An epoch is one complete pass through the training dataset during model optimization. The number of epochs is a hyperparameter: it is chosen before training, while model parameters are learned from data. section 8.1.1 section 8.1
Use the epoch count to balance learning progress against generalization:
- Increasing the number of epochs can improve learning up to a point. section 8.1.1
- Too many epochs can cause overfitting: the model memorizes training data and performs poorly on unseen data. section 8.1.1
- Early stopping can interrupt training when a validation metric stops improving for a specified number of epochs. section 8.1.1
Training configuration
A training run sets the number of epochs alongside the learning rate and batch size. The epoch count controls how often the entire dataset is presented, while the other hyperparameters shape update size and gradient calculation. section 8.1.1
