Short Answer

How to Choose Dev and Test Data When Future Data Will Differ

Question: If the data you expect to see in production is different from your training data, what principle should guide selection of the development and test sets?

Sample answer: Choose dev and test examples so they match the future data the system is meant to handle. Their job is to measure performance on the distribution you care about, not to mirror the training set.

Key points:

  • Dev and test sets should represent the future operating environment.
  • They should be drawn from the distribution the model is expected to perform well on.

Rubric: The answer must say that dev and test sets should be selected to match the future data distribution the model will face and optimize for.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI

Related