__init__ for layers in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the purpose of the __init__ method in a PyTorch layer class?

The __init__ method sets up the layer by defining its parameters and sub-layers. It prepares the layer to be used later in the forward pass.

beginner

Why do we call super().__init__() inside the __init__ method of a PyTorch layer?

Calling super().__init__() initializes the parent class nn.Module, which is necessary for PyTorch to track parameters and manage the layer properly.

beginner

How do you define a fully connected layer inside the __init__ method?

You create an instance of nn.Linear with input and output sizes, for example: self.fc = nn.Linear(input_size, output_size).

intermediate

What happens if you forget to call super().__init__() in your layer's __init__?

PyTorch will not properly register the layer's parameters, which can cause errors during training or the model not learning correctly.

beginner

Can you store multiple layers inside the __init__ method? How?

Yes, you can define multiple layers as attributes, for example: self.conv1 = nn.Conv2d(...) and self.fc1 = nn.Linear(...). This organizes the model parts clearly.

What is the first thing you should do inside the __init__ method of a PyTorch layer?

A Call <code>super().__init__()</code>

B Define the forward method

C Initialize the optimizer

D Load the dataset

Calling super().__init__() initializes the base nn.Module class, which is essential for the layer to work properly.

How do you define a linear layer with 10 inputs and 5 outputs inside __init__?

A self.fc = nn.Linear(5, 10)

B self.fc = nn.Linear(10, 5)

C self.fc = nn.Conv2d(10, 5)

D self.fc = nn.Linear()

The first argument is input size (10), the second is output size (5) for nn.Linear.

What happens if you do not call super().__init__() in your layer?

A Parameters are not registered properly

B The model trains faster

C The forward method runs twice

D Nothing happens

Without calling super().__init__(), PyTorch cannot track parameters, causing training issues.

Where do you define the layers like convolution or linear in a PyTorch model?

A Outside the class

B Inside the <code>forward</code> method

C Inside the <code>__init__</code> method

D In the training loop

Layers are defined in __init__ so they are created once and reused during forward passes.

Can you define multiple layers inside __init__?

A Only if you use a list

B No, only one layer per class

C Only if they are the same type

D Yes, by assigning them as attributes

You can define many layers as attributes like self.conv1, self.fc1, etc.

Explain the role of the __init__ method when creating a custom PyTorch layer.

Think about what happens before the model processes data.

Describe what could go wrong if you forget to call super().__init__() in your PyTorch layer's __init__.

Consider how PyTorch manages model parts internally.