Optimization Algorithms

The optimization algorithms currently implemented are: Particle Swarm Optimization (PSO), Adaptive Diversity Particle Swarm Optimization (ADPSO), Genetic Algorithms (GA), Simultaneous Perturbation Stochastic Approximation (SPSA) and Systematic Search (Systematic).

These are available from the optimizers2-sharp library. When using optirunner, you can list the available optimization algorithms and their settings using the --list command line option. There are a few settings common to all optimization algorithms (although some might not apply to all algorithms). Table 3.3 lists these settings.

Table 3.3: Optimization Settings

Name	Description
population-size	The population size (applies to population based methods and has a special meaning for the SPSA and Systematic algorithms).
max-iterations	The maximum number of iterations to run the optimization (does not apply to Systematic).
convergence-threshold	If specified, adds an additional stopping criterion which is based on measurement of convergence. If the convergence is below the threshold, the optimization will be stopped. Convergence is measured by: $$\vert\max_{i \in W}{f_i(\mathbf{x})} - \min_{i \in W}{f_i(\mathbf{x})}\vert$$ Where $W$ is the set of the last convergence-window iterations and $f_i(\mathbf{x})$ is the best fitness in iteration $i$.
convergence-window	The window (iterations) over which to measure convergence.
min-iterations	The minimum number of iterations to run before the convergence stopping criterion is used.

Each of the available algorithms and their settings will be briefly described.