Case Study

How should a data-limited team use an Eyeball dev set with 10 mistakes?

Case context: A team has an Eyeball dev set in which its classifier makes 10 mistakes. The team cannot afford to place more data into the set and wants to choose which error categories to address first.

Question: Diagnose the limitation of the team's evidence and decide how the team should use the 10 mistakes for project prioritization.

Sample answer: The team should recognize that 10 mistakes form a very small Eyeball dev set. It should not treat the apparent impact of each error category as an accurate estimate because the sample is too small. Since no additional data is available, the team should still inspect the errors and use the observed categories as preliminary guidance for prioritization, while remaining cautious about the estimates.

Key points:

  • The set of 10 mistakes is very small.
  • Category-impact estimates will be difficult to make accurately.
  • The team cannot add more data.
  • The errors should still be examined.
  • Priorities based on the sample should be treated as preliminary.

Rubric: Full credit requires identifying the small-sample limitation, rejecting claims of accurate category-impact estimation, and recommending cautious use of the available errors for prioritization because they are better than nothing.

0

1

Updated 2026-07-19

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Machine Learning Strategy

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Yearning @ DeepLearning.AI