liveBook · Manning

Unsupervised learning

As explained in Grokking Machine Learning, Second Edition

A machine-learning paradigm that discovers structure or patterns in unlabeled data without target outcomes.

Unsupervised learning finds useful structure in data without a target label to predict. It can reveal which items are similar, simplify complex data, or generate new examples that resemble existing ones. section 2.10 section 2.17

A dataset is unlabeled when it has no particular column designated for prediction; whether a feature counts as a label depends on the problem. Unlabeled pet images, for example, do not identify each image as a dog or cat, but an algorithm can still compare and group them. section 2.6 section 2.10

The main branches use unlabeled data in different ways:

  • Clustering groups similar data points into clusters, such as houses grouped by location, price, or size. section 2.11
  • Dimensionality reduction describes data with fewer features while trying to lose as little information as possible. section 2.10 section 2.17
  • Generative learning creates new data points resembling the dataset, including realistic faces, music, or poetry. section 2.10 section 2.17

Grouping emails without labels

Suppose nine emails are recorded only by size and recipient count. Plotting those two features reveals three natural groups. A clustering algorithm can form the groups; people can then label them as Personal, Social, or Promotions. The same approach can use message words, senders, attachments, or links. section 2.11

Figure 2.4 An unsupervised learning algorithm can still extract information from data. For example, it can group similar elements together.

Shows how an unsupervised algorithm groups similar elements even when the groups have no known names. section 2.10

How Grokking Machine Learning, Second Edition covers it

How deep it goes
Defines three branches and works through clustering intuitively, but stops short of detailed algorithm implementations.
Examples
Uses pet images, housing records, and a nine-email dataset to illustrate similarity, grouping, and feature-based clusters.
What it assumes
Assumes only basic awareness of datasets and labels, while distinguishing labeled from unlabeled data explicitly.
What you can do after
Apply clustering to group similar unlabeled items and recognize dimensionality reduction or generation as alternative uses.
How it connects
Connects unlabeled data with supervised learning, clustering, dimensionality reduction, generative learning, and preprocessing.

Also explained in

Compare how 22 books explain unsupervised learning →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Grokking Machine Learning, Second Edition. The section links point to the parts of the book it draws on.