A practical guide on correctly using train, validation, and test sets in machine learning. Covers the validation overfitting problem that arises from repeated tuning, and recommends k-fold cross-validation and nested cross-validation as solutions. Explains that the test set should only be used once for final unbiased

8m read timeFrom blog.dailydoseofds.com
Post cover image
Table of contents
What are RL environments, and how to build themHow to actually use train, validation, and test sets

Sort: