Gradient of an arbitrary function

autograd
1

Hi all, I am a new Pytorch user. I am working on a network structure which does not use the input value directly but first will calculate a function of the input value.

I am doing the simplest case when I give the input to a function and ask PyTorch to calculate the derivative of the function wrt the input. Here is the code,

Blockquote

def get_torch_G2(Rij, Rs, eta):
return Variable(0.5 * torch.exp(-eta * (Rij - Rs).pow(2)) * (torch.cos(np.pi * Rij / 12)+1), requires_grad=True)

dd = torch.tensor([2.2], requires_grad=True)
bb = get_torch_G2(dd, 0, 1)
bb.backward()
dd.grad.clone()

Blockquote

It gives me the error “‘NoneType’ object has no attribute ‘clone’”. Can anyone help me with my problem please?

2

Jus remove Variable and it will work

import torch
def get_torch_G2(Rij, Rs, eta):
    return 0.5 * torch.exp(-eta * (Rij - Rs).pow(2)) * (torch.cos(3.1416 * Rij / 12)+1)
dd = torch.tensor([2.2], requires_grad=True)
bb = get_torch_G2(dd, 0, 1)
bb.backward()
dd.grad.clone()
1 Like