Schöffelen 2019 (MOUS)

pnpl.datasets.Schoffelen2019 wraps the “Mother of Unification Studies” (MOUS) sentence-comprehension MEG dataset (Schöffelen et al., 2019) released on the Radboud Data Repository (DSC_3011020.09_236_v1).

• ~200 participants, two cohorts:

  • subjects with the A* prefix performed the auditory task

  • subjects with the V* prefix performed the visual (reading) task

• every participant performed a shared rest block

• CTF MEG, raw .ds directories

• ~2 GB raw CTF per task

Warning

MOUS is not open access. You need an approved data-sharing agreement with the dataset owner before you can download.

Auth

export RADBOUD_USERNAME="you@orcid.org"
export RADBOUD_PASSWORD="..."

Quickstart

from pnpl.datasets import Schoffelen2019
from pnpl.tasks.schoffelen2019 import TrialEpoching

ds = Schoffelen2019(
    data_path="./data/schoffelen",
    task=TrialEpoching(tmin=0.0, tmax=1.0, label_type="trigger"),
    include_subjects=["A2002"],
    include_tasks=["auditory"],
    preprocessing="notch+bp+ds",
    download=True,
    standardize=True,
)

x, y = ds[0]
print(x.shape, y.item())

A good starting point is a single A* (or V*) subject with the rest task — rest is smaller and needs no stimulus alignment.

Layout differences from the other datasets

MOUS doesn’t have a session axis. The BIDS layout is sub-XXXX/meg/sub-XXXX_task-{auditory,visual,rest}_meg.ds, so the loader synthesizes a constant session = "01" for the standard (subject, session, task, run) 4-tuple convention. It also auto-skips tasks the subject didn’t perform (auditory for V*, visual for A*).

BIDS axes

Axis	Values
subject	A2002–A2125 (auditory cohort), V1001–V1117 (visual cohort)
session	always "01" (synthesized)
task	"auditory", "visual", "rest" (gated per-subject)
run	always "01"

Selected arguments

• task — currently pnpl.tasks.schoffelen2019.TrialEpoching.

• preprocessing, preprocessing_config — see Preprocessing.

• include_subjects, include_tasks, include_run_keys (and their exclude_* counterparts).

• standardize, clipping_boundary, channel_means, channel_stds.

• create_h5_if_missing, preload_h5.

Available tasks

TrialEpoching(tmin, tmax, label_type) epochs around each trial onset (rows whose type is "trial" in events.tsv).

• label_type="trigger" (default) — label each trial with the first UPPT001 trigger code that appears inside it.

• label_type="binary" — constant label of 1 (useful for self-supervised windowing without conditional labels).