Case Study

Using component error counts to decide where to improve a document-processing pipeline.

Case context: A team is building a document-processing pipeline with two stages: a page-segmentation module and a topic classifier. After reviewing 120 mistakes on the development set, they find that 84 errors come from the page-segmentation module and 36 errors come from the topic classifier.

Question: Which component should the team prioritize for improvement, and why?

Sample answer: The team should prioritize the page-segmentation module. It is responsible for most of the development-set mistakes, with 84 of the 120 errors, while the topic classifier accounts for only 36. Improving the component that causes the larger share of errors is more likely to reduce overall pipeline failures.

Key points:

  • Prioritize the page-segmentation module.
  • It causes 84 of the 120 observed errors.
  • The topic classifier is responsible for the remaining 36 errors.

Rubric: The answer must recommend focusing on the page-segmentation module and justify that choice by noting that it is responsible for 84 of the 120 errors, whereas the topic classifier is responsible for 36.

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