Summary - nn.Module

In this group you learned nn.Module, __init__, forward, and nn.Linear.

Summary: nn.Module

In this group:

Every model is an nn.Module; layers in __init__, computation in forward; called as m(x).
super().__init__() is required; nn.ModuleList for lists of layers (a plain list hides the parameters).
nn.Sequential for simple chains; torch.nn.functional for parameter-free operations.
nn.Linear: weight (out, in), y = xWᵀ + b; initialisation with nn.init.
parameters(), named_parameters(), freezing and state_dict.

Exercise Your own classifier

Write the class Siniflandirici(girdi, gizli, sinif): two nn.Linear in __init__, a ReLU between them in forward. For Siniflandirici(4, 16, 3), print the output shape for the input torch.zeros(10, 4) and the parameter count on one line.

Show example solution

import torch
from torch import nn

class Siniflandirici(nn.Module):
    def __init__(self, girdi, gizli, sinif):
        super().__init__()
        self.l1 = nn.Linear(girdi, gizli)
        self.l2 = nn.Linear(gizli, sinif)

    def forward(self, x):
        return self.l2(torch.relu(self.l1(x)))

m = Siniflandirici(4, 16, 3)
print(m(torch.zeros(10, 4)).shape, sum(p.numel() for p in m.parameters()))

Test yourself

What is the recommended way to save a model?
- print(model)
- Printing the model.parameters() list
- The whole Python session with pickle
- torch.save(model.state_dict(), path)
Show answer

torch.save(model.state_dict(), path) — A state_dict contains only tensors; the model class is rebuilt in code and loaded.

Load the saved weights into the model.

model.____(torch.load("model.pt"))

Show answer

model.load_state_dict(torch.load("model.pt"))

Match the tools.

nn.Module

Base class
nn.Sequential

Classless chain
nn.ModuleList

Registered list
state_dict

Weight dict

Next lesson: Training loop

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies