Xpress NonLinear multistart
The feature is an additive feature that minimizes the development overhead and effort of implementing parallel multistart searches. The purpose of multistart is two-fold. Traditionally, multistart is a so called globalization feature. It is important to correctly understand what this technology offers, and what it does not. It offers a convenient and efficient way of exploring a larger feasible space building on top of existing local solver algorithms by the means of perturbing initial points and/or parameters or even the problem statement itself. Multistart can also be viewed as a left-alone feature. In a typical situation, versions of a model react favourably to a set of control settings, dependent on data. Multistart allows for a simple way of combining different control setting scenarios, increasing the robustness of the model.
The base problem is defined as the baseline: as the model is normally loaded it without any multistart information, including problem description, callbacks and controls. A run or a job is defined as a problem instance that needs to be solved as part of multistart.
On completion, the current problem is set up to match that of the winner, allowing examination of the winning strategy and solution using the normal means.
The original prob object is not reused, all runs are mode on a copy of the problem, allowing full customization from the callbacks, including changes to structre.
Callbacks are inherited to the multistart jobs from the master problem and can be customized from the the multistart callbacks. XSLinterrupt has a global scope, and a calling it terminates the multistart search.
Although not intended as the primary use, multistart allows the execution of all supported problem classes, so for example alternate MIP strategies can be used in parallel.
The mutistart job pool is maintained and can be extended until the first maxim / minim with XSLP_MULTISTART on. This allows for doing optimizations runs aimed at generating multistart jobs. The multistart pool is dynamic and new jobs can be added on the fly from the jobstart and jobend callbacks.
© 2001-2019 Fair Isaac Corporation. All rights reserved. This documentation is the property of Fair Isaac Corporation (“FICO”). Receipt or possession of this documentation does not convey rights to disclose, reproduce, make derivative works, use, or allow others to use it except solely for internal evaluation purposes to determine whether to purchase a license to the software described in this documentation, or as otherwise set forth in a written software license agreement between you and FICO (or a FICO affiliate). Use of this documentation and the software described in it must conform strictly to the foregoing permitted uses, and no other use is permitted.