i(0) + y.i(1), &x).unwrap();
    let expected_grad = Tensor::from_slice(&[1f32, 1f32]);
    let expected_hessian = Tensor::from_slice(&[0f32, 0f32, 0f32, 0f32]).reshape([2, 2]);

    assert_eq!(grad, expected_grad);
    assert_eq!(hessian, expected_hessian);
}

#[test]
fn test_gradient_and_hessian_case2() {
    let x = Tensor::from_slice(&[5.0f32, 1.0f32]);
    let (grad, hessian) = gradient_and_hessian(&|y| y.i(0) * y.i(0) + y.i(0) * y.i(1), &x).unwrap();
    let expected_grad = Tensor::from_slice(&[11f32, 5f32]);
    let expected_hessian = Tensor::from_slice(&[2f32, 1f32, 1f32, 0f32]).reshape([2, 2]);

    assert_eq!(grad, expected_grad);
    assert_eq!(hessian, expected_hessian);
}