SelfSupervisedLearningSession<T>

Models & Types · AiDotNet.SelfSupervisedLearning

Manages a self-supervised learning training session.

For Beginners

An SSL session manages the entire training lifecycle: initialization, training loop, evaluation, and checkpointing. It provides callbacks for monitoring progress and supports resuming from checkpoints.

Constructors

Constructor Summary
SelfSupervisedLearningSession(ISelfSupervisedLearningMethod<T> method, SelfSupervisedLearningConfig<T> config) Initializes a new SSL training session.

Properties

Property Summary
ISelfSupervisedLearningMethod<T> Method { get; } Gets the SSL method being used.
bool IsDistributed { get; } Gets whether this session is using distributed training.
bool IsTraining { get; } Gets whether training is in progress.
int CurrentEpoch { get; } Gets the current epoch number.
int GlobalStep { get; } Gets the global step counter.
int Rank { get; } Gets the rank of this worker in distributed training.
int WorldSize { get; } Gets the world size (number of workers) for distributed training.

Methods

Method Summary
SelfSupervisedLearningMetricReport<T> Evaluate(Tensor<T> data) Runs evaluation on the current encoder.
SelfSupervisedLearningResult<T> Train(Func<IEnumerable<Tensor<T>>> dataLoader, Tensor<T> validationData, int[] validationLabels) Trains the SSL method for the specified number of epochs.
SelfSupervisedLearningSession<T> FromCheckpoint(string checkpointPath, INeuralNetwork<T> encoder, Func<INeuralNetwork<T>, ISelfSupervisedLearningMethod<T>> methodFactory) Creates a session from a pretrained checkpoint.
SelfSupervisedLearningTrainingHistory<T> GetHistory() Gets the current training history.
int GetEffectiveBatchSize() Gets the effective batch size considering distributed training.
void CacheTrainingFeaturesForKNN(IEnumerable<Tensor<T>> trainingData, int[] trainingLabels) Caches training data for k-NN evaluation.
void Reset() Resets the session for a new training run.
void SaveCheckpoint(string checkpointPath) Saves a checkpoint to disk.
void Stop() Stops training gracefully.
void SynchronizeParameters() Synchronizes model parameters across all distributed workers.