Class SSLPretrainingPipeline<T>

Namespace

AiDotNet.SelfSupervisedLearning

Assembly

AiDotNet.dll

High-level pipeline for SSL pretraining.

public class SSLPretrainingPipeline<T>

Type Parameters

T

The numeric type used for computations.

Inheritance

object

SSLPretrainingPipeline<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

For Beginners: This pipeline provides a simple, high-level interface for SSL pretraining. Just provide your encoder and data, and it handles the rest: method selection, augmentation, training loop, and evaluation.

Example usage:

var pipeline = new SSLPretrainingPipeline<double>(encoder)
    .WithMethod(SSLMethodType.SimCLR)
    .WithConfig(config => config.PretrainingEpochs = 100);

var result = pipeline.Train(dataLoader);

Constructors

SSLPretrainingPipeline(INeuralNetwork<T>, int)

Initializes a new SSL pretraining pipeline.

public SSLPretrainingPipeline(INeuralNetwork<T> encoder, int encoderOutputDim)

Parameters

encoder INeuralNetwork<T>

The encoder network to pretrain.

encoderOutputDim int

Output dimension of the encoder.

Methods

Train(Func<IEnumerable<Tensor<T>>>, Tensor<T>?, int[]?)

Trains the encoder using SSL.

public SSLResult<T> Train(Func<IEnumerable<Tensor<T>>> dataLoader, Tensor<T>? validationData = null, int[]? validationLabels = null)

Parameters

dataLoader Func<IEnumerable<Tensor<T>>>

Function that yields batches of unlabeled data.

validationData Tensor<T>

Optional validation data for monitoring.

validationLabels int[]

Optional validation labels for k-NN evaluation.

Returns

SSLResult<T>

Training result with pretrained encoder.

WithConfig(Action<SSLConfig>)

Configures the training parameters.

public SSLPretrainingPipeline<T> WithConfig(Action<SSLConfig> configure)

Parameters

configure Action<SSLConfig>

Returns

SSLPretrainingPipeline<T>

WithEncoderCopyFactory(Func<INeuralNetwork<T>, INeuralNetwork<T>>)

Sets the function to create encoder copies (for momentum methods).

public SSLPretrainingPipeline<T> WithEncoderCopyFactory(Func<INeuralNetwork<T>, INeuralNetwork<T>> createCopy)

Parameters

createCopy Func<INeuralNetwork<T>, INeuralNetwork<T>>

Returns

SSLPretrainingPipeline<T>

WithMethod(SSLMethodType)

Sets the SSL method to use.

public SSLPretrainingPipeline<T> WithMethod(SSLMethodType method)

Parameters

method SSLMethodType

Returns

SSLPretrainingPipeline<T>

Events

OnProgress

Event raised during training for progress updates.

public event Action<int, int, T>? OnProgress

Event Type

Action<int, int, T>