Class SVMOptions<T>

Namespace

AiDotNet.Models.Options

Assembly

AiDotNet.dll

Configuration options for Support Vector Machine classifiers.

public class SVMOptions<T> : ClassifierOptions<T>

Type Parameters

T

The data type used for calculations.

Inheritance

object

ModelOptions

ClassifierOptions<T>

SVMOptions<T>

Inherited Members

ClassifierOptions<T>.TaskType

ClassifierOptions<T>.DecisionThreshold

ClassifierOptions<T>.UseClassWeights

ClassifierOptions<T>.ClassWeights

ModelOptions.Seed

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

Support Vector Machines (SVMs) find the optimal hyperplane that maximizes the margin between classes. They are particularly effective in high-dimensional spaces and can handle non-linear classification using kernel functions.

For Beginners: SVMs are like drawing the best possible line between classes!

Imagine you have red and blue dots on paper. You want to draw a line that:

1. Separates the red dots from the blue dots
2. Stays as far as possible from both sets of dots

The "margin" is the gap between the line and the nearest dots on each side. SVMs find the line that maximizes this margin.

Key concepts:

• C parameter: How much to penalize misclassifications (higher = stricter)
• Kernel: How to measure similarity between points (linear, RBF, polynomial)
• Support vectors: The dots closest to the line that define its position

SVMs work great when:

• You have clear separation between classes
• You have many features but not tons of samples
• You need a robust classifier

Properties

C

Gets or sets the regularization parameter C.

public double C { get; set; }

Property Value

double

The regularization strength. Default is 1.0.

Remarks

C controls the trade-off between achieving a low training error and a low testing error (generalization). Higher C means less regularization and the model will try harder to classify all training points correctly.

For Beginners: C controls how strictly the SVM fits the training data.

• C = 0.01: Very flexible boundary, allows many misclassifications
• C = 1.0: Balanced (default), good starting point
• C = 100: Very strict, tries to classify all training points correctly

If your model overfits (high training accuracy, low test accuracy), try lower C. If your model underfits (low training accuracy), try higher C.

CacheSize

Gets or sets the cache size in MB for kernel computations.

public double CacheSize { get; set; }

Property Value

double

The cache size in megabytes. Default is 200.

Coef0

Gets or sets the independent term (coef0) in kernel function.

public double Coef0 { get; set; }

Property Value

double

The coefficient. Default is 0.0.

Remarks

Used in polynomial and sigmoid kernels.

Degree

Gets or sets the degree for polynomial kernel.

public int Degree { get; set; }

Property Value

int

The polynomial degree. Default is 3.

Remarks

Only used when Kernel is Polynomial. Higher degrees can model more complex relationships but may overfit.

Gamma

Gets or sets the kernel coefficient gamma.

public double? Gamma { get; set; }

Property Value

double?

The gamma value, or null for "scale" (1 / (n_features * variance)). Default is null.

Remarks

Gamma defines how far the influence of a single training example reaches. Low gamma means 'far' (smoother decision boundary), high gamma means 'close' (more complex, wiggly boundary).

For Beginners: Gamma controls the "reach" of each training point.

• Low gamma (0.001): Each point influences a large area (smooth boundary)
• High gamma (10.0): Each point only influences nearby points (complex boundary)

If null, gamma is set automatically based on your data. Too high gamma = overfitting Too low gamma = underfitting

Kernel

Gets or sets the kernel type.

public KernelType Kernel { get; set; }

Property Value

KernelType

The kernel function to use. Default is RBF.

Remarks

The kernel determines how similarity between samples is computed. Different kernels can capture different types of relationships.

For Beginners: Kernels let SVMs work with non-linear boundaries!

• Linear: Straight line/plane separation (fast, works when data is linearly separable)
• RBF (Radial Basis Function): Flexible curves (most popular, handles complex patterns)
• Polynomial: Curved boundaries with polynomial degree
• Sigmoid: S-shaped curves (similar to neural networks)

Start with RBF - it works well for most problems. Use Linear if you have many features or need speed.

MaxIterations

Gets or sets the maximum number of iterations.

public int MaxIterations { get; set; }

Property Value

int

The maximum iterations, or -1 for unlimited. Default is 1000.

OneVsRest

Gets or sets whether to use one-vs-rest or one-vs-one for multi-class.

public bool OneVsRest { get; set; }

Property Value

bool

True for one-vs-rest, false for one-vs-one. Default is false (OvO).

Remarks

For multi-class classification: - One-vs-One (OvO): Train n*(n-1)/2 classifiers, one for each pair of classes - One-vs-Rest (OvR): Train n classifiers, each separating one class from all others

For Beginners: How to handle more than 2 classes?

OvO (default):

• Creates classifiers for every pair of classes
• More classifiers but each is simpler
• Often works better in practice

OvR:

• Creates one classifier per class
• Each classifier separates one class from "everything else"
• Fewer classifiers but each may be harder to train

Probability

Gets or sets whether to calculate probability estimates.

public bool Probability { get; set; }

Property Value

bool

True to enable probability estimation. Default is false.

Remarks

When true, the SVM will be trained to output probability estimates. This uses Platt scaling and requires extra cross-validation during training.

For Beginners: Should the SVM output probabilities?

With Probability = true:

• PredictProbabilities() returns actual probabilities (0.0 to 1.0)
• Training takes longer (needs calibration)

With Probability = false:

• PredictProbabilities() returns estimates based on decision function
• Faster training

Enable this if you need calibrated probability outputs.

RandomState

Gets or sets the random state for probability estimation.

public int? RandomState { get; set; }

Property Value

int?

The random seed, or null for non-deterministic. Default is null.

Shrinking

Gets or sets whether to use shrinking heuristic.

public bool Shrinking { get; set; }

Property Value

bool

True to use shrinking (default), false otherwise.

Remarks

Shrinking can speed up training by focusing on active support vectors. It usually helps but can sometimes slow things down.

Tolerance

Gets or sets the tolerance for stopping criterion.

public double Tolerance { get; set; }

Property Value

double

The tolerance. Default is 0.001.

Remarks

The solver iterates until the change in the objective function is below this threshold. Lower values mean more precise solutions but longer training.