API reference

Generated from the source by mkdocstrings, so a signature or docstring here is the one in ecgbench/ at the commit the site was built from.

Import surface

Everything below is re-exported from the top-level package:

from ecgbench import ECGDataset, load_config, run_splits

Two things about that top level are worth knowing before you read the rest.

The package root is half lazy, on purpose

ecgbench/__init__.py imports the catalogue and config eagerly — they are pure-Python and cheap — and resolves everything else through a module-level __getattr__ against _LAZY_IMPORTS. So import ecgbench never pulls in torch, wfdb or mlcroissant; those arrive only when you touch a name that needs them.

A consequence for this reference: the pages below document each name at the module that defines it (ecgbench.dataset.ECGDataset), because a lazy attribute has no static definition at the package root to document. Both import paths work at runtime.

Adding a public symbol that needs a heavy dependency means adding it to _LAZY_IMPORTS and __all__, never to the imports at the top of the file.

Config slugs, not catalogue slugs

Every function here that takes a dataset argument wants the underscored config slug — ptbxl, the name of the YAML in ecgbench/data/configs/ — not the dashed catalogue slug ptb-xl. See the two namespaces.

Where to start

Doing	Start at
Loading records for training	ECGDataset
Looking up what exists	metadata (merged view, any alias) or catalogue (front matter only)
Describing a new dataset	DatasetConfig
Running the whole pipeline	run_splits
Checking a split is the canonical one	verify_splits