4.5 Multilayer Neural Networks for Regression (Parts 1-2)

Code

• Part 2: Neural networks for regression in PyTorch, 4.5-mlp-regression-part2.ipynb

What we covered in this video lecture

So far, we have talked about classification a lot — the most common use case for deep neural networks. However, we can use deep neural networks for regression, too! In this lecture, we saw that we can turn a multilayer perceptron classifier into a regression model with only two minor changes:

Removing the logistic sigmoid / softmax activation function from the output layer; swapping the cross entropy loss with the mean squared error loss.

Additional resources if you want to learn more

If you plan to implement neural networks for regression, also consider using the R-squared coefficient as an evaluation metric. The R2 score is available in TorchMetrics, a library that we will cover in the next unit (unit 5.3).

a = torch.tensor([1.23, 0.51, -2.5])
b = torch.tensor([-1.0, 0.55, -1.2])
torch.nn.functional.mse_loss(a, b)

Watch Video 1 Mark complete and go to Unit 4.6 →