Bit of fun with gradients


(Duane Nielsen) #1

Thought I’d share this code. Learned a few things about autograd doing this.

import torch

# unbroken gradient, backward goes all the way to x
x = torch.ones(2, 2, requires_grad=True)
y = 2 * x + 2
z = y * y * 3
out = z.mean()
out.backward()
print(x.grad)
baseline_x = x.grad

# broken gradient, ends at _y
x = torch.ones(2, 2, requires_grad=True)
y = 2 * x + 2

_y = torch.tensor(y.detach(), requires_grad=True)
z = _y * _y * 3
out = z.mean()
out.backward()
print(x.grad)
print(_y.grad)

# we can however, take the grad of _y and put it in manually!
y.backward(_y.grad)
print(x.grad)
x_from_manipulated_graph = x.grad

assert torch.eq(x_from_manipulated_graph, baseline_x).all()

The first computation is standard autograd.

The second computation shows how to break the computation graph apart, pull a tensor out of the middle, and manually put the gradient back in.

Very cool that pytorch allows you to do this kind of stuff.

What was interesting to me is that I thought all the intermediate variables in the computation graph hold gradient values… but actually it seems that this is not the case. It looks like grads are only kept for terminals.


#2

x_from_manipulated_graph and baseline_x??


(Duane Nielsen) #3

Thanks, missed out a couple lines. The last line should make a bit more sense now.