Loading
How to find structure in data that has no labels. The course covers k-means, density and hierarchical clustering, how to score a clustering, and how PCA and autoencoders compress data into fewer dimensions.
01
k-means alternates two steps: assign each point to its nearest centre, then move each centre to the mean of its points. The result depends on the start, so it is run from several starts.
02
DBSCAN grows clusters from points that have many close neighbours. It finds clusters of any shape and marks isolated points as noise.
Practise it
03
Agglomerative clustering starts with one cluster per point and merges the two closest clusters again and again. The linkage rule, such as single or complete, defines what closest means.
Practise it
04
Without labels, a score must compare how close each point is to its own cluster and to the next one. The silhouette coefficient does this for each point.
Practise it
05
PCA finds the directions of largest variance and projects the data onto the first few. It is the best linear compression in squared error.
06
An autoencoder compresses each input into a small code and rebuilds it. With linear layers it learns the same subspace as PCA, and with nonlinear layers it can follow curved structure.
Practise it