Using PyTorch for computing derivative of a function

Question

When I execute the code below:

import torch 

def g(x):
    return 4*x + 3
x=3.0
g_hat=torch.tensor(g(x), requires_grad= True)
g_hat.backward()

I get the following output:

>>>g_hat.grad
tensor(1.)

But this is not the result that I was expecting to get from my code above... what I want to do is, I want to find the value of dg/dx, at x = 3.0 (so in the example above, the correct output should be tensor(4.)). How can I achieve this with PyTorch? or if I can't carry out this task with PyTorch, how can I do this type of task easily on Python?

Thank you,

Igor Rivin · Accepted Answer · 2020-06-23 04:16:44Z

1

x = torch.autograd.Variable(torch.Tensor([1.0]),requires_grad=True)
y = 4 * x + 3
y.backward()
x.grad

Works fine.

answered Jun 23, 2020 at 4:16

Igor Rivin

4,8742 gold badges31 silver badges45 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

jodag Over a year ago

The autograd.Variable API has been deprecated since version 0.4.0. To improve clarity I recommend x = torch.Tensor([1.0], requires_grad=True) instead.

chico0913 Over a year ago

Hello, thank you for your answer. I have a follow up question. can PyTorch do vector calculus? for example, computing a second derivative with respect to a vector? If yes, how can I compute a second derivative of a function with respect to a vector on Pytorch?

chico0913 Over a year ago

@IgorRivin thank you so much! This is very helpful! :)

Collectives™ on Stack Overflow

Using PyTorch for computing derivative of a function

1 Answer 1

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related