Search⌘ K

Model Tuning Intuition 201

Explore the concept of model tuning focusing on decision trees using ten-fold cross-validation to balance bias and variance. Understand how to use hyperparameter sets to improve prediction accuracy and estimate generalization error without test data leakage.

Back to the darts

This lesson combines several topics. Assume there’s some training data, access to the CART classification tree algorithm, and some hyperparameter values. This is everything needed to perform ten-fold cross-validation (CV). Regarding the bias-variance tradeoff, each CV iteration is conceptually a dart thrown at the dartboard.

Performing CV for each set of hyperparameter values is a best practice. Assuming there are four sets of hyperparameter values, the following image visualizes cross-validation in terms of the bias-variance tradeoff:

Visualizing cross-validation runs using the dartboard analogy
Visualizing cross-validation runs using the dartboard analogy

Notice in the illustration that each ...