Online k-means — how does it work?
hardAnswer
- For each incoming point, find nearest centroid and update it: ← with decaying learning rate.
- Streaming-friendly, constant memory.
- Sensitive to arrival order and cluster drift.
- Modern alternative: mini-batch k-means; production streaming systems (Kafka + Flink) prefer mini-batch for robustness.
- Concept drift → reset centroids periodically.
Check yourself — multiple choice
- Same as batch
- Per-point update μ ← μ + η(x - μ); constant memory; order-sensitive + drift-sensitive → prefer mini-batch or reset periodically
- Random
- Not real
Online k-means: SGD-like centroid updates; sensitive to order / drift.
#clustering
Practise Unsupervised Learning
214 interview questions in this topic.