Hill Climbing for Parameter Tuning

• Problem Description: PA

  • Examples: Troubleshooting Records (21 attributes)
  • Background Knowledge: Parameterized Max Rules
  • Hypothesis Language: Vector of 14 Numeric Parameters
    
  • I Initial state: Vector of Numeric Parameter
  • O Operators
    • Add 2.5% of parameter value to each parameter
    • Subtract 2.5% of parameter value to each parameter
  • E Cost of Misclassification

If there's an X > *Y*

Class 1.

If distance from station > *Z*

Class 2.

If distance from station <= *Z*

and there's no X <= *A*

Class 3.

Previous slide

Next slide

Back to first slide

View graphic version