Classical machine learning

From regression to boosted trees, and the overfitting ideas that follow you to the end.

Framing the problemClassification, regression or clustering, and what you are measuring before you write a line.
Linear and logistic regressionWritten by hand once so you understand it, then in three lines of scikit-learn.
Splitting the dataTrain, validation and test, and why leakage destroys every number you report after it.
MetricsAccuracy lies on imbalanced data; precision, recall and the confusion matrix do not.
Trees, forests and boostingA decision tree, a random forest, then XGBoost, and why they beat networks on tables.
Feature engineeringEncoding and scaling, and when a new feature earns its keep instead of adding noise.
Clustering and dimensionalityk-means and PCA, and when clustering is a discovery instead of a pretty illusion.
Project: imbalanced classificationA naive baseline, a model that beats it, and a report that argues with numbers.