Table of Contents

Class RadialBasisFunctionOptions

Namespace
AiDotNet.Models.Options
Assembly
AiDotNet.dll

Configuration options for Radial Basis Function (RBF) models, a type of artificial neural network that uses radial basis functions as activation functions for approximating complex non-linear relationships.

public class RadialBasisFunctionOptions : NonLinearRegressionOptions
Inheritance
RadialBasisFunctionOptions
Inherited Members

Remarks

Radial Basis Function networks are a specialized type of neural network that utilize radially symmetric functions (typically Gaussian) centered at specific points in the feature space. These networks excel at function approximation, interpolation, and classification tasks. RBF networks consist of an input layer, a hidden layer with RBF activation functions, and an output layer. Each neuron in the hidden layer represents a radial basis function centered at a particular point. The output of the network is typically a linear combination of these basis functions. RBF networks are known for their ability to model complex non-linear relationships while often requiring less training time than traditional multilayer perceptrons. They are particularly effective for problems where the data exhibits localized patterns or when smooth interpolation between data points is desired.

For Beginners: Radial Basis Function networks are a special kind of AI model that's good at finding patterns in data.

Think about weather prediction:

  • Traditional models might try to find one formula that works for the whole world
  • But an RBF network places "experts" at different locations
  • Each "expert" (or center) specializes in predicting weather in their local area
  • The final prediction combines opinions from nearby experts, with closer ones having more influence

What this technique does:

  • It places a number of "centers" throughout your data
  • Each center is like a spotlight that illuminates the nearby data points
  • The model learns how strong each spotlight should be
  • Predictions are made by seeing how much light falls on new data points

This is especially useful when:

  • Your data has clusters or regions with different patterns
  • You need a model that can adapt to different "neighborhoods" in your data
  • You want smooth transitions between these different regions
  • The relationship between inputs and outputs changes across your data space

For example, in image recognition, different RBF centers might specialize in detecting different shapes or textures, and the combined output helps identify the complete image.

This class lets you configure how the RBF network is structured and initialized.

Properties

NumberOfCenters

Gets or sets the number of RBF centers (hidden neurons) in the network.

public int NumberOfCenters { get; set; }

Property Value

int

The number of centers, defaulting to 10.

Remarks

This parameter determines how many radial basis functions (centers) will be used in the hidden layer of the network. Each center represents a specific point in the feature space around which a radial basis function (typically Gaussian) is placed. The centers are usually determined through clustering algorithms such as k-means, or they may be randomly sampled from the training data. More centers allow the network to capture more complex patterns and local variations in the data, but increase computational cost and may lead to overfitting with limited training data. Fewer centers result in a simpler model that generalizes more broadly but might miss important local patterns.

For Beginners: This setting controls how many "experts" or "centers" the model uses to analyze your data.

The default value of 10 means:

  • The model will place 10 different centers throughout your data space
  • Each center specializes in making predictions for data points near it
  • The final prediction combines information from multiple centers

Think of it like placing weather stations across a country:

  • Each weather station (center) is good at predicting local conditions
  • More stations give you more detailed coverage but require more maintenance
  • Too few stations might miss important local weather patterns
  • Too many stations might be wasteful and could start reporting noise

You might want more centers (like 50 or 100):

  • When you have a lot of training data
  • When your data has many distinct regions or clusters
  • When you need to capture very detailed local patterns
  • When your problem is complex and requires fine-grained analysis

You might want fewer centers (like 5 or 3):

  • When you have limited training data
  • When you want to avoid overfitting
  • When your data has simple patterns
  • When you need faster prediction times
  • When you want a more interpretable model

Finding the right number of centers often requires experimentation to balance model complexity against generalization ability.

Edit this page