BacktrackingLineSearch Class

public sealed class BacktrackingLineSearch : OneDimensionalOptimizer

Public NotInheritable Class BacktrackingLineSearch
	Inherits OneDimensionalOptimizer

public ref class BacktrackingLineSearch sealed : public OneDimensionalOptimizer

[<SealedAttribute>]
type BacktrackingLineSearch = 
    class
        inherit OneDimensionalOptimizer
    end

Inheritance

Object  →  ManagedIterativeAlgorithm<Double, Double, SolutionReport<Double, Double>>  →  ManagedIterativeAlgorithm<Double>  →  OneDimensionalOptimizer  →  BacktrackingLineSearch

Use the BacktrackingLineSearch class as a line search method in a multidimensional optimization algorithm. This class is not suitable for general optimization in one dimension.

The backtracking algorithm is the line search method of choice for most applications, particularly when the gradient of the objective function is relatively expensive. The algorithm uses a custom termination criterion based on the Wolfe conditions. These conditions guarantee convergence of quasi-Newton algorithms in most situations.

The algorithm uses the Armijo conditions as termination criteria. The parameter of the Armijo conditions can be set or retrieved through the DescentFactor property, while the rate of backtracking can be controlled through the MinContractionFactor and MaxContractionFactor properties.

Reference