Model definition

Define a model with nn.Module.

Model definition

Our model has two parameters: the slope w and the intercept b. Both are created with requires_grad=True so that autograd computes their gradients. The model is an ordinary Python function:

w = torch.zeros(1, requires_grad=True)
b = torch.zeros(1, requires_grad=True)

def model(x):
    return x * w + b      # (10, 1) * (1,) + (1,) -> (10, 1)

Initially the parameters are zero; the model predicts 0 for every input.

The parameter list

Collecting the parameters to train in a list lets the update and gradient-zeroing steps be written with a single loop. nn.Module keeps this list for us with parameters():

parametreler = [w, b]
for p in parametreler:
    print(p.shape, p.requires_grad)

Writing the model as a function is instructive in small examples; in large models nn.Module manages the layers and parameters.

Exercise Try the model

While the parameters are zero, print the sum of model(x) with item(). Then, inside torch.no_grad(), set w to 3 and b to 2 (w.fill_(3), b.fill_(2)) and print with torch.equal that the prediction equals y.

Show example solution

import torch

x = torch.arange(10, dtype=torch.float32).unsqueeze(1)
y = 3 * x + 2
w = torch.zeros(1, requires_grad=True)
b = torch.zeros(1, requires_grad=True)

def model(x):
    return x * w + b

print(model(x).sum().item())
with torch.no_grad():
    w.fill_(3)
    b.fill_(2)
print(torch.equal(model(x).detach(), y))

Test yourself

What does <code>model(x)</code> give while the parameters are zero?
- 0 for every example
- x itself
- y
- An error
Show answer

0 for every example — x · 0 + 0 = 0.

Compute the model's prediction.

def model(x):
    return x * w ____ b

Show answer

def model(x):
    return x * w + b

Match the pieces.

w

Slope
b

Intercept
requires_grad=True

To be learned
model(x)

Prediction

Next lesson: Loss computation

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies