SumKernel<T>

Models & Types · AiDotNet.Kernels

Implements a Sum kernel that combines multiple kernels by adding their outputs.

Implements: IKernelFunction<T>

For Beginners

The Sum kernel allows you to combine multiple kernels by adding their similarity scores together. This creates a more expressive model that can capture different types of patterns simultaneously.

In mathematical terms: k_sum(x, x’) = k1(x, x’) + k2(x, x’) + … + kn(x, x’)

How It Works

Why use Sum kernels?

  1. Multiple pattern types: Combine a smooth kernel (RBF) with a periodic kernel

to model data with both smooth trends and seasonal patterns.

  1. Additive structure: If your function can be decomposed into additive components,

each handled by a different kernel.

  1. Noise modeling: Add a WhiteNoise kernel to model observation noise.

Example combinations:

  • RBF + WhiteNoise: Smooth function with measurement noise
  • RBF + Periodic: Smooth trend with seasonal variation
  • Linear + RBF: Linear trend with non-linear deviations
  • RBF(short) + RBF(long): Capture both local and global patterns

Example

using AiDotNet.Kernels;
using AiDotNet.Tensors.LinearAlgebra;

var kernel = new SumKernel<double>();
var a = new Vector<double>(new[] { 1.0, 2.0, 3.0 });
var b = new Vector<double>(new[] { 1.5, 1.0, 2.5 });
double similarity = kernel.Calculate(a, b);
Console.WriteLine($"SumKernel: K(a, b) = {similarity:F4}");

Constructors

Constructor Summary
SumKernel(IKernelFunction<T>[] kernels) Initializes a new Sum kernel from an array of kernels.

Methods

Method Summary
IKernelFunction<T>[] GetKernels() Gets the component kernels in this sum.
T Calculate(Vector<T> x1, Vector<T> x2) Calculates the Sum kernel value between two vectors.