Unsupervised learning Introduction
Clustering
Group observations using a notion of similarity.
Before this
The intuition
Clustering finds groups without supplied target labels. A method such as k-means alternates between assigning points to nearby centers and updating those centers.
A concrete example
Points [1, 1] and [1, 2] are close under Euclidean distance; [9, 9] is far away. A two-cluster solution can separate the distant point from the pair.
What to remember
Groups depend on the distance measure, feature scaling, and algorithm. They are not necessarily natural categories.
Check your understanding
Why can measuring one feature in millimetres instead of metres change a clustering?
Reveal explanation
Without scaling, that feature can dominate distance calculations.