1. Kernel Regression Playground
Compare 1-NN, k-NN, and kernel regression on a smooth signal, an abrupt step, and a sparse region with an outlier. Discuss locality, smoothing bias, and sensitivity.
Open activity2. Distance Metrics Lab
Use an abstract feature space, a restaurant example from the slides, and a feature-unit trap. Change L1/L2/L∞ and feature weights, then watch the neighbor ranking change.
Open activity3. Curse of Dimensionality
Jump among 2-D, 10-D, 30-D, and 100-D regimes. Explore neighborhood volume, sample demand for k local neighbors, and concentration of distances.
Open activity4. Gradient Descent Lab
Fit lines for well-spread data, a flat-direction case, and an outlier case. Compare batch and stochastic updates, learning rates, and the closed-form optimum.
Open activity5. Generalization and Regularization
Try sine, cubic, and linear targets. Vary polynomial degree, training-set size, and regularization while watching train/test RMSE and coefficient growth.
Open activity