Which recordings belong to an EEG2025 mini release?

Estimated reading time:4 minutes

Inspect the R5 challenge catalogue without reading signal payloads. The mini list defines eligible participants; catalogue recording counts depend on task availability. Challenge data are 100 Hz, 0.5–50 Hz filtered derivatives and must not be confused with the original HBN OpenNeuro recordings.

Before you start

Use an installed EEGDash environment and an internet connection to the catalogue. This page requests metadata only; it needs neither a GPU nor a signal download. EEGDASH_CACHE_DIR overrides the shared EEGDash cache resolver. Reading a recording’s .raw later is a separate step that acquires its signal and sidecars.

The question is whether a proposed cohort contains the participants and tasks you expect. There is no prediction target or train/test score on this page. p_factor is requested to inspect available participant metadata, not to construct a new target or infer that every participant has a valid value.

from eegdash.paths import get_default_cache_dir

from eegdash import EEGChallengeDataset
from eegdash.const import SUBJECT_MINI_RELEASE_MAP

Select the challenge release

release="R5" selects the challenge’s release mapping; mini=True restricts eligibility to its curated participant list. No task filter is used, so several tasks and runs may belong to the same person. The challenge loader also selects the preprocessed challenge storage location. Replacing it with an OpenNeuro query changes the data product even when participant IDs match.

description_fields requests fields useful for checking cohort eligibility. The description can contain additional source metadata. Missing age, sex or p-factor values require an explicit exclusion policy in a later supervised analysis; this discovery step deliberately does not silently drop records.

dataset = EEGChallengeDataset(
    release="R5",
    mini=True,
    cache_dir=get_default_cache_dir(),
    description_fields=["subject", "task", "run", "age", "sex", "p_factor"],
)

╭────────────────────── EEG 2025 Competition Data Notice ──────────────────────╮
│ This object loads the HBN dataset that has been preprocessed for the EEG     │
│ Challenge:                                                                   │
│   * Downsampled from 500Hz to 100Hz                                          │
│   * Bandpass filtered (0.5-50 Hz)                                            │
│                                                                              │
│ For full preprocessing applied for competition details, see:                 │
│   https://github.com/eeg2025/downsample-datasets                             │
│                                                                              │
│ The HBN dataset have some preprocessing applied by the HBN team:             │
│   * Re-reference (Cz Channel)                                                │
│                                                                              │
│ IMPORTANT: The data accessed via `EEGChallengeDataset` is NOT identical to   │
│ what you get from EEGDashDataset directly.                                   │
│ If you are participating in the competition, always use                      │
│ `EEGChallengeDataset` to ensure consistency with the challenge data.         │
╰──────────────────────── Source: EEGChallengeDataset ─────────────────────────╯
[10/16/26 15:55:35] INFO     Auto-corrected misrouted             dataset.py:562
                             storage.base for dataset
                             EEG2025r5mini: None ->
                             s3://nemar/EEG2025r5mini

Count participants separately from recordings

A row of description describes a recording, not an independent person. The eligibility list, unique matched subjects, and number of recording objects therefore answer different questions. Grouping by task exposes repeated runs: recordings can exceed participants without indicating duplicate data.

Not every eligible participant has every task. Inspect the table before choosing a bounded cohort for the next tutorial.

metadata = dataset.description
print("Eligible mini participants:", len(SUBJECT_MINI_RELEASE_MAP["R5"]))
print("Matched participants:", metadata.subject.nunique())
print("Matched recordings:", len(dataset.datasets))
print(
    metadata.groupby("task").agg(
        recordings=("subject", "size"), participants=("subject", "nunique")
    )
)

Eligible mini participants: 20
Matched participants: 20
Matched recordings: 240
                         recordings  participants
task
DespicableMe                     20            20
DiaryOfAWimpyKid                 20            20
FunwithFractals                  20            20
RestingState                     20            20
ThePresent                       20            20
contrastChangeDetection          60            20
seqLearning6target                6             6
seqLearning8target               14            14
surroundSupp                     40            20
symbolSearch                     20            20

Select participants explicitly before accessing .raw: the constructor above discovers metadata, whereas .raw triggers acquisition of signal payloads.

print("First recording provenance:", dataset.records[0]["bids_relpath"])

First recording provenance: sub-NDARAH793FBF/eeg/sub-NDARAH793FBF_task-symbolSearch_eeg.bdf

Use the result to define a bounded cohort

The printed bids_relpath identifies a concrete source recording. Keep that path, release, subject and task with any later window metadata. A model’s sample count should come from its windows; a cohort’s participant count should come from unique subject identifiers.

Next, choose three listed subjects and a single task, add those filters to the constructor, and inspect their annotations through .raw. Estimate the resulting download before widening the query. For regression, check observed p-factor availability before feature extraction; for reaction time, retain observed stimulus and response events. This metadata table alone cannot establish either model performance or data quality.

Related data-loading example: Braindecode BIDS Dataset Example.