stable_pretraining.callbacks

The callbacks module provides various monitoring and evaluation tools for self-supervised learning training.

Online Monitoring

OnlineProbe(name, input, target, probe, loss_fn)	Online probe for evaluating learned representations during self-supervised training.
OnlineKNN(name, input, target, queue_length, ...)	Weighted K-Nearest Neighbors online evaluator using queue discovery.
OnlineWriter(names, path, during[, ...])	Writes specified batch data to disk during training and validation.
RankMe(name, target, queue_length, target_shape)	RankMe (effective rank) monitor using queue discovery.
LiDAR(name, target, queue_length, target_shape)	LiDAR (Linear Discriminant Analysis Rank) monitor using queue discovery.

Training Utilities

EarlyStopping([mode, milestones, ...])	Early stopping mechanism with support for metric milestones and patience.
TrainerInfo()	Links the trainer to the DataModule for enhanced functionality.
LoggingCallback()	Displays validation metrics in a color-coded formatted table.
ModuleSummary()	Logs detailed module parameter statistics in a formatted table.

Model Persistence

SklearnCheckpoint()

Callback for saving and loading sklearn models in PyTorch Lightning checkpoints.

Evaluation

ImageRetrieval(pl_module, name, input, ...)

Image Retrieval evaluator for self-supervised learning.