Machine Learning
DBSCAN
Density-based clustering that needs no k: it floods outward wherever points are packed tightly and labels the sparse leftovers as noise.
Click the canvas to add points
corebordernoiseeverything p reachesthe core that connects them
Click a point to move q, then p · dashed circles are ε
Clusters found—
—
View
Random data
Case
clusters found—
noise points—
p is—
q is—
same cluster—
next click sets—
What to observe
- Press Run. DBSCAN starts at a dense point and floods outwardthrough neighbors within ε, so a cluster takes whatever shape the data has, even a crescent.
- Grey points are noise: they never gather enough ε-neighbors to join anyone. Shrink ε and everything shatters into noise; grow it and separate clusters merge.
- In the Reachability view, One way: a core reaches the border on its fringe, but the border reaches nothing back, because a hop must leave from a core. Reachability is not symmetric.
- Two borders: neither reaches the other, yet they share a cluster, because one core reaches both. That relation, connected, is what defines a cluster.
Shortcuts:space run/pause · s step · r reset · f fullscreen