AttributeError: 'float' object has no attribute 'detach'

1

I am trying to do this :

my_file3 = x_mean[1].detach().cpu().numpy()

and it shows me this error :

AttributeError: ‘float’ object has no attribute ‘detach’

2

Is the variable part of a graph? If it is not part of a graph or does not have a gradient you won’t be able to detach it from anything because it wasn’t attached in the first place.

3

Adding on to what @Dwight_Foster said, you can do my_file3 = (x_mean[1].detach() if x_mean[1].requires_grad else x_mean[1]).cpu().numpy() to make it work in any case.

4

Hi Himanshu,
I tried what you have suggested but it is showing me this now
AttributeError: ‘float’ object has no attribute ‘requires_grad’

5

Are you sure that x_mean[1] is a torch.Tensor object?

Even if you take a member of a tensor that is a list, it will still have .requires_grad as False by default since it is of the torch.Tensor class:

>>> import torch
>>> x_mean = torch.ones((50))
>>> x_mean.requires_grad
False
>>> x_mean[1].requires_grad
False
>>> type(x_mean[1])
<class 'torch.Tensor'>

Maybe you are appending the .item() of the original tensor somewhere, which changes the class to a normal Python float:

>>> x_mean=[1.,2.,3.,4.,5.]
>>> type(x_mean[1])
<class 'float'>
>>> x_mean[1].requires_grad
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'float' object has no attribute 'requires_grad'