Module 16

Working with PyTorch

tensors, nn.Module, DataLoader, the training loop

When you finish this module you can say:

I can write a model, a dataset, and a training loop in PyTorch, and evaluate the model.

By the end of this module you'll be able to write Module 15's digit recognizer in PyTorch and train it on Colab.

Lessons

7 lessons · ≈ 69 min

1. Tensors: a numpy array + automatic differentiation9 min
2. Writing an nn.Module9 min
3. Dataset and DataLoader9 min
4. The training loop and optim10 min
5. Context managers and decorators: no_grad10 min
6. Using a pretrained model10 min
7. Project: a digit recognizer, with PyTorch12 min · project

Practice exercises

Short writing exercises, independent of the lessons: Go to the exercises

1. Loss and gradient
2. Your own predictor
3. A decorator that traces calls