Linear regression derived, implemented and diagnosed — the template every supervised model follows.
10h · 8 lessons · 1 challenge
Before this: Calculus & Optimization, NumPy & Vectorization
Linear regression is the smallest complete example of supervised learning: a hypothesis, a loss, a gradient, an update, a diagnosis. Learn it properly and every model after it is a variation on parts you already know.
You derive the normal equation, implement gradient descent against it, and then spend as much time on residual analysis as on fitting -- because reading a model is the part interviews probe.
A residual plot appears; name the defect — heteroscedasticity, missing interaction, outlier leverage, wrong link. Twelve rounds, increasing subtlety.
Worth 350 XP.
