Validation Curve

noun

say it: “val-ih-DAY-shun KURV” /ˌvæləˈdeɪʃən ˈkɝv/

A graph of a model's scores on training data and on held-out data as one of its settings changes, used to find the setting between underfitting and overfitting.

What is a validation curve?

A validation curve plots a model's training score and validation score as one setting changes, such as a tree's max_depth. Where both are low, the model underfits; where training keeps rising and validation falls, it overfits. scikit-learn's validation_curve() works out the scores with cross-validation.

In a sentence

  • “The validation curve peaked at a depth of 4, then fell as the tree started to overfit.”
  • “Her validation curve showed training scores climbing while validation scores dropped.”

Level: Freelancer · More AI Basics terms at this level

Updated