CMAEvolutionStrategyOptimizer.CovarianceRefreshInterval Property

public int CovarianceRefreshInterval { get; set; }

Public Property CovarianceRefreshInterval As Integer
	Get
	Set

public:
property int CovarianceRefreshInterval {
	int get ();
	void set (int value);
}

member CovarianceRefreshInterval : int with get, set

Property Value

Full covariance with positive updates may reuse its sampling and whitening factors for this many generations. Covariance and evolution paths still update every generation. A larger interval reduces decomposition work but changes the search trajectory.

Active covariance updates always refresh every generation to retain their current covariance metric. Diagonal covariance also refreshes every generation because this takes only linear work. These policies can refresh sooner than the supplied maximum.

The first generation and each new solve refresh the factors. Covariance is checked for finite values every generation, including when factors are reused.

Reference