Racing Population Individuals and Algorithms

by Yung Siang Liau for PaGMO / PyGMO

The main objective of this project is two-fold: firstly, to enable individuals (and meta-heuristics algorithms) in PaGMO with stochastic quality indicators be distinguished efficiently via racing algorithms; secondly, to implement interesting functionalities in PaGMO leveraging the racing capability. The most straightforward enhancement is to enable stochastic optimization problems to be solved more efficiently without too much modification to the existing algorithms. A perhaps more interesting consequent of the racing module would be an in-house automatic parameter configuration system for PaGMO optimizers / meta-heuristics. This could potentially mean that in future, algorithms implemented in PaGMO will no longer require (heavy) manual parameter tuning.