Training and validation scores converge to the same poor plateau as sample size grows. What should you try before…
Classic MLLevel: IntermediateValue of collecting more data
Training and validation scores converge to the same poor plateau as sample size grows. What should you try before assuming more similar data will solve it?
Hints
- Compare the gap with the absolute score level.
- A small gap can coexist with poor learning.
Answer
Investigate model capacity, feature quality and target noise. The curve suggests the current representation/learner is constrained; it does not prove that unlimited more data will remove that constraint.
Source and licence: scikit-learn 1.7.2 — learning curve · BSD-3-Clause · answer: Quantoros editorial