Unsupervised Learning

No labels, so no notion of a right answer to fit. What is left is structure: points that sit together (clustering), items that co-occur (association rules), and directions along which the data actually varies (SVD).

Clustering

• Clustering Foundations - what a cluster is, and how the families differ.
• Hierarchical Agglomerative Clustering - the dendrogram, and the four linkage rules.
• K-Means - the two-step loop, worked by hand.
• K-Means Limitations & Bisecting K-Means - where the loop fails and what to do instead.

Association rule mining

• Foundations - support, confidence, lift, and the itemset lattice.
• The Apriori Algorithm - the downward closure property, pass by pass.
• Maximal & Closed Itemsets - compressing the frequent itemsets.
• Rule Generation - from frequent itemsets to rules worth reading.
• FP-Growth - the same answers without generating candidates.

Dimensionality

• Singular Value Decomposition - factorising a matrix into the directions that matter.
• Cheat Sheet - the whole topic on one page.