Short Answer

State the primary risk of fixing only misclassified dev set labels.

Question: What is the primary risk to your evaluation metrics if you only correct the labels of dev set examples that your system misclassified?

Sample answer: The primary risk is introducing bias into the dev set evaluation. Specifically, it will artificially inflate the measured performance of your system because you are only correcting errors in one direction (changing misclassified to correct) while ignoring errors that favor the model.

Key points:

  • It introduces bias into the dev set evaluation.
  • It artificially inflates the measured performance metrics of the system.

Rubric: The answer should state that it introduces evaluation bias, specifically an optimistic bias that artificially inflates the system's measured performance.

0

1

Updated 2026-05-26

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Related