optimization_method

Change the optimization method used to compute hyperparameters

Specification

  • Alias: None

  • Arguments: STRING

  • Default: global

Description

Select the optimization method to compute hyperparameters of the Gaussian Process by specifying one of these arguments:

  • global (default) - DIRECT method

  • local - CONMIN method

  • sampling - generates several random guesses and picks the candidate with greatest likelihood

  • none - no optimization, pick the center of the feasible region

The none option, and the starting location of the local optimization, default to the center, in log(correlation length) scale, of the of feasible region.

Surfpack picks a small feasible region of correlation parameters.

Note that we have found the global optimization method to be the most robust.