Explain how a dev-error learning curve can support a decision to collect more data.
Question: In a concise analytical response, explain how the desired performance level and visual extrapolation of dev error are used to estimate the benefit of more training data.
Sample answer: First, add the desired performance level to the learning curve. Then visually extrapolate the dev-error curve to estimate how much closer increasing the training-set size could bring dev error to that target. The result should be treated as a plausibility judgment rather than a guarantee. In the passage's example, the curve suggested that doubling the training-set size might be sufficient to reach the desired performance.
Key points:
- Add the desired performance level to the learning curve.
- Visually extrapolate the dev-error curve.
- Estimate how much closer more data could bring performance to the target.
- Treat the estimate as plausible rather than guaranteed.
- Recognize that doubling the training-set size was the example conclusion.
Rubric: A strong response identifies the target line, explains visual extrapolation of the dev-error curve, connects the projection to the estimated benefit of more data, and characterizes the conclusion as plausible rather than certain.
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 does visually extrapolating the dev-error curve help estimate?
Visual extrapolation gives a guaranteed prediction of performance after adding data.
Add the desired _____ level to the learning curve before extrapolating dev error.
Match each learning-curve element with its role in estimating the value of more data.
Order the reasoning process for estimating whether more data could reach the desired performance.
Explain how a dev-error learning curve can support a decision to collect more data.
A projected dev-error curve reaches the target after doubling the data. What should the team conclude?
Why should the desired performance level be shown on the learning curve?
Which conclusion best reflects the passage's example learning curve?
A dev-error extrapolation may justify saying that a data increase looks plausible, not certain.