Table of Contents

Class BayesianOptimizerOptions<T, TInput, TOutput>

Namespace
AiDotNet.Models.Options
Assembly
AiDotNet.dll

Configuration options for the Bayesian optimization algorithm.

public class BayesianOptimizerOptions<T, TInput, TOutput> : OptimizationAlgorithmOptions<T, TInput, TOutput>

Type Parameters

T

The data type used by the kernel function.

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

Remarks

Bayesian optimization is a sequential design strategy for global optimization of black-box functions that doesn't require derivatives. It's particularly useful for optimizing expensive-to-evaluate functions.

For Beginners: Bayesian optimization is like a smart search strategy. Imagine you're trying to find the highest point in a hilly landscape while blindfolded - you can only ask for the height at specific locations. Instead of checking every possible spot (which would take too long), Bayesian optimization uses what it learns from previous measurements to make educated guesses about where the highest point might be. It balances between exploring new areas and focusing on promising regions, making it efficient for finding optimal solutions when each evaluation is time-consuming or expensive.

Properties

AcquisitionFunction

Gets or sets the type of acquisition function to use.

public AcquisitionFunctionType AcquisitionFunction { get; set; }

Property Value

AcquisitionFunctionType

The acquisition function type, defaulting to UpperConfidenceBound.

Remarks

The acquisition function determines the strategy for selecting the next point to evaluate. Common options include Upper Confidence Bound (UCB), Expected Improvement (EI), and Probability of Improvement (PI).

For Beginners: This determines the strategy the algorithm uses to decide where to look next. The default (UpperConfidenceBound) balances exploring uncertain areas and exploiting promising ones. Think of it like:

  • Upper Confidence Bound: "I'll check places that might be good or that I'm uncertain about"
  • Expected Improvement: "I'll check places likely to be better than the best I've found so far"
  • Probability of Improvement: "I'll check places most likely to be at least slightly better than my current best"
Different strategies work better for different problems, but the default is a good starting point.

AcquisitionOptimizationSamples

Gets or sets the number of samples used when optimizing the acquisition function.

public int AcquisitionOptimizationSamples { get; set; }

Property Value

int

The number of acquisition optimization samples, defaulting to 1000.

Remarks

The acquisition function helps decide where to sample next. This parameter controls how thoroughly the algorithm searches for the best next point to evaluate.

For Beginners: After taking some measurements, the algorithm needs to decide where to look next. This parameter (default: 1000) controls how carefully it considers its options. Think of it like considering 1000 possible locations and picking the most promising one. A higher number means more thorough consideration but takes longer to compute. For most problems, the default value provides a good balance between computation time and optimization quality.

ExplorationFactor

Gets or sets the exploration factor (kappa) used in the acquisition function.

public double ExplorationFactor { get; set; }

Property Value

double

The exploration factor, defaulting to 2.0.

Remarks

This parameter balances exploration (searching new areas) versus exploitation (refining known good areas). Higher values favor exploration, while lower values favor exploitation.

For Beginners: This controls the balance between trying new areas (exploration) and focusing on promising areas already discovered (exploitation). The default value of 2.0 provides a balanced approach. Think of it like deciding whether to visit new restaurants or return to ones you already know are good. A higher value means more adventurous exploration, while a lower value means sticking closer to what's already known to work well.

InitialSamples

Gets or sets the number of initial random samples to evaluate before starting the optimization process.

public int InitialSamples { get; set; }

Property Value

int

The number of initial samples, defaulting to 5.

Remarks

These initial samples help build the first Gaussian Process model before the algorithm starts making informed decisions about where to sample next.

For Beginners: This is like taking a few random measurements in our hilly landscape example before starting to make educated guesses. The default value of 5 means the algorithm will first check 5 random locations to get a basic understanding of what the landscape looks like. Too few initial samples might give an incomplete picture, while too many might waste evaluations on potentially uninteresting areas.

IsMaximization

Gets or sets whether the objective should be maximized (true) or minimized (false).

public bool IsMaximization { get; set; }

Property Value

bool

KernelFunction

Gets or sets the kernel function used by the Gaussian Process model.

public IKernelFunction<T> KernelFunction { get; set; }

Property Value

IKernelFunction<T>

The kernel function, defaulting to Gaussian kernel (also known as Radial Basis Function kernel).

Remarks

The kernel function determines how the algorithm measures similarity between points in the search space, which affects how it generalizes from observed data points to unobserved points.

For Beginners: The kernel function helps the algorithm understand how similar different points are to each other. The default Gaussian kernel (also called Radial Basis Function kernel) assumes that points close to each other will have similar values, with the similarity decreasing smoothly as distance increases. This is like assuming that in our hilly landscape, nearby locations tend to have similar heights. The Gaussian kernel works well for many problems, especially when the underlying function is smooth.

LowerBound

Gets or sets the lower bound for the search space.

public double LowerBound { get; set; }

Property Value

double

The lower bound value, defaulting to -10.

Remarks

This value defines the minimum value for each dimension in the search space.

For Beginners: This sets the minimum value the algorithm will consider for each variable (default: -10). In our landscape analogy, this would be like setting the western and southern boundaries of the area you're willing to explore. You should set this based on what makes sense for your specific problem - for example, if you're optimizing a learning rate that can't be negative, you might set this to 0 instead.

UpperBound

Gets or sets the upper bound for the search space.

public double UpperBound { get; set; }

Property Value

double

The upper bound value, defaulting to 10.

Remarks

This value defines the maximum value for each dimension in the search space.

For Beginners: This sets the maximum value the algorithm will consider for each variable (default: 10). Continuing our landscape analogy, this would be like setting the eastern and northern boundaries of your search area. Setting appropriate bounds helps the algorithm focus on realistic values and can significantly improve optimization efficiency.

Edit this page