Hands-On Machine Learning for Algorithmic Trading

Early stopping

We encountered early stopping as a regularization technique in Chapter 10, Decision Trees and Random Forests. It is probably the most commonly used form of regularization in DL and is popular because it is both effective and simple to use.

It works as a regularization mechanism by monitoring the model's performance on a validation set during training. When the performance ceases to improve for a certain number of observations, the algorithm stops to prevent overfitting.

Early stopping can be viewed as a very efficient hyperparameter selection algorithm that automatically determines the correct amount of regularization, whereas parameter penalties require hyperparameter tuning to identify the ideal weight decay.