What to do with a too-small development set
Question: Give a brief explanation of how to handle a development set that is too small to divide further, and name the ways it may still be used.
Sample answer: If the development set is too small to split, keep all of it together and use it as the manual-review development set. That set can then support detailed error inspection, comparison among candidate models, and hyperparameter adjustment.
Key points:
- The development set is too small to divide.
- Keep the full set together.
- Use it for detailed error inspection.
- Use it for model comparison.
- Use it for hyperparameter adjustment.
Rubric: A strong response says the set should stay intact, identifies it as the manual-review development set, and names all three uses. Partial credit should be given if the answer correctly recommends keeping the set whole and lists one or two of the uses.
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 use of a development set that is too small to split?
A very small development set can be kept as one set instead of being split.
If the development set cannot be divided, use the _____ development set as the eyeball set.
Match each item to its role in a small validation-set workflow.
Put the steps for deciding what to do with a very small development set in order
What to do with a too-small development set
How to handle a validation set that is too small to split
Three functions of a very small validation set
Best use of a dev set that is too small to split
If the full development set is used as an eyeball set, it cannot be used for hyperparameter selection.