Short Answer

Why is a mixed dev/test split misleading in a warehouse-vision project?

Question: Answer in one to three sentences using the characteristics from the evaluation-data example.

Sample answer: If you randomly mix 84,000 warehouse photos with 6,000 smartphone photos, the dev/test sets end up being about 93.3% warehouse images. That would not match the future smartphone images the model is meant to handle, so the evaluation would be misleading.

Key points:

  • Random mixing makes the dev/test sets about 93.3% warehouse images.
  • The split does not match the future smartphone-image distribution.

Rubric: The response must state that random shuffling leaves the dev/test sets dominated by warehouse images at about 93.3%, which means they do not reflect the future smartphone-image distribution.

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