Table of Contents

Class CMAESOptimizerOptions<T, TInput, TOutput>

Namespace
AiDotNet.Models.Options
Assembly
AiDotNet.dll

Configuration options for the CMA-ES (Covariance Matrix Adaptation Evolution Strategy) optimization algorithm.

public class CMAESOptimizerOptions<T, TInput, TOutput> : GeneticAlgorithmOptimizerOptions<T, TInput, TOutput>

Type Parameters

T

The type of data being optimized.

TInput
TOutput
Inheritance
OptimizationAlgorithmOptions<T, TInput, TOutput>
CMAESOptimizerOptions<T, TInput, TOutput>
Inherited Members

Remarks

CMA-ES is a powerful evolutionary algorithm for difficult non-linear, non-convex optimization problems. It adapts its search strategy during optimization by learning dependencies between variables and step sizes, making it effective for complex problems where other methods might fail.

For Beginners: CMA-ES is like a smart search algorithm that tries to find the best solution to a complex problem. Imagine you're looking for the lowest point in a mountain range with fog everywhere - you can only see what's right around you. CMA-ES works by sending out multiple "scouts" (solutions) in different directions, then learning from their findings to decide where to search next. It's particularly good at handling tricky landscapes where simple "always go downhill" approaches would get stuck. This class lets you configure how that search process works.

Properties

InitialStepSize

Gets or sets the initial step size that controls how far the algorithm explores from the starting point.

public double InitialStepSize { get; set; }

Property Value

double

The initial step size as a decimal, defaulting to 0.5.

Remarks

The step size determines the initial spread of the search distribution. A larger value encourages more exploration early in the search process, while a smaller value focuses on exploitation near the starting point.

For Beginners: This setting determines how big the initial "steps" are when the algorithm starts searching. The default (0.5) means it will look at solutions that are moderately different from your starting point. If you set this higher (like 2.0), it will explore more widely at first, which is good if you think the best solution might be far from your starting point. If you set it lower (like 0.1), it will focus more closely around your starting point, which works well if you're already close to the optimal solution.

MaxGenerations

Gets or sets the maximum number of generations (iterations) the algorithm will run.

public int MaxGenerations { get; set; }

Property Value

int

The maximum number of generations, defaulting to 100.

Remarks

This parameter limits the total computational budget of the optimization process. The algorithm will stop either when it converges or when it reaches this number of generations.

For Beginners: This is simply how many rounds of optimization the algorithm will perform before stopping. Each generation involves creating and evaluating multiple solutions, then learning from the results to improve the next generation. The default (100) is enough for many problems, but complex problems might need more generations to find good solutions. If the algorithm finds an excellent solution earlier, it may stop before reaching this limit.

PopulationSize

Gets or sets the number of candidate solutions evaluated in each generation.

public int PopulationSize { get; set; }

Property Value

int

The population size, defaulting to a formula based on problem dimensionality: 4 + (int)(3 * Math.Log(100)).

Remarks

The population size affects both the quality of solutions and computational cost. Larger populations provide more exploration but require more function evaluations. The default formula (4 + 3*ln(D)) is a common heuristic where D is the problem dimension.

For Beginners: This setting controls how many different potential solutions the algorithm tries in each round (or "generation"). The default value is calculated based on a formula that works well for most problems. Think of it like sending out search parties - more parties can cover more ground but require more resources. For simple problems, you might reduce this value to speed things up, while for complex problems with many variables, you might increase it to find better solutions.

StopTolerance

Gets or sets the convergence threshold that determines when the algorithm should stop.

public double StopTolerance { get; set; }

Property Value

double

The stop tolerance as a decimal, defaulting to 1e-12 (0.000000000001).

Remarks

The algorithm stops when the change in the best solution falls below this threshold, indicating that further optimization is unlikely to yield significant improvements.

For Beginners: This setting determines how precise the algorithm needs to be before it decides it's found a good enough solution and stops. The default value (0.000000000001) is very small, meaning the algorithm will keep going until it's making virtually no progress. Think of it like deciding when to stop polishing a surface - at some point, more polishing doesn't make a noticeable difference. You might increase this value (making it less strict) if you want faster results and don't need the absolute best solution.

Edit this page