moabb.datasets.Wang2025#

class moabb.datasets.Wang2025(subjects=None, sessions=None)[source]#

Bases: BaseDataset

Dataset Snapshot

Wang2025

Four-class motor imagery EEG from 15 healthy subjects (left hand, right hand, tongue, both feet) recorded with 62 channels at 256 Hz over two days of calibration and online feedback sessions.

Imagery, 4 classes (left_hand vs right_hand vs tongue vs feet)

AuthorsJiaheng Wang, Lin Yao, Yueming Wang

🇨🇳 Zhejiang University, CN·2025

Imagery Code: Wang2025 15 subjects 2 sessions 62 ch 256 Hz 4 classes 4.0 s trials

Class Labels: left_hand, right_hand, tongue, feet

Read Paper Compare Similar Report Issue

Overview

Four-class motor imagery dataset (ZJU-MI-EEG / MI4)

Dataset description

15 healthy subjects performed cued four-class motor imagery (left hand, right hand, tongue, both feet) on two days, mapped to two sessions. Each day has a 240-trial calibration run and a 160-trial online feedback run (MI4 subset of the ZJU-MI-EEG Hugging Face dataset; 62 channels, 256 Hz). In the paper each session consists of six calibration runs and four online feedback runs of 40 trials (10 per class); the release merges them into one calibration and one feedback file per day, which are the two runs exposed here. The paper reports a 62-channel g.USBamp amplifier (g.tec) sampled at 256 Hz and high-pass filtered above 0.1 Hz, whereas the release ships a 62channels_gNautilus.ced montage file; the amplifier model is therefore recorded as reported by the paper. Feedback trials lasted up to 10 s online, but the release stores every trial as the -1 to 4 s window around the cue.

Each .mat run stores EEG_data (62, 1280, n_trials) in microvolts (-1 to 4 s around the cue) and integer labels. The loader concatenates the trials, marks each cue with a stim event and converts to volts; the interval ends at 4 - 1/256 s.

Open in Overview tab →

Citation & Impact

Paper DOI10.1109/TNSRE.2025.3591254
CitationsLoading…
Public APICrossref | OpenAlex

Stimulus Protocol

4s task window per trial · 4-class imagery paradigm · 2 runs/session across 2 sessions

HED Event Tags

HED tags4/4 events annotated

Source: MOABB BIDS HED annotation mapping.

Agent-action

4

Sensory-event

4

left_hand

Sensory-eventAgent-action

right_hand

Sensory-eventAgent-action

tongue

Sensory-eventAgent-action

feet

Sensory-eventAgent-action

HED tree view

Tree · left_hand

├─ Sensory-event
│  ├─ Experimental-stimulus
│  └─ Visual-presentation
└─ Agent-action
   └─ Imagine
      ├─ Move
      └─ Left
         └─ Hand

Tree · right_hand

├─ Sensory-event
│  ├─ Experimental-stimulus
│  └─ Visual-presentation
└─ Agent-action
   └─ Imagine
      ├─ Move
      └─ Right
         └─ Hand

Tree · tongue

├─ Sensory-event
│  ├─ Experimental-stimulus
│  └─ Visual-presentation
└─ Agent-action
   └─ Imagine
      ├─ Move
      └─ Tongue

Tree · feet

├─ Sensory-event
│  ├─ Experimental-stimulus
│  └─ Visual-presentation
└─ Agent-action
   └─ Imagine
      ├─ Move
      └─ Foot

Channel Summary

Total channels62

EEG62

Montage10-10

Sampling256 Hz

Filter0.1 Hz high-pass

Notch / line50 Hz

This diagram is automatically generated from MOABB metadata. Please consult the original publication to confirm the experimental protocol details.

Overview

Four-class motor imagery dataset (ZJU-MI-EEG / MI4) [1].

Dataset description

15 healthy subjects performed cued four-class motor imagery (left hand, right hand, tongue, both feet) on two days, mapped to two sessions. Each day has a 240-trial calibration run and a 160-trial online feedback run (MI4 subset of the ZJU-MI-EEG Hugging Face dataset; 62 channels, 256 Hz). In the paper [1] each session consists of six calibration runs and four online feedback runs of 40 trials (10 per class); the release merges them into one calibration and one feedback file per day, which are the two runs exposed here. The paper reports a 62-channel g.USBamp amplifier (g.tec) sampled at 256 Hz and high-pass filtered above 0.1 Hz, whereas the release ships a 62channels_gNautilus.ced montage file; the amplifier model is therefore recorded as reported by the paper. Feedback trials lasted up to 10 s online, but the release stores every trial as the -1 to 4 s window around the cue.

Each .mat run stores EEG_data (62, 1280, n_trials) in microvolts (-1 to 4 s around the cue) and integer labels. The loader concatenates the trials, marks each cue with a stim event and converts to volts; the interval ends at 4 - 1/256 s.

References

[1] (1,2)

Wang, J., Yao, L., and Wang, Y. (2025). Enhanced Online Continuous Brain-Control by Deep Learning-Based EEG Decoding. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 33, 2834-2846. DOI: 10.1109/TNSRE.2025.3591254

Code Examples

from moabb.datasets import Wang2025
dataset = Wang2025()
data = dataset.get_data(subjects=[1])
print(data[1])

Metadata

Dataset summary

#Subj	15
#Chan	62
#Classes	4
#Trials / class	200
Trials length	3.99609375 s
Freq	256 Hz
#Sessions	2
#Runs	2
Total_trials	12000

Participants

Population: healthy
Age: 24 years
BCI experience: 10 BCI-naive; 5 without online BCI experience

Equipment

Amplifier: g.USBamp (g.tec medical engineering, Austria), 62 channels
Montage: 10-10

Data Access

DOI: 10.1109/TNSRE.2025.3591254
Data URL: https://huggingface.co/datasets/Jiaheng-Wang/ZJU-MI-EEG
Repository: Hugging Face

Experimental Protocol

Paradigm: imagery
Feedback: visual
Stimulus: visual

Notes

Added in version 1.8.

__init__(subjects=None, sessions=None)[source]#: Initialize function for the BaseDataset.

property all_subjects#: Full list of subjects available in this dataset (unfiltered).

convert_to_bids(path=None, subjects=None, overwrite=False, format='EDF', verbose=None, generate_figures=False)[source]#

Convert the dataset to BIDS format.

Saves the raw EEG data in a BIDS-compliant directory structure. Unlike the caching mechanism (see CacheConfig), the files produced here do not contain a processing-pipeline hash (desc-<hash>) in their names, making the output a clean, shareable BIDS dataset.

Parameters:

path (str | Path | None) – Directory under which the BIDS dataset will be written. If None the default MNE data directory is used (same default as the rest of MOABB).
subjects (list of int | None) – Subject numbers to convert. If None, all subjects in subject_list are converted.
overwrite (bool) – If True, existing BIDS files for a subject are removed before saving. Default is False.
format (str) – The file format for the raw EEG data. Supported values are "EDF" (default), "BrainVision", and "EEGLAB".
verbose (str | None) – Verbosity level forwarded to MNE/MNE-BIDS.
generate_figures (bool) – If True, generate interactive neural signature HTML figures in {bids_root}/derivatives/neural_signatures/. Requires plotly (pip install moabb[interactive]). Default is False.

Returns:

bids_root – Path to the root of the written BIDS dataset.

Return type:

pathlib.Path

Examples

>>> from moabb.datasets import AlexMI
>>> dataset = AlexMI()
>>> bids_root = dataset.convert_to_bids(path="/tmp/bids", subjects=[1])

Notes

Use CacheConfig to configure caching for get_data(). Use moabb.datasets.bids_interface.get_bids_root to get the BIDS root path.

Added in version 1.5.

data_path(subject, path=None, force_update=False, update_path=None, verbose=None)[source]#: Return [s1_calibration, s1_feedback, s2_calibration, s2_feedback].

download(subject_list=None, path=None, force_update=False, update_path=None, accept=False, verbose=None)[source]#

Download all data from the dataset.

This function is only useful to download all the dataset at once.

When the dataset declares a nemar_id and the download provider is not "upstream", the files come from NEMAR’s sourcedata/ – the original pre-BIDS distribution, byte-identical to what the upstream host serves. On any NEMAR failure this falls back to the dataset’s own downloader with a warning (unless the provider is pinned to "nemar"). See sourcedata_path() and moabb.set_download_provider().

Parameters:

subject_list (list of int | None) – List of subjects id to download, if None all subjects are downloaded. On the NEMAR path each subject is resolved through the deposit’s sourcedata_provenance.json; deposits enriched before that manifest recorded subjects fetch the whole tree.
path (None | str) – Location of where to look for the data storing location. If None, the environment variable or config parameter MNE_DATASETS_(dataset)_PATH is used. If it doesn’t exist, the “~/mne_data” directory is used. If the dataset is not found under the given path, the data will be automatically downloaded to the specified folder.
force_update (bool) – Force update of the dataset even if a local copy exists.
update_path (bool | None) – If True, set the MNE_DATASETS_(dataset)_PATH in mne-python config to the given path. If None, the user is prompted. Not used on the NEMAR path.
accept (bool) – Accept licence term to download the data, if any. Default: False. Only relevant to the dataset’s own downloader; NEMAR mirrors are already public.
verbose (bool, str, int, or None) – If not None, override default verbose level (see mne.verbose()).

get_additional_metadata(subject: str, session: str, run: str)[source]#

Load additional metadata for a specific subject, session, and run.

This method is intended to be overridden by subclasses to provide additional metadata specific to the dataset. The metadata is typically loaded from an events.tsv file or similar data source.

Parameters:

subject (str) – The identifier for the subject.
session (str) – The identifier for the session.
run (str) – The identifier for the run.

Returns:

A DataFrame containing the additional metadata if available, otherwise None.

Return type:

None | pandas.DataFrame

get_block_repetition(paradigm, subjects, block_list, repetition_list)[source]#

Select data for all provided subjects, blocks and repetitions.

subject -> session -> run -> block -> repetition

See also

get_data

Parameters:

subjects (List of int) – List of subject number
block_list (List of int) – List of block number
repetition_list (List of int) – List of repetition number inside a block

Returns:

data – dict containing the raw data

Return type:

Dict

get_data(subjects=None, cache_config=None, process_pipeline=None, n_jobs=1)[source]#

Return the data corresponding to a list of subjects.

The returned data is a dictionary with the following structure:

data = {"subject_id": {"session_id": {"run_id": run}}}

subjects are on top, then we have sessions, then runs. A sessions is a recording done in a single day, without removing the EEG cap. A session is constitued of at least one run. A run is a single contiguous recording. Some dataset break session in multiple runs.

Processing steps can optionally be applied to the data using the *_pipeline arguments. These pipelines are applied in the following order: raw_pipeline -> epochs_pipeline -> array_pipeline. If a *_pipeline argument is None, the step will be skipped. Therefore, the array_pipeline may either receive a mne.io.Raw or a mne.Epochs object as input depending on whether epochs_pipeline is None or not.

Parameters:

subjects (List of int) – List of subject number
cache_config (dict | CacheConfig) – Configuration for caching of datasets. See CacheConfig for details.
process_pipeline (sklearn.pipeline.Pipeline | None) – Optional processing pipeline to apply to the data. To generate an adequate pipeline, we recommend using moabb.make_process_pipelines(). This pipeline will receive mne.io.BaseRaw objects. The steps names of this pipeline should be elements of StepType. According to their name, the steps should either return a mne.io.BaseRaw, a mne.Epochs, or a numpy.ndarray. This pipeline must be “fixed” because it will not be trained, i.e. no call to fit will be made.
n_jobs (int) – Number of jobs to run in parallel over subjects (passed to joblib.Parallel). Default 1 (sequential). Per-subject processing (reading, filtering, resampling, epoching) is independent, so this gives a near-linear speedup for datasets with many subjects.

Returns:

data – dict containing the raw data

Return type:

Dict

property metadata[source]#

Return structured metadata for this dataset.

Returns the DatasetMetadata object from the centralized catalog, or None if metadata is not available for this dataset.

Returns:: The metadata object containing acquisition parameters, participant demographics, experiment details, and documentation. Returns None if no metadata is registered for this dataset.
Return type:: DatasetMetadata | None

Examples

>>> from moabb.datasets import BNCI2014_001
>>> dataset = BNCI2014_001()
>>> dataset.metadata.participants.n_subjects
9
>>> dataset.metadata.acquisition.sampling_rate
250.0

sourcedata_path(subject=None, path=None, force_update=False, verbose=None)[source]#

Get the dataset’s original pre-BIDS distribution from NEMAR.

Where data_path() fetches the original files from the upstream host, this fetches the copy NEMAR mirrors under sourcedata/. The files keep their upstream names, so the two are interchangeable in content – but NEMAR stays reachable when the upstream host is slow, rate-limited, behind a bot gate, or retired.

Parameters:

subject (int | str | None) – Restrict the download to one subject, resolved through the deposit’s sourcedata_provenance.json. Deposits enriched before that manifest recorded subjects fall back to the whole tree with a warning. When None the whole tree is fetched.
path (None | str) – Base path where MOABB stores datasets.
force_update (bool) – Re-fetch even when a local copy is present.
verbose (bool, str, int, or None) – If not None, override default verbose level.

Returns:

Local path to the sourcedata directory.

Return type:

str

Raises:

ValueError – If the dataset declares no nemar_id.
moabb.datasets.download.NemarDownloadError – If the download fails or the deposit publishes no sourcedata/.