All courses

Advanced · module A2

Unsupervised Learning and Applied Machine Learning

By the end: Apply clustering, dimensionality reduction and anomaly detection, and build recommenders and forecasts

Start here
Covers
Finding structure without labels, and the specialised problems of recommendation, time series and limited labels
Contents
4 chapters, 15 topics
Time
About 3 hours
Level
Advanced
Your progress
0 / 15

This module builds on Supervised Machine Learning in Depth. We suggest looking at it first, but the choice is yours.

Chapter 1 Clustering

  1. 1.1

    k-Means in Depth

    k-means++, the elbow method and silhouette scores

    Code: k-means from scratch

    J = Σ‖x − μₖ‖²

    15 minBeing written
  2. 1.2

    Hierarchical Clustering

    Agglomerative clustering, linkage and dendrograms

    Code: a dendrogram

    10 min Activity
  3. 1.3

    Density-Based Clustering

    DBSCAN, OPTICS and HDBSCAN

    Interactive: eps and min samples

    12 min Activity
  4. 1.4

    Gaussian Mixture Models and EM

    Soft clustering and the expectation-maximisation algorithm

    Code: fit a mixture

    p(x) = Σπₖ N(x|μₖ,Σₖ)

    15 min Activity

Chapter 2 Dimensionality Reduction

  1. 2.1

    Principal Component Analysis

    Variance, covariance, eigenvectors and choosing components

    Code: PCA from scratch

    15 min Read Activity
  2. 2.2

    t-SNE and UMAP

    Visualising high-dimensional data, and how not to misread it

    Code: visualise digits

    12 minBeing written
  3. 2.3

    Other Methods

    ICA, NMF and autoencoders for compression

    Code: compare methods

    10 minBeing written

Chapter 3 Patterns and Anomalies

  1. 3.1

    Anomaly Detection

    Statistical methods, isolation forest, one-class SVM and LOF

    Code: detect fraudulent transactions

    15 min Activity
  2. 3.2

    Association Rules

    Support, confidence, lift, Apriori and FP-Growth

    Code: market basket analysis

    lift = conf(A→B) / supp(B)

    12 min Activity

Chapter 4 Applied Machine Learning

  1. 4.1

    Recommender Systems

    Content-based filtering, collaborative filtering and matrix factorisation

    Code: a film recommender

    R ≈ UVᵀ

    18 minBeing written
  2. 4.2

    Time Series Forecasting

    Trend, seasonality, stationarity, ARIMA and machine-learning forecasters

    Code: forecast monthly demand

    18 min Activity
  3. 4.3

    Semi-Supervised and Self-Supervised Learning

    Learning with few labels

    Code: self-training

    12 min Activity
  4. 4.4

    Feature Engineering at Scale

    Domain features, interactions, target encoding and feature selection

    Code: improve a model with features

    15 minBeing written
  5. 4.5

    AutoML

    Automated model search and its limits

    Lab: compare AutoML with a hand-built model

    10 minBeing written
  6. 4.6

    Module Review and Assessment

    Coding tasks and a scored quiz

    Graded assessment

    20 minBeing written