competitive fitness on empirical data

• decision tree to classify clue words
  • coadapt training case distribution
    • training cases have a fitness
    • init to 0
    • increment if incorrectly classified
    • decrement if correctly classified
    • tournament selection on training cases!
  • need absolute fitness, also
  • overlearning
• competitive classifiers
  • spiral test problem
  • tournament selection
  • only count cases that others get wrong
  • no need for fitness measurements
• both: emphasize harder training cases
• boostin' Bayes