Braindecode Tutorials

Curated collection of 31 executable Jupyter notebooks showing how to use braindecode for end-to-end EEG / biosignal deep learning: data loading, preprocessing, model training, fine-tuning of foundation models, and Hugging Face Hub integration.

Each notebook is mirrored from the examples/ directory of the upstream repo. They are auto-converted from the sphinx-gallery .py source on every release; the canonical rendered versions live at https://braindecode.org/stable/auto_examples.

How to use this dataset repo

Three options, in increasing order of convenience:

Option	What you do
Open in Colab	Click any ▶ Colab link below. Free GPU runtime; no install.
Download	huggingface-cli download braindecode/tutorials --repo-type dataset --local-dir tutorials/
Browse rendered HTML	https://braindecode.org/stable/auto_examples (sphinx-gallery output).

Each notebook installs braindecode in the first cell, so it's runnable on a fresh kernel.

Catalogue

Getting started — datasets & I/O

Notebook	Topic	Run
plot_moabb_dataset_example.ipynb	Load a MOABB dataset (BNCI 2014-001)	▶ Colab
plot_mne_dataset_example.ipynb	Wrap an mne.io.Raw as a braindecode dataset	▶ Colab
plot_bids_dataset_example.ipynb	Load a BIDS-formatted EEG dataset	▶ Colab
plot_custom_dataset_example.ipynb	Roll your own dataset class	▶ Colab
plot_load_save_datasets.ipynb	Cache datasets to disk and reload them	▶ Colab
plot_split_dataset.ipynb	Train/valid/test splits at session and subject level	▶ Colab
plot_benchmark_preprocessing.ipynb	Compare preprocessing strategies head-to-head	▶ Colab
plot_tuh_discrete_multitarget.ipynb	TUH multi-target discrete labels	▶ Colab
plot_hub_integration.ipynb	Push / pull braindecode datasets to the Hub	▶ Colab

Model building & training

Notebook	Topic	Run
plot_basic_training_epochs.ipynb	Train any braindecode model in a few cells	▶ Colab
plot_train_in_pure_pytorch_and_pytorch_lightning.ipynb	Pure-PyTorch and Lightning training loops	▶ Colab
plot_bcic_iv_2a_moabb_trial.ipynb	BCI Competition IV 2a — trial-wise decoding	▶ Colab
plot_bcic_iv_2a_moabb_cropped.ipynb	BCI Competition IV 2a — cropped decoding	▶ Colab
plot_bcic_iv_2a_eegprep_cleaning.ipynb	EEG-Prep cleaning before training	▶ Colab
plot_load_pretrained_models.ipynb	Load BENDR / BIOT / EEGPT from the Hub and fine-tune	▶ Colab
plot_how_train_test_and_tune.ipynb	Cross-validation and hyper-parameter tuning	▶ Colab
plot_hyperparameter_tuning_with_scikit-learn.ipynb	scikit-learn GridSearchCV over braindecode	▶ Colab
plot_preprocessing_classes.ipynb	The Preprocessor API	▶ Colab
plot_channel_interpolation.ipynb	Bad-channel interpolation	▶ Colab
plot_regression.ipynb	EEG regression instead of classification	▶ Colab

Advanced training

Notebook	Topic	Run
plot_data_augmentation.ipynb	EEG-specific augmentations	▶ Colab
plot_data_augmentation_search.ipynb	Search over augmentation policies	▶ Colab
plot_relative_positioning.ipynb	Self-supervised pretext tasks	▶ Colab
plot_temporal_generalization.ipynb	Temporal-generalisation matrices	▶ Colab
plot_finetune_foundation_model.ipynb	Fine-tune a Hub foundation model end-to-end	▶ Colab
plot_moabb_benchmark.ipynb	Run a full MOABB benchmark	▶ Colab
plot_exca_config.ipynb	Reproducible experiments with exca	▶ Colab

Applied examples

Notebook	Topic	Run
plot_sleep_staging_chambon2018.ipynb	Sleep staging with Chambon2018	▶ Colab
plot_sleep_staging_usleep.ipynb	Sleep staging with U-Sleep	▶ Colab
plot_sleep_staging_eldele2021.ipynb	Sleep staging with Eldele2021	▶ Colab
plot_tuh_eeg_corpus.ipynb	Working with the TUH EEG corpus	▶ Colab

How the notebooks are produced

Sphinx-gallery automatically generates a .ipynb next to each rendered HTML page during the docs build. The conversion script for this dataset repo simply collects those generated notebooks:

# in the braindecode repo
cd docs && make html         # builds HTML + .ipynb files
python ../hf_assets/tutorials_index/build_notebooks.py  # gathers .ipynb

See build_notebooks.py for the gather + push logic.

Cost note

Hosting .ipynb files in a public dataset repo is free on Hugging Face. Compute (Colab) is also free for the small datasets used in these tutorials. No paid tier is required.

Citation

@article{aristimunha2025braindecode,
  title   = {Braindecode: a deep learning library for raw electrophysiological data},
  author  = {Aristimunha, Bruno and others},
  journal = {Zenodo},
  year    = {2025},
  doi     = {10.5281/zenodo.17699192},
}