7a9bb0afce
* Create rule S6978 * First version of the rule * Make it a bug * Add example & typo * Small correction --------- Co-authored-by: ghislainpiot <ghislainpiot@users.noreply.github.com> Co-authored-by: Ghislain Piot <ghislain.piot@sonarsource.com>
83 lines
2.6 KiB
Plaintext
83 lines
2.6 KiB
Plaintext
This rule raises an issue when a class is a Pytorch module and does not call the `++super().__init__()++` method in its constructor.
|
|
|
|
== Why is this an issue?
|
|
|
|
To provide the AutoGrad functionality, the Pytorch library needs to set up the necessary data structures in the base class. If the `++super().__init__()++` method is not called, the module will not be able to keep track of its parameters and other attributes.
|
|
|
|
For example, when trying to instantiate a module like `nn.Linear` without calling the `++super().__init__()++` method, the instantiation will fail when it tries to register it as a submodule of the parent module.
|
|
|
|
[source, python]
|
|
----
|
|
import torch.nn as nn
|
|
|
|
class MyCustomModule(nn.Module):
|
|
def __init__(self, input_size, output_size):
|
|
self.fc = nn.Linear(input_size, output_size)
|
|
|
|
model = MyCustomModule(10, 5) # AttributeError: cannot assign module before Module.__init__() call
|
|
----
|
|
|
|
|
|
== How to fix it
|
|
Add a call to `++super().__init__()++` at the beginning of the constructor of the class.
|
|
|
|
=== Code examples
|
|
|
|
==== Noncompliant code example
|
|
|
|
[source,python,diff-id=1,diff-type=noncompliant]
|
|
----
|
|
import torch.nn as nn
|
|
|
|
class MyCustomModule(nn.Module):
|
|
def __init__(self, input_size, output_size):
|
|
self.fc = nn.Linear(input_size, output_size) # Noncompliant: creating an nn.Linear without calling super().__init__()
|
|
----
|
|
|
|
==== Compliant solution
|
|
|
|
[source,python,diff-id=1,diff-type=compliant]
|
|
----
|
|
import torch.nn as nn
|
|
|
|
class MyCustomModule(nn.Module):
|
|
def __init__(self, input_size, output_size):
|
|
super().__init__()
|
|
self.fc = nn.Linear(input_size, output_size)
|
|
----
|
|
|
|
== Resources
|
|
=== Documentation
|
|
|
|
* Pytorch documentation - https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module[torch.nn.Module]
|
|
|
|
|
|
ifdef::env-github,rspecator-view[]
|
|
|
|
(visible only on this page)
|
|
|
|
== Implementation specification
|
|
|
|
Verify that subclasses of `torch.nn.Module` call the `++super().__init__()++` method in their constructor.
|
|
It also applies to classes that inherit from a class that inherits from `torch.nn.Module`.
|
|
|
|
=== Message
|
|
|
|
Primary : Add a call to `++super().__init__()++`
|
|
|
|
Secondary : Inheritance happens here
|
|
|
|
=== Issue location
|
|
|
|
Primary : First line of the constructor
|
|
|
|
Secondary : (The inheritance position of the class)
|
|
|
|
=== Quickfix
|
|
|
|
Add a call to `++super().__init__()++` at the beginning of the constructor.
|
|
If direct child of `torch.nn.Module`, easy, no need to try to add parameters.
|
|
If indirect, try to find the parameters and match them with the same name ? Or abort if too complicatted.
|
|
|
|
endif::env-github,rspecator-view[]
|