Tensor TownISSUE #10 of 120

clustering · k-means · dbscan

NeuraVSThe Overfit Ogre

Neura saysClustering groups similar points without labels — k-means and DBSCAN are the staples.

k-means partitions data into k clusters by repeatedly assigning points to the nearest center and recomputing centers — fast, but you must pick k and it assumes round, similar-sized clusters. DBSCAN groups by density, finding arbitrary shapes and labeling outliers as noise without needing k. Pick by your data's shape and whether you know the cluster count.

Power-ups you unlock

k-means: partition into k clusters (pick k)
Assumes round, similar-sized clusters
DBSCAN: density-based, any shape, finds noise
Choose by shape + whether k is known

The Overfit Ogre attacks — common mistakes

Using k-means on non-spherical clusters
Guessing k with no validation (elbow/silhouette)
Ignoring feature scaling before clustering

Boss battlePick k-means or DBSCAN for tight blobs vs irregular shapes with outliers.

Example code

<!doctype html><html><head><meta charset="utf-8"></head>
<body style="background:#06040d;color:#e6e0ff;font-family:monospace;padding:20px"><pre>round, known-count clusters → k-means
irregular shapes + outliers → DBSCAN (density)</pre></body></html>

▶ Open the interactive comic issue

‹ Classification Vs Regression Dimensionality Reduction · Pca · T-Sne · Umap ›