Document Motor processing and hierarchy

README.md

@@ -1,66 +1,48 @@
 ---
-license: unknown
+pretty_name: Motor Processed EEG
+task_categories:
+- time-series-classification
 tags:
-  - eeg
-  - motor-imagery
-  - in-ear-eeg
+- eeg
+- ear-eeg
+- motor-task
+- bids
 ---
 
-# Motor-Processed
+# Motor Processed EEG
 
-Preprocessed HDF5 export of the **Motor** (Wu et al. 2020, J. Neural Eng.
-17:066010, in-ear sensing for motor task classification) BIDS dataset — 6
-subjects, 1–4 runs each, one motor-imagery trial per row (`left` / `right`
-hand). Trial windows are taken directly from each event's own
-onset/duration in `events.tsv` (always 4.0 s here) — **not** the paper's own
-[-1, +2] s classification epoch (see note below).
+Paper-aligned scalp-EEG and ear-EEG HDF5 files for binary left/right motor-task classification, prepared for the [Ear-EEG FM Benchmark](https://github.com/zhikaili1150/Ear-EEG-FM-Benchmark).
 
-## Preprocessing
-
-**Only** a 0.1–100 Hz band-pass + 50 Hz notch filter is applied, on the full
-continuous recording before slicing into trials (to avoid per-trial filter
-edge effects). Nothing else: **no resampling, no channel renaming**.
-Channel names are kept exactly as in the source BIDS `channels.tsv`.
+## Files
 
-Generated by `dataset/preprocess_motor.py` in the EEGFM repo.
+| File | Samples | Shape | Size | SHA-256 |
+|---|---:|---|---:|---|
+| `motor-eeg.h5` | 1,083 | `(1083, 122, 750)` | 396,762,364 bytes | `104fd2333f44aa7ca3a9976d5e65dbb00223ac45294c9e5f94b8bbbc24818bfb` |
+| `motor-ear-eeg.h5` | 1,097 | `(1097, 6, 750)` | 20,130,804 bytes | `bc88050fcfd5242bc0166224b25f399234774029661d9ab970320f8d57fb032f` |
 
-**Note on trial length**: the 4.0 s trial length here is the raw
-`duration` column in `events.tsv` (the movement-hold period marked in the
-recording), *not* a value stated by the paper as its classification window.
-Wu et al. 2020's own EEGNet pipeline (§2.2.1) uses a [-1, +2] s = 3 s window
-around arrow onset, 250 Hz, 4-38 Hz filter — independent of this `duration`
-column. That paper-matched variant lives separately in this repo's source
-project as `motor-eeg.h5` / `motor-ear-eeg.h5`; do not confuse it with the
-files here.
+The leading dimension differs because peak-to-peak rejection is applied independently to scalp and ear channels.
 
-## Files
+## Processing and hierarchy
 
-| file | channels | trials (N) | sfreq | trial length |
-|---|---|---|---|---|
-| `motor-scalp-eeg.h5` | 122 scalp EEG (drops 8 ear + Trigger + channel "10") | 1114 | 1000 Hz | 4.0 s (4000 samples) |
-| `motor-in-ear-eeg.h5` | 6: LF, LB, LOD, RF, RB, ROD (LOU/ROU excluded — used as the physical reference in the paper's ear-IPSI/CONTRA montage, not as signal channels) | 1114 | 1000 Hz | 4.0 s (4000 samples) |
+- Source hierarchy: `subject / session / run / block / trial`
+- Source trials: 1,114 experimental trials across 6 subjects; each trial lasts 10 s
+- H5 sample: one 3 s window from `-1` to `+2` s relative to event onset
+- Signal processing: 4–38 Hz band-pass, 250 Hz output sampling rate, 750 time points
+- Labels: `left` (`0`) and `right` (`1`)
+- Rejection: peak-to-peak amplitude greater than 800 uV
+- `trial_id`: experimental trial identifier from `events.tsv:trial_id`
+- `split_group_id`: paper-defined experimental block from `events.tsv:block_id`; treat it as indivisible within its recording when constructing data splits
+- Data unit: uV, stored as `float32`
 
-Labels are perfectly balanced: 557 `left` / 557 `right` in both files (row-aligned — same trial order/index in both).
+The HDF5 schema version is `0.4.0`. Each file also stores subjects, sessions, runs, recording IDs, event/sample IDs, channel names, window bounds, preprocessing metadata, and class names.
 
-## HDF5 schema (v0.2)
+## Download
 
+```bash
+huggingface-cli download Zachary1150/Motor-Processed \
+    --repo-type dataset --local-dir data
 ```
-/data          (N, C, T)   float32   signal, µV
-/durations     (N,)        int64     valid samples per trial (== T here, fixed-length)
-/nan_fraction  (N,)        float32   fraction of non-finite samples in the trial
-/labels        (N,)        int64     index into attrs['class_names'] (['left','right'])
-/subject       (N,)        str       'sub-01' ...
-/session       (N,)        str       'ses-01'
-/task          (N,)        str       'motor'
-/run           (N,)        str       'run-01' ... 'run-04'
-/trial_id      (N,)        int64     row index in the source events.tsv
-/ch_names      (C,)        str       channel names, as in source BIDS channels.tsv
 
-attrs: sfreq, n_class, class_names, unit ('uV'), eegfm_version,
-       source_bids_path, bids_dataset_name, preprocess_config_json, created_at
-```
+## Source
 
-`preprocess_config_json` (per file) records the exact filter settings, e.g.:
-```json
-{"filter_low": 0.1, "filter_high": 100.0, "notch": 50.0, "trial_window": "events_tsv onset/duration (label-defined length)"}
-```
+The original recordings accompany *An investigation of in-ear sensing for motor task classification* and are distributed through [IEEE DataPort](https://ieee-dataport.org/open-access/ear-eeg-recording-brain-computer-interface-motor-task). Users are responsible for complying with the original dataset terms.