API Reference

This reference is generated automatically from the physioex docstrings, so it always mirrors the installed code (the physioex-dev API). For the high-level picture of how these pieces fit together, see the library architecture pages.

physioex.data

The raw-EDF, lazy-loading dataset layer, the preprocessing pipeline, the on-disk cache, EDF/annotation readers and sleep-event helpers.

physioex.data.BasePhysioDataset	Abstract base class for lazy-loading EDF-based sleep datasets.
physioex.data.MultiDataset	Wraps multiple BasePhysioDataset instances into a single unified dataset.
physioex.data.SubjectSpec	Metadata describing how to locate a single subject's raw data.
physioex.data.PreprocessingStep	Base class for all preprocessing operations.
physioex.data.PreprocessingPipeline
physioex.data.CompiledPipeline
physioex.data.CompiledStep	Result of compiling a PreprocessingStep against a specific input fs.
physioex.data.Identity	Passthrough; does nothing.
physioex.data.BandpassFilter	IIR Butterworth bandpass filter (zero-phase via sosfiltfilt).
physioex.data.NotchFilter	IIR notch filter (powerline removal).
physioex.data.Resample	Resample to target sampling rate.
physioex.data.ZScoreNormalize	Per-subject per-channel z-score normalization.
physioex.data.XSleepNetSpectrogram	STFT spectrogram with power-dB scaling, matching Huy Phan's SeqSleepNet / XSleepNet / SleepTransformer preprocessing.
physioex.data.get_preset	Instantiate a preset by name.
physioex.data.available_presets	Return sorted list of available preset names.
physioex.data.ChannelCache
physioex.data.dict_collate_fn	Collate a list of dict-returning dataset items.
physioex.data.stack_channels	Convert a collated dict batch to a (B, L, C, ...) tensor for model input.
physioex.data.is_dict_batch	Return True if batch looks like a dict-style PhysioEx batch.
physioex.data.EDFHeader
physioex.data.ResolvedChannel
physioex.data.probe_edf_header	Read the EDF structural header only (no signal data).
physioex.data.resolve_channels	Resolve user channel requests against the EDF's available channels.
physioex.data.SleepEvent
physioex.data.map_events_to_epochs	Map a flat list of events to per-epoch event lists.
physioex.data.events_to_dicts	Convert a list of SleepEvent objects to a list of dicts.
physioex.data.dicts_to_events	Reconstruct SleepEvent objects from a list of dicts.

Datasets are resolved by name from the registry:

physioex.data.datasets.get_dataset
physioex.data.datasets.available_datasets

physioex.models

Foundation encoders (a uniform (B, L, C, T) -> (B, L, D) wrapper around heterogeneous pretrained backbones), embedding/pretrained helpers, and the classic sleep-staging architectures.

physioex.models.FoundationEncoder	Abstract adapter: frozen encoder that returns embeddings.
physioex.models.CBraModEncoder	CBraMod accepts any number of EEG channels, pools over them.
physioex.models.BENDREncoder	BENDR maps input channels to a fixed 20-channel 10-20 layout.
physioex.models.LaBraMEncoder	LaBraM with channel-aware position embeddings; lazy model build.
physioex.models.BIOTEncoder	BIOT with 16-channel double-banana layout + q95 normalization.
physioex.models.SleepFMEncoder	SleepFM pads input to 10 channel slots and masks unused ones.
physioex.models.TFCEncoder	TF-C: single-channel time + frequency transformer.
physioex.models.REVEEncoder	REVE with position bank for channel-aware embeddings.
physioex.models.SJEEncoder	Signal-JEPA with fixed 19-channel 10-20 layout.
physioex.models.NeuroLMEncoder	NeuroLM with patch-based tokenization and VQ channel vocabulary.
physioex.models.load_from_pretrained	Load a pretrained model from HuggingFace Hub.
physioex.models.extract_embeddings	Extract and cache contextualized embeddings for all subjects.
physioex.models.load_embeddings	Load cached embeddings, downloading from HuggingFace if needed.
physioex.models.linear_probe	5-fold subject-wise cross-validated linear probe on cached embeddings.

Classic architectures:

physioex.models.chambon2018.Chambon2018Net	Chambon et al. (2018) sleep staging network.
physioex.models.tinysleepnet.TinySleepNet
physioex.models.seqsleepnet.SeqSleepNet
physioex.models.sleeptransformer.SleepTransformer	SleepTransformer (Phan et al. 2022).

physioex.train

A hand-written PyTorch training/evaluation loop (not Lightning), with metric and statistics helpers, pluggable logging, and the three console-script entry points.

physioex.train.trainer.Trainer
physioex.train.trainer.seed_everything	Seed all relevant RNGs for reproducibility.
physioex.train.multidevicetrainer.Trainer
physioex.train.logger.Logger	Backend-agnostic experiment logger.
physioex.train.logger.NoOpLogger	A logger that discards everything.
physioex.train.logger.build_logger	Instantiate a logger backend.
physioex.train.models.load.load_model	Load a model with dual-format checkpoint support.

Metrics, statistics and progress helpers are documented at module level:

physioex.train.metrics
physioex.train.stats	Per-subject metric aggregation (mean / std / confidence interval).
physioex.train.progress

physioex.explain

Post-hoc attribution, the foundational Conceptual Spectral Decomposition (CSD), and prototype discovery.

physioex.explain.foundational	Explainability methods for foundation model embeddings.
physioex.explain.prototypes.posthoc