3h 12m total
Build real predictive models from the ground up. You'll start with the ML mindset and workflow — what learning from data actually means, and when NOT to reach for ML — then wrangle and explore data with pandas and NumPy. From there you train your own supervised models (linear and logistic regression, gradient descent, k-nearest neighbours, decision trees), evaluate them honestly (train/validation/test, cross-validation, precision/recall/F1, ROC-AUC, the bias-variance tradeoff, overfitting, regularization), engineer features and assemble leak-free scikit-learn Pipelines, then move to unsupervised learning (k-means, PCA) and finally neural networks with PyTorch — finishing with an end-to-end capstone that saves and serves a trained model. This is classical, foundational machine learning: you train the models yourself. It assumes Python basics (see the Python course) and teaches the ML on top.
