The Explorers
Every big idea in this course gets a living, draggable version. No installation, no login — open one on your phone during class or at midnight before the project is due. If you can predict what the picture will do before you drag, you understand the concept.
The Overfitting Explorer
A hidden demand curve seen through noisy weeks. Drag the polynomial degree and watch training error fall forever while test error turns back up — then add data, cut noise, and resample to see which cures work.
The Leakage Time Machine
A real model retrained at each moment on a timeline — a bank's phone campaign, a hotel's bookings. Drag the moment a prediction is made and watch which columns exist yet, how good the model looks, and whether there is still a decision left to make. The one question that catches every leak, made draggable.
The Threshold & Profit Explorer
Two overlapping score distributions, a draggable cutoff, a live confusion matrix, ROC curve, and profit curve. Change the offer cost, the value of a save, or the contact budget and watch the best threshold walk away from 0.5.
The Gradient Boosting Explorer
Start from a flat guess and add one tiny tree at a time, each fit to what the ensemble still gets wrong. Drag the learning rate and depth, step through the ensemble, and watch held-out error bottom out and turn back up — the point where early stopping stops.
The Conformal Explorer
A model, a calibration set, and a 90% band sized by the 90th-percentile residual. Draw new cases a hundred times and watch coverage settle on the target — then underfit the model, shrink the calibration set, or shift the data and see which promises survive.
The K-Means Explorer
Customers on a recency × frequency plane. Place centroids, step assignment and update one iteration at a time, compare k-means++ to random starts, read the elbow and silhouette — and catch k-means inventing segments in a single blob.