Definition
Typically, a learning curve displays training and validation scores on the y-axis against the number of training samples or iterations on the x-axis. It helps diagnose whether a model suffers from high bias (underfitting) or high variance (overfitting). By observing the gap between training and validation curves, practitioners can decide whether to collect more data, simplify the model, or adjust regularization parameters to improve generalization.
Summary
A learning curve plots a model’s performance metric against the amount of training data or training epochs to visualize the learning progress.
Key Concepts
- Training score
- Validation score
- Underfitting vs Overfitting
Use Cases
- Diagnosing model performance issues
- Determining sample size requirements
- Monitoring training convergence