Datasets:

windwerfer
/

neurofeed-eeg-windows

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
corpus: string
split: string
policy: string
subjects: list<item: string>
  child 0, item: string
created_utc: string
note: string
montage_id: string
parallel_to_muse4: string
updated_utc: string
updated_by: string
n: int64
to
{'split': Value('string'), 'subjects': List(Value('string')), 'n': Value('int64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              corpus: string
              split: string
              policy: string
              subjects: list<item: string>
                child 0, item: string
              created_utc: string
              note: string
              montage_id: string
              parallel_to_muse4: string
              updated_utc: string
              updated_by: string
              n: int64
              to
              {'split': Value('string'), 'subjects': List(Value('string')), 'n': Value('int64')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Neurofeed EEG Windows

Derived 2-second @ 256 Hz EEG windows for training and evaluating Muse- and Crown-oriented models. This is not raw clinical EEG, and it is not a mirror of OpenNeuro or PhysioNet recordings.

Repo: windwerfer/neurofeed-eeg-windows

What you get

Each config folder holds subject/session packs as NumPy .npz files (plus matching manifests, small QC npz where present, and subject-wise split JSON). There is no Arrow/parquet table export — load the npz files directly.

Config	Montage / shape	Subjects (train/val/test)	Windows (approx.)	Status
`muse4_vigilance_sleep_edf`	Muse4 `(N,4,512)`	124 subjects / 125 nights · 86 / 19 / 19	~519k	primary release candidate
`muse4_attention_ds001787`	Muse4 proxy `(N,4,512)`	16 · 12 / 2 / 2	~5.2k	research / reproducibility (`ship_candidate: false`)
`muse4_attention_ds003969`	Muse4 proxy `(N,4,512)`	64 · 60 / 2 / 2	~51k	research / reproducibility (`ship_candidate: false`)
`crown8_attention_ds001787`	Crown8 `(N,8,512)`	16 · 12 / 2 / 2	~5.2k	research / reproducibility (`ship_candidate: false`)
`crown8_attention_ds003969`	Crown8 `(N,8,512)`	64 · 60 / 2 / 2	~51k	research / reproducibility (`ship_candidate: false`)
`muse4_engagement_a_eng`	Muse4 proxy `(N,4,512)`	~133 persons / 150 packs	~19.7k	research only; not a shipping head
`crown2_vigilance_hmc`	Crown2 `(N,2,512)` C3/C4	151 · 106 / 23 / 22	~453k	Crown vig ship proxy (HMC-only; CBraMod test 0.670)
`crown4_vigilance_hmc`	Crown4 HMC proxy `(N,4,512)` C3/C4/F6/PO4	151 · 106 / 23 / 22	~453k	Crown vig ship proxy (HMC-only; F6≈F4, PO4≈O2; CBraMod test 0.680)

Channel order

Muse4: AF7, AF8, TP9, TP10
Crown8: CP3, C3, F5, PO3, PO4, F6, C4, CP4
Crown2 (HMC vig): C3, C4
Crown4 HMC proxy (vig): C3, C4, F6, PO4 (F6≈F4-M1, PO4≈O2-M1)

Keep Muse4 / Crown2 / Crown4 / Crown8 in separate configs. Do not mix montages in one training example. Crown vig configs are HMC-only (not Sleep-EDF) and must not be mixed with muse4_vigilance_sleep_edf.

Proxy / remapping caveat

Sleep-EDF, HMC, OpenNeuro attention, and engagement sources are remapped onto Muse4 or Crown8 montages. They are useful transfer / research data, but they are not native Muse S or Neurosity Crown hardware recordings for those upstream studies. Treat manifests as the authority for provenance, preprocessing, units, and checksums.

Honest shipping status

Vigilance Muse4 (muse4_vigilance_sleep_edf) is the primary public release candidate for Muse-oriented vigilance / sleep-stage transfer work.
Vigilance Crown HMC (crown2_vigilance_hmc, crown4_vigilance_hmc) are ship candidates for Crown vig proxy heads (HMC CC-BY-4.0; CBraMod test macro-F1 0.670 / 0.680). Honest proxy: crown4 uses F6≈F4, PO4≈O2. Not Sleep-EDF; do not mix with muse4 vig.
Attention configs historically show weak subject-held-out performance and stay research-only (Crown attention still not shippable).
Engagement (muse4_engagement_a_eng) is a heterogeneous multi-source proxy with task-order and device confounds — research/reproducibility only, not a shipping engagement head.

Labels encode source protocols. Do not treat them as clinical diagnosis, a universal “attention ability,” or a validated psychological state.

How to load

Files live under each config folder (e.g. muse4_vigilance_sleep_edf/windows/*.npz). Example:

import numpy as np
from huggingface_hub import hf_hub_download, snapshot_download

# one pack
path = hf_hub_download(
    repo_id="windwerfer/neurofeed-eeg-windows",
    filename="muse4_vigilance_sleep_edf/windows/SC4001_windows.npz",
    repo_type="dataset",
)
z = np.load(path)
X, y = z["X"], z["y"]          # X: (N, 4, 512) float; y: (N,)
print(z.files)                 # often includes starts, label_names, stage_*, ...

# or pull a whole config
snapshot_download(
    repo_id="windwerfer/neurofeed-eeg-windows",
    repo_type="dataset",
    allow_patterns="muse4_vigilance_sleep_edf/**",
)

Schemas and frozen subject splits are also in this repo under schemas/ and each config’s splits/.

Attribution and licenses

This repository is licensed as other / mixed open upstream. Each config redistributes derived windows only; upstream licenses still apply. See per-config ATTRIBUTION.md files and the packaging docs.

Config	Upstream (summary)	Upstream license notes
`muse4_vigilance_sleep_edf`	PhysioNet Sleep-EDF Expanded; Haaglanden Medisch Centrum (HMC) sleep staging	Sleep-EDF ODC-By; HMC CC-BY-4.0
`muse4_attention_ds001787` / `crown8_attention_ds001787`	OpenNeuro ds001787	CC0-1.0
`muse4_attention_ds003969` / `crown8_attention_ds003969`	OpenNeuro ds003969	CC0-1.0
`muse4_engagement_a_eng`	OpenNeuro ds007169 / ds007262 / ds007554; PhysioNet EEGMAT; processed STEW HF mirror	OpenNeuro pieces CC0-1.0; EEGMAT ODC-By-1.0; STEW mirror CC-BY-4.0 (see `muse4_engagement_a_eng/ATTRIBUTION.md`)
`crown2_vigilance_hmc` / `crown4_vigilance_hmc`	PhysioNet HMC sleep staging v1.1	CC-BY-4.0 (see per-config `ATTRIBUTION.md`)

Always attribute the original sources when you publish models or papers that use these windows.

What is NOT included

Raw EDF / BDF / PSG recordings (no raw/ trees)
Private muse-eeg-heads-cache or other private caches
L-FAME, LUNA, SEED-VIG
Gated REVE base weights / position caches
Any claim that proxy labels are native Muse labels or clinical truth

Obtain upstream raw data under each source’s own terms if you need to reproduce the export from scratch.

Downloads last month: 91

Total file size:

2.74 GB