pytorch_autograd_basic_gradient_computation_with_backpropagation.py

1import torch
2
3x = torch.ones(5)  # input tensor
4y = torch.zeros(3)  # expected output
5w = torch.randn(5, 3, requires_grad=True)
6b = torch.randn(3, requires_grad=True)
7z = torch.matmul(x, w)+b
8loss = torch.nn.functional.binary_cross_entropy_with_logits(z, y)
9
10print(f"Gradient function for z = {z.grad_fn}")
11print(f"Gradient function for loss = {loss.grad_fn}")
12
13loss.backward()
14print(w.grad)
15print(b.grad)