Example index and related work

The examples run as part of this site, see the gallery.

Every example builds a system with known ground truth, runs pyPTE, and asserts the result. If an assertion fails, the claim it makes is no longer true.

uv sync --group examples
uv run python -m examples.two_node_coupling      # add --quick for a faster run

example	what it establishes
two_node_coupling	direction is recovered; SNR asymmetry fakes it
kuramoto_network	known directed graphs recovered, AUC 0.96–1.00
neural_mass_network	recording length vs precision, on Jansen-Rit columns
significance_testing	what a surrogate test does, drawn out
epoched_analysis	full epoched two-condition pipeline
tvb_connectome	a real directed connectome, where recovery mostly fails

See examples/README.md for details.

Related work

The example models here are self-contained on purpose, so they run in CI without extra dependencies. If you want richer or more citable ground-truth signals, these are the simulators worth knowing about:

• The Virtual Brain (tvb-library, GPL-3.0-or-later) — the reference whole-brain platform, and the one worth reaching for first. It ships a genuinely directed 76-region connectome from tract-tracing rather than a symmetric DTI matrix, which is what a directional measure needs as ground truth, plus ~30 population models and EEG/MEG forward monitors. Sanz Leon et al., Front. Neuroinform. 7:10 (2013). → examples/tvb_connectome.py runs pyPTE against that connectome, and is the most informative example here because the result is largely negative.

• PyRates (GPL-3.0) — a much smaller dependency footprint, with explicit per-edge directed coupling and delays. No bundled connectome, so you supply the topology, which is what the self-contained models here already do; the gain over them is citability rather than capability. Gast et al., PLOS ONE 14(12):e0225900 (2019).

• neurolib (MIT) — whole-brain modelling with several neural mass models. No example here, deliberately: its bundled HCP connectome is symmetric, so it carries no net direction for a directional measure to recover, and the project has had little activity since December 2024.

• brainmass (Apache-2.0), part of the BrainX ecosystem — differentiable neural mass models on JAX. No example here, deliberately: at v0.1.1 the API is still moving, and it pulls in the whole JAX stack for a demonstration the models above already provide.

pyPTE depends on none of them.