TypeError: unsupported operand type(s) for *: 'Tensor' and 'NoneType'

autograd
1

Hello, can some figure out this error?
thanks

TypeError Traceback (most recent call last)
in

—> 44 qx = k * qx
45 Div = Divergence( x_trg , qx )
46 ########################################################

TypeError: unsupported operand type(s) for *: ‘Tensor’ and ‘NoneType’

2

for qx is defined as: qx = Gradient( x_trg , T )

3

Based on the error message it seems that qx is None. Could you check, if Gradient is returning a value or if the return statement is either not used or is returning None?

1 Like
4

Yeak I chesked, gradient returns None, and I don’t know how to make it returning value.
gradient function is defined with:
torch.autograd.grad (outputs , inputs , grad_outputs=None , retain_graph=None , create_graph=False , only_inputs=True ):

5

torch.autograd.grad should yield the desired gradients, so maybe you are missing to assign them to a return value?

1 Like
6

thanks for replying, here is gradient function:

image
image708×166 6.26 KB

7

I’m not sure why a None value is returned. Could you post a minimal code snippet to reproduce this issue, i.e. valid gradients are expected while you are seeing Nones?

8

Actually the model is composed of several files ( it’s physics-informed neural network), I send you in message the portion of the code that precedes the calling of gradient.

9

Could you please post the code here, as you shouldn’t share private code with anyone who is not collaborating on your project?

10

yeah, I can’t share it here, it’s private code :confused:

11

That’s understandable. Could you come up with a minimal “fake” example, which would reproduce the issue you are seeing, but wouldn’t show anything from your architecture etc.?

1 Like