Labs / Computer Science
K-Means Clustering
Pick how many clusters you want, then hit Step (or let it auto-run) to watch the algorithm work. Every round each point recolors to its nearest centroid, and every centroid slides to the average position of the points that chose it — repeat until nothing moves. Click empty space to drop in your own points.
iter 0inertia 0.0sizes —
What to try
- Does the inertia readout ever go up from one step to the next, or only down?
- Hit Reseed a few times without changing anything. Do the centroids always settle in the same place?
- The data was made from four blobs. What happens when you ask for only k = 2 clusters — or as many as k = 5?
- Click to add a lone point far from every cluster. How much does one outlier drag its centroid around?
- How many iterations does it usually take to converge? Does more k mean more steps, or fewer?