Cross Validation

KFold

  • Repeat for m = 1..L
    • Split data into roughly equal sizes. Disjoint subsets
    • Get model with min Emperical Risk
    • Test it with validation set
    • Avg it for the folds for this value of m
  • Find optimal class for that m that had min avg validation risk (aka training error)
  • Compute using the original training data

Leave One Out

  • Each D contains a single training example
  • For tiny datasets