Understanding the Curves
Training Error (blue): Error on the training dataset. Generally decreases as model learns.
Validation Error (orange): Error on held-out validation data. Indicates generalization performance.
Scenarios:
- Good Fit: Both errors decrease and converge at low values
- Overfitting: Training error continues decreasing but validation error increases - model memorizes training data
- Underfitting: Both errors remain high - model is too simple
- Early Stopping: Stop training when validation error starts increasing to prevent overfitting