tutorials
Notebooks + explainers, the way I wish they had been explained to me. Each tutorial has a runnable Colab/Binder link so you can try it without installing anything.
-
Machine Learning Training, from scratch
Train/val/test splits, loss curves, overfitting, regularisation, the things every ML course glosses over.
-
scikit-learn, the parts you'll actually use
Pipelines, cross-validation, and the five models that solve 80% of tabular problems.
-
Introduction to Neural Networks
Build a 2-layer network in NumPy, then the same thing in PyTorch. Backprop, by hand, once.
-
Convolutional Neural Networks
Convolutions, pooling, receptive fields. Build a CNN that classifies images, then read what it actually learned.
-
Physics-Informed Neural Networks (PINNs)
Putting a PDE into the loss function. The idea, the gotchas, and a working example for the heat equation.