Case Study

Evaluate a team's decision to immediately implement a new algorithm without prior error analysis.

Case context: A machine learning team comes up with an exciting new idea to improve their computer vision model. Eager to see results, they decide to immediately start writing code for the new algorithm, which they estimate will take about a month. One engineer suggests they should first manually review 100 misclassified images to see if the new algorithm addresses the actual errors, which would take about two hours.

Question: Based on Andrew Ng's advice, what should the team decide to do, and what are the specific risks of ignoring the dissenting engineer's suggestion?

Sample answer: The team should decide to pause implementation and manually review the 100 misclassified images. As Andrew Ng advises, spending under two hours on error analysis is a small investment compared to a month of coding. The specific risk of ignoring the engineer and skipping the analysis is that the team might spend a full month building the new algorithm, only to discover afterward that it provides little to no benefit because it doesn't address the actual underlying errors in the model.

Key points:

  • The team should perform the two-hour manual review.
  • Jumping straight into implementation is a common mistake.
  • The risk is wasting a month of effort for little benefit.

Rubric: The response must recommend performing the manual review and cite the risk of wasting a month of effort for little benefit. It should also contrast the two-hour cost with the one-month risk.

0

1

Updated 2026-05-27

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