Training, validation and test data
Separate datasets to build, tune and fairly measure a model.
Training data teaches the model; validation data tunes it; test data, kept apart, measures how it performs on data it hasn’t seen. Leakage between sets inflates results.
On the exam
Test data must stay separate to give an unbiased estimate of real-world performance.
Related terms
- Data drift and model driftWhen a model gets worse because the world changed, not the code.
- Data lineage and provenanceWhere data came from and what happened to it on the way.
Test yourself with today’s question of the day.