Deciding Whether a Planned Improvement Is Worth the Time
Question: According to project-planning advice in machine learning, what is the main value of estimating how much a proposed change might improve accuracy before spending several weeks building it?
Sample answer: It gives a numerical basis for deciding whether the change is worth the engineering time or whether that time should be used on a more valuable task.
Key points:
- Provides a numerical basis for comparing options.
- Helps prevent wasting development time.
- Helps decide whether another task is a better use of time.
Rubric: The answer should explain that estimating the likely gain gives a numerical basis for deciding whether the work is worth the development time or whether another task should be prioritized.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
What is the best first step before spending several weeks on a proposed model change?
Error analysis can provide a quantitative basis for deciding whether a proposed improvement is worth the engineering effort.
Before spending two weeks on a feature, the safest first step is to _____ its likely effect on validation accuracy.
Match each error-analysis idea to how it helps with project investment decisions.
Order the steps for using error analysis to decide whether a model improvement is worth funding.
What does an error-counting review help you estimate quickly?
Build first, estimate later
Error analysis provides a fast estimate of the possible _____ gained by adding a new improvement to a machine learning system.
Match each error-analysis finding to the investment decision it most directly supports.
Order the steps for deciding whether fixing one error type deserves a month of engineering work.
Using Error Analysis to Judge Whether an Improvement Is Worth the Effort
Deciding Whether to Add a Commercial OCR Module
Deciding Whether a Planned Improvement Is Worth the Time