Interactive Companions

Drag the input point along the fitted line and watch the predicted score update — the exact hours/score example and numbers from class, live.

Drag a vector, watch the angle and the similarity score move together — the same idea behind the movie-recommender assignment, small enough to see.

Drag the new customer and vote by hand, paint your own data to see decision regions, and see why training accuracy flatters k = 1 — with the lecture's six customers and its ties.

Draw a test set from the housing districts a thousand times, at random and stratified by income — watch the random income mix wobble and the stratified one hold still, for 20,640 districts or only 200.

Drag points and watch every term of the correlation formula as a shaded rectangle — starting from the dogs-and-cats data from class, with the cosine view and the cases where r misleads.

Click a housing district and watch its nearest neighbors jump when you scale, drag one typo through min-max scaling and standardization, and see why scaling changes the units but a log changes the shape.

Take six scikit-learn classes apart — their roles, hyperparameters and learned parameters, and the errors when you call the wrong method — then chain them into a pipeline and watch what each step learns from eleven housing districts.