Case Study

Deciding on dev set structure for stock market anomaly detection.

Case context: A machine learning team is developing an algorithm to detect micro-anomalies in high-frequency financial trading data. These anomalies occur at millisecond scales and are impossible for human analysts to identify or evaluate. The team is debating whether to set aside time and resources to manually analyze an Eyeball dev set of misclassified anomalies.

Question: Based on the difficulty of this task for humans, how should the team decide on the inclusion of an Eyeball dev set, and why?

Sample answer: The team should omit the Eyeball dev set. Since human analysts cannot identify or evaluate these anomalies, manually examining misclassified examples would not be helpful because they cannot diagnose why the algorithm made a mistake.

Key points:

  • Recommendation to omit the Eyeball dev set.
  • Lack of human capability to perform the task well.
  • Inability to diagnose why the algorithm made a classification error.

Rubric: The response must recommend omitting the Eyeball dev set and justify it by explaining that because humans cannot do the task well, analyzing errors to determine why the algorithm failed is too difficult.

0

1

Updated 2026-05-27

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