LCDB 1.1: A Database Illustrating Learning Curves Are More Ill-Behaved Than Previously Thought
This reveals a widespread issue in machine learning for researchers and practitioners relying on learning curves, though it is incremental as it builds on prior database efforts.
The paper tackles the assumption that learning curves are well-behaved by constructing LCDB 1.1, a large-scale database with modern learners, and finds that approximately 15% of learning curves show significant ill-behavior, almost double previous estimates, impacting tasks like model selection.
Sample-wise learning curves plot performance versus training set size. They are useful for studying scaling laws and speeding up hyperparameter tuning and model selection. Learning curves are often assumed to be well-behaved: monotone (i.e. improving with more data) and convex. By constructing the Learning Curves Database 1.1 (LCDB 1.1), a large-scale database with high-resolution learning curves including more modern learners (CatBoost, TabNet, RealMLP and TabPFN), we show that learning curves are less often well-behaved than previously thought. Using statistically rigorous methods, we observe significant ill-behavior in approximately 15% of the learning curves, almost twice as much as in previous estimates. We also identify which learners are to blame and show that specific learners are more ill-behaved than others. Additionally, we demonstrate that different feature scalings rarely resolve ill-behavior. We evaluate the impact of ill-behavior on downstream tasks, such as learning curve fitting and model selection, and find it poses significant challenges, underscoring the relevance and potential of LCDB 1.1 as a challenging benchmark for future research.