General ML

These chapters introduce the methods used later in the book on standard teaching datasets, where the answer is known and the focus can stay on the method. They build on one another, but each notebook runs on its own.

Foundations

• Software Toolkit: the Python packages used in the examples and what each one does.

• Data Splits and Leakage: training, validation, and test sets, and how information leaks between them.

• Model Evaluation: baselines, confusion matrices, and the metrics used throughout the book.

• Neural Networks: a dense network for handwritten digits, with every step of the training loop explained.

• Convolutional Neural Networks: the same task with a network that exploits the image’s spatial structure.

Further methods

• CIFAR-10 CNN Progression: a harder image task, a deeper network, and overfitting.

• Tree Models and Ensembles: gradient-boosted trees on flattened images.

• Transfer Learning: reusing an ImageNet network when labelled data are scarce.

• Hyperparameter Tuning: choosing architecture and learning rate on validation data.

• Generative Models: a generative adversarial network that produces new digits.

Each neural-network chapter has a PyTorch notebook, which writes the training loop out explicitly, and a Keras 3 notebook, which runs the same model through the shorter compile() and fit() interface on the same PyTorch backend. Each notebook ends with suggestions for changes to try.