Table of Contents

Interface IClassifier<T>

Namespace
AiDotNet.Interfaces
Assembly
AiDotNet.dll

Defines the common interface for all classification algorithms in the AiDotNet library.

public interface IClassifier<T> : IFullModel<T, Matrix<T>, Vector<T>>, IModel<Matrix<T>, Vector<T>, ModelMetadata<T>>, IModelSerializer, ICheckpointableModel, IParameterizable<T, Matrix<T>, Vector<T>>, IFeatureAware, IFeatureImportance<T>, ICloneable<IFullModel<T, Matrix<T>, Vector<T>>>, IGradientComputable<T, Matrix<T>, Vector<T>>, IJitCompilable<T>

Type Parameters

T

The numeric type used for calculations (e.g., float, double).

Inherited Members
Extension Methods

Remarks

Classification is a type of machine learning algorithm used to predict categorical values (discrete classes) rather than continuous values. This interface extends IFullModel with classification-specific functionality.

For Beginners: Classification is about putting things into categories.

For example, classification can be used to:

  • Predict whether an email is spam or not spam (binary classification)
  • Identify handwritten digits (0-9) from images (multi-class classification)
  • Determine which diseases a patient might have (multi-label classification)
  • Rate customer satisfaction as Poor/Fair/Good/Excellent (ordinal classification)

Unlike regression algorithms (which predict numbers), classification algorithms predict which category or categories something belongs to.

Properties

ClassLabels

Gets the class labels learned during training.

Vector<T>? ClassLabels { get; }

Property Value

Vector<T>

A vector containing the unique class labels, or null if not yet trained. The index corresponds to the class index used in predictions.

Remarks

Class labels provide the mapping between numeric class indices (0, 1, 2, ...) and the original label values from the training data. This is useful for interpreting predictions.

For Beginners: When you train a classifier, it learns what categories exist.

For example, if you train on sentiment data with labels "positive", "neutral", "negative":

  • The model internally uses indices: 0, 1, 2
  • ClassLabels stores the mapping: [positive, neutral, negative]
  • When the model predicts class 2, you can look up that it means "negative"

This is especially important when your original labels aren't already 0, 1, 2, etc.

NumClasses

Gets the number of classes that this classifier can predict.

int NumClasses { get; }

Property Value

int

The total number of distinct classes in the classification problem.

Remarks

For binary classification, this returns 2. For multi-class classification, this returns the number of distinct classes learned during training.

For Beginners: This tells you how many different categories the model can predict.

Examples:

  • Spam detection: 2 classes (spam, not spam)
  • Digit recognition: 10 classes (0-9)
  • Sentiment analysis: 3 classes (negative, neutral, positive)

The number of classes is determined during training based on the target labels.

TaskType

Gets the type of classification task this classifier is configured for.

ClassificationTaskType TaskType { get; }

Property Value

ClassificationTaskType

A ClassificationTaskType value indicating whether this is binary, multi-class, multi-label, or ordinal classification.

Remarks

The task type affects how predictions are interpreted and which loss functions and metrics are appropriate for the model.

For Beginners: The task type tells you what kind of classification problem this model is designed to solve.

  • Binary: Two categories (yes/no, spam/not-spam)
  • MultiClass: Multiple exclusive categories (exactly one answer)
  • MultiLabel: Multiple overlapping categories (can have multiple answers)
  • Ordinal: Ordered categories (like ratings or satisfaction levels)

Knowing the task type helps you understand what kind of output to expect.

Edit this page