Hooke-Jeeves Pattern Search
Version
1.0
Description
Hooke-Jeeves Pattern Search is a non-gradient based algorithm for unconstrained problems. This uses a simple search pattern where the current design is perturbed (plus and minus) in each dimension one at a time. If a better design is found, the current design point is moved to the new design. Otherwise, the perturbation step size is reduced and the pattern search is repeated until the step size becomes smaller than the convergence criteria.
References
- "Direct search solution of numerical and statistical problems," Hooke, R., Jeeves, T.A., Journal of the Association for Computing Machinery (ACM) 8 (2), pp.212-229
- See also Pattern Search
Control Parameters
| Name | Default Value | Description |
|---|---|---|
| MaxEvaluations | 1000 | Maximum number of function evaluations. The value specified must be a positive integer. |
| MaxIterations | 100 | Maximum number of iterations that algorithm can run for. The value must be a positive integer. |
| StepSizeReductionFactor | 0.5 | Step size reduction factor if pattern search did not find a better design. This value should be between 0 and 1. A larger value gives greater possibility of convergence at a cost of more function evaluations. A smaller value reduces the number of function evaluations but increases the risk of premature termination. |
| StepSizeTolerance | 1E-06 | Termination step size is the criterion for stopping the optimization. If the step size is reduced below this number while trying to find a better design, the optimization stops. Option must have a value greater than 0 but less than StepSizeReductionFactor. |