physioex.models.extract_embeddings

physioex.models.extract_embeddings(model, dataset, model_name, dataset_name, L, device='cpu', overwrite=False, upload=False, cache_dir=None)

Extract and cache contextualized embeddings for all subjects.

For each subject in the dataset, extracts per-epoch embeddings using sliding-window encoding (same as voting evaluation) and saves them to disk. Optionally uploads to HuggingFace Hub.

Parameters:

• model (Module) – Model with encode(x) -> (B, L, D) method.

• dataset – A BasePhysioDataset instance.

• model_name (str) – Identifier for cache directory (e.g. "seqsleepnet-phan").

• dataset_name (str) – Dataset name for cache (e.g. "sleepedf").

• L (int) – Sequence length the model was trained with.

• device (str) – Device string ("cpu" or "cuda:0").

• overwrite (bool) – If True, re-extract even if cached.

• upload (bool) – If True, upload all embeddings to HuggingFace Hub.

• cache_dir (str | None) – Override cache root directory.

Returns:

Path to the embeddings directory.

Return type:

Path