braindecode/example_dataset-eegwindows

EEG Dataset

This dataset was created using braindecode, a deep learning library for EEG/MEG/ECoG signals.

Dataset Information

Property	Value
Recordings	1
Type	Windowed (from Raw object)
Channels	26
Sampling frequency	250 Hz
Total duration	0:06:26
Windows/samples	48
Size	19.22 MB
Format	zarr

Quick Start

from braindecode.datasets import BaseConcatDataset

# Load from Hugging Face Hub
dataset = BaseConcatDataset.pull_from_hub("username/dataset-name")

# Access a sample
X, y, metainfo = dataset[0]
# X: EEG data [n_channels, n_times]
# y: target label
# metainfo: window indices

Training with PyTorch

from torch.utils.data import DataLoader

loader = DataLoader(dataset, batch_size=32, shuffle=True, num_workers=4)

for X, y, metainfo in loader:
    # X: [batch_size, n_channels, n_times]
    # y: [batch_size]
    pass  # Your training code

BIDS-inspired Structure

This dataset uses a BIDS-inspired organization. Metadata files follow BIDS conventions, while data is stored in Zarr format for efficient deep learning.

BIDS-style metadata:

• dataset_description.json - Dataset information
• participants.tsv - Subject metadata
• *_events.tsv - Trial/window events
• *_channels.tsv - Channel information
• *_eeg.json - Recording parameters

Data storage:

• dataset.zarr/ - Zarr format (optimized for random access)

sourcedata/braindecode/
├── dataset_description.json
├── participants.tsv
├── dataset.zarr/
└── sub-<label>/
    └── eeg/
        ├── *_events.tsv
        ├── *_channels.tsv
        └── *_eeg.json

Accessing Metadata

# Participants info
if hasattr(dataset, "participants"):
    print(dataset.participants)

# Events for a recording
if hasattr(dataset.datasets[0], "bids_events"):
    print(dataset.datasets[0].bids_events)

# Channel info
if hasattr(dataset.datasets[0], "bids_channels"):
    print(dataset.datasets[0].bids_channels)

---

Created with braindecode