Tensor deployment with autograd in pure rust.
short story long an example of python code
import torch
x = torch.tensor([1., 2., 3., 4., 5., 6.]).reshape(6, 1)
y = torch.tensor([3., 5., 7., 9., 11., 13.]).reshape(6, 1)
w = torch.tensor([-0.1], requires_grad=True)
b = torch.tensor([0.], requires_grad=True)
epochs = 1000
lr = 0.01
for _ in range(epochs):
# forward propagation
y_hat = x@w + b
# loss calculation
loss = 0.5 * ((y_hat - y)**2).mean()
# back propagation
w.grad.zero_()
b.grad.zero_()
loss.sum().backward()
# optimization
dw = w.grad
db = b.grad
with torch.no_grad():
w = w - lr * dw
b = b - lr * db
print(f"w = {w}\nb = {b}")equivalent rust code
use perceptron::prelude::*;
fn main() {
let x = tensor![1., 2., 3., 4., 5., 6.].reshape((6, 1));
let y = tensor![3., 5., 7., 9., 11., 13.].reshape((6, 1));
let mut w = rand_tensor![1, 1].requires_grad(true);
let mut b = rand_tensor![1].requires_grad(true);
let epochs = 1000;
let lr = 0.01;
for _ in 0..epochs {
// forward propagation
let y_hat = x.dot(&w) + &b;
// loss calculation
let loss = 0.5 * (&y_hat - &y).pow(2.0).mean();
// back propagation
loss.backward();
// optimization
no_grad!({
w = &w - lr * w.grad().unwrap();
b = &b - lr * b.grad().unwrap();
});
w.zero_grad();
b.zero_grad();
}
println!("w = {w}\n\nb = {b}");
}This code is working in Rust the same as Python. There are many additional features that can be tested.
In conclusion, please use Display instead of Debug if you can. You can see that there is no difference between Python and Rust.
This project may not be completed due to the lack of interest from GitHub's Rust and ML developers. Feel free to leave a comment if you are interested.