Short Answer

How should class proportions be chosen for balanced learning-curve samples?

Question: When you create very small subsets to plot a learning curve for data with uneven labels or many classes, what principle should guide how many examples of each class go into a subset?

Sample answer: Make the subset's class mix match the class mix of the full training set as closely as possible.

Key points:

  • Do not draw tiny subsets purely at random when the data are highly skewed or have many labels.
  • Keep each subset’s label proportions as close as you can to those in the complete training data.

Rubric: The response must say that each class should appear in the subset in roughly the same proportion as in the original training data.

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