NeuroAtlas documentation¶

NeuroAtlas is a Python package and command-line tool, neuroatlas, for benchmarking EEG foundation models. It downloads the datasets and model weights, extracts embeddings with a frozen model, fits a probe on every fold and reports the results. It covers 12 benchmarks in sleep, epilepsy, brain age and brain-computer interfaces, with 42 datasets and 44 model checkpoints.

Installation¶

Install the package from PyPI (Python 3.10 or newer, Linux):

pip install neuroatlas-bench

Installation covers conda and venv, GPUs, and the extra packages for BCI datasets and time-series models.

A first run¶

Choose a project folder, download one dataset and one model, and run a benchmark:

neuroatlas config init ~/neuroatlas
neuroatlas data download sleep_edf_expanded --mirror aws
neuroatlas models download biot_pretrained
neuroatlas run sleep_stage -m biot_pretrained --debug
neuroatlas results sleep_stage

Contents¶

Getting started: Installation, a quick start, and a walkthrough of every command on open data.
User guide: Configuration, data, models, the benchmarks, running them locally or on a cluster, and reading the results.
Command reference: Every neuroatlas command with its options and examples.
Citation: How to cite NeuroAtlas in your work.