67 lines
1.6 KiB
Plaintext
67 lines
1.6 KiB
Plaintext
|
This rule raises when a `torch.autograd.Variable` is instantiated.
|
||
|
|
|
||
|
|
== Why is this an issue?
|
||
|
|
|
||
|
|
The Pytorch Variable API has been deprecated. The behavior of Variables is now provided by the Pytorch tensors and can be controlled with the `requires_grad` parameter.
|
||
|
|
|
||
|
|
The Variable API now returns tensors anyway, so there should not be any breaking changes.
|
||
|
|
|
||
|
|
== How to fix it
|
||
|
|
|
||
|
|
Replace the call to `torch.autograd.Variable` with a call to `torch.tensor` and set the `requires_grad` attribute to `True` if needed.
|
||
|
|
|
||
|
|
=== Code examples
|
||
|
|
|
||
|
|
==== Noncompliant code example
|
||
|
|
|
||
|
|
[source,python,diff-id=1,diff-type=noncompliant]
|
||
|
|
----
|
||
|
|
import torch
|
||
|
|
|
||
|
|
x = torch.autograd.Variable(torch.tensor([1.0]), requires_grad=True) # Noncompliant
|
||
|
|
x2 = torch.autograd.Variable(torch.tensor([1.0])) # Noncompliant
|
||
|
|
----
|
||
|
|
|
||
|
|
==== Compliant solution
|
||
|
|
|
||
|
|
[source,python,diff-id=1,diff-type=compliant]
|
||
|
|
----
|
||
|
|
import torch
|
||
|
|
|
||
|
|
x = torch.tensor([1.0], requires_grad=True)
|
||
|
|
x2 = torch.tensor([1.0])
|
||
|
|
----
|
||
|
|
|
||
|
|
|
||
|
|
== Resources
|
||
|
|
=== Documentation
|
||
|
|
|
||
|
|
* Pytorch documentation - https://pytorch.org/docs/stable/autograd.html#variable-deprecated[Variable API]
|
||
|
|
|
||
|
|
|
||
|
|
ifdef::env-github,rspecator-view[]
|
||
|
|
|
||
|
|
(visible only on this page)
|
||
|
|
|
||
|
|
== Implementation specification
|
||
|
|
|
||
|
|
Should be pretty straighforward to implement.
|
||
|
|
|
||
|
|
=== Message
|
||
|
|
|
||
|
|
Primary : Replace this call with a call to "torch.tensor".
|
||
|
|
|
||
|
|
|
||
|
|
=== Issue location
|
||
|
|
|
||
|
|
Primary : Name of the function call
|
||
|
|
|
||
|
|
=== Quickfix
|
||
|
|
|
||
|
|
Might be tricky to know how to call the `torch.tensor` function.
|
||
|
|
If there is an import like `from torch import tensor`, then replace with `tensor(...)`
|
||
|
|
If not, then replace with `torch.tensor(...)`
|
||
|
|
|
||
|
|
|
||
|
|
endif::env-github,rspecator-view[]
|