braindecode/braindecode

Braindecode – Deep‑learning toolbox for brain‑signal data

What it is – Braindecode is an open‑source Python library that lets researchers apply modern deep‑learning models directly to raw electrophysiological recordings such as EEG, ECoG, and MEG. It bundles data‑fetching utilities, preprocessing pipelines, visualisation helpers, and a collection of ready‑made neural‑network architectures (e.g., convolutional nets) plus data‑augmentation tricks that are useful for brain‑signal decoding.

Who it’s for –

• Neuroscientists who want to experiment with deep learning without building the whole ML stack from scratch.
• Machine‑learning researchers interested in neuro‑physiological datasets.

Key components

• Dataset fetchers – one‑click download of popular EEG benchmarks (via the MOABB library).
• Pre‑processing & augmentation – functions for filtering, epoching, normalising, and augmenting raw signals.
• Model zoo – implementations of several state‑of‑the‑art architectures (e.g., ShallowConvNet, Deep4Net) that have been validated on EEG decoding tasks.
• Visualization – tools to plot raw signals, model predictions, and learned feature maps.
• Integration with MNE‑Python – leverages the widely‑used MNE ecosystem for handling neuro‑imaging data.

Installation

# 1. Install PyTorch (no torchvision needed)
#    Follow the instructions at https://pytorch.org/

# 2. Optional: install MOABB if you need benchmark EEG datasets
pip install moabb

# 3. Install Braindecode itself
pip install braindecode

For the latest development version, see the project’s CONTRIBUTING guide.

Documentation & support – Full API docs, tutorials, and examples are hosted at https://braindecode.org (stable and development builds). The repository also includes CI badges for docs, tests, and code coverage, indicating active maintenance.

Citation – If you use the library in research, cite the Zenodo DOI (10.5281/zenodo.17699192) and the original 2017 Human Brain Mapping paper that introduced the core EEG‑CNN model.

License – BSD‑3‑Clause for the core code, with some auxiliary files under CC‑BY‑NC, MIT, or Apache‑2.0.