What Is Machine Learning · L2 Mechanics

The train/validation/test split, the bias-variance tradeoff, and how generalization is actually measured.

Training set (~80%)

used to fit model parameters

Validation set (~10%)

used to tune hyperparameters and detect overfitting

Test set (~10%)

touched once, at the very end, for final reporting

Splitting data is the single most important discipline in applied ML.

The entire point of ML is generalization — performing well on data the model has never seen. A model that just memorizes its training set is useless; that's why we never evaluate a model on the data it was trained on.