Learn Before
Essay

Why a Better Score Does Not Always Produce the Best Output

Question: In your own words, why might a system return candidate Y even if the scoring rule assigns a higher value to candidate X? What does this say about the difference between the scorer and the search method?

Sample answer: The scoring rule only evaluates and compares candidate outputs; it does not choose the final answer by itself. A separate search procedure is responsible for finding a candidate with a high score. If that search procedure is approximate, it may miss the true highest-scoring candidate and stop at candidate Y instead of candidate X. So when ScoreB(X) is greater than ScoreB(Y), the problem is not necessarily the score definition. It may mean the search method failed to find the best-scoring option. This shows that both parts matter: the scorer defines what is preferred, and the search algorithm must be strong enough to locate that preferred output.

Key points:

  • The scoring rule ranks candidate outputs but does not directly select the final one
  • The search procedure may be approximate and fail to find the best-scoring candidate
  • If ScoreB(X) > ScoreB(Y), the scoring rule can still be correct
  • The failure can come from the search method rather than the scorer

Rubric: Full credit: clearly separates the role of the scoring rule from the role of search and explains the failure as a search limitation, not a scoring error. Partial credit: identifies one role correctly but does not fully distinguish the two. No credit: misses the distinction or incorrectly blames the scoring rule.

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