Unsupervised Learning

Methods

clustering

How to evaluate

• Does the method perform well in the context of business goal? How well a particular unsupervised method performs will largely depend on why one is doing unsupervised learning in the first place. stackoverflow
• Human evaluation
• Leave some data out, add the data afterwards, is the model robust?
• Try a downstream auxiliary task. The performance of the downstream task can serve as a surrogate for the performance of the unsupervised learner.

Readings

• Evaluation Metrics for Unsupervised Learning Algorithms paper
• Stanford notes on unsupervised learning