|
Download README.md from madteam/thinking-bciciv2a: direct link, hf CLI and curl.
- Browser
- Download file 2.91 kB
-
https://huggingface.co/datasets/madteam/thinking-bciciv2a/resolve/main/README.md
- Command line
-
hf download hf://datasets/madteam/thinking-bciciv2a/README.md
-
curl -L -o README.md https://huggingface.co/datasets/madteam/thinking-bciciv2a/resolve/main/README.md
2.91 kB
| license: other | |
| language: | |
| - en | |
| - vi | |
| tags: | |
| - eeg | |
| - motor-imagery | |
| - bci | |
| - bciciv-2a | |
| pretty_name: BCICIV-2a EEG Motor Imagery | |
| size_categories: | |
| - 100M<n<1G | |
| # BCICIV-2a EEG Motor Imagery | |
| ## Provenance | |
| This repository contains one consolidated CSV shard derived from the local | |
| research copy identified as the Kaggle dataset | |
| [`aymanmostafa11/eeg-motor-imagery-bciciv-2a`](https://www.kaggle.com/datasets/aymanmostafa11/eeg-motor-imagery-bciciv-2a), | |
| which describes the BCI Competition IV 2a motor-imagery recordings associated | |
| with the Graz BCI laboratory. The uploader has represented that they have the | |
| right to redistribute this research copy. The repository owner must preserve | |
| any upstream attribution and terms that apply to the source dataset. | |
| This card does not assert a SPDX license. `license: other` is intentional until | |
| the upstream terms are independently mapped to a recognized license identifier. | |
| ## Contents | |
| - `BCICIV_2a_all_patients.csv` | |
| - 2,448 epochs from 9 subject identities | |
| - 22 EEG channels per epoch | |
| - 201 samples per epoch at 250 Hz | |
| - labels: `left`, `right`, `foot`, `tongue` | |
| - identity fields: `patient` (subject) and `epoch` (session boundary) | |
| The CSV is a tabular, already-windowed representation. It must not be pooled | |
| with Brennan, D01, Chisco, D09, or SparrKULee. | |
| ## Intended use | |
| Research and reproducibility testing for the THINKING EEG pipeline, including | |
| schema validation, checksum verification, leakage-safe subject/session splits, | |
| and controlled training experiments. | |
| ## Limitations and safety | |
| - This dataset is not a clinical diagnostic resource. | |
| - The scientific gate for the current BCICIV-2a encoder is **false**; no | |
| deployability or scientific headline is claimed by this repository. | |
| - Do not infer identity beyond the explicit `patient` and `epoch` fields. | |
| - Do not use random epoch splits for scientific evaluation. | |
| - Consumers must verify the immutable catalog revision and shard SHA-256 before | |
| training. | |
| - The repository contains EEG-derived human-subject data. Users are responsible | |
| for complying with the source terms, applicable law, institutional review | |
| requirements, and any restrictions imposed by the data custodian. | |
| ## Integrity | |
| Measured source shard: | |
| - bytes: `214848010` | |
| - SHA-256: `fa52c442278b63c9add50939b453f0668719ff1137dd596b5b1cac90fe3aa6cd` | |
| The GitHub catalog must pin the actual 40-character Hugging Face commit SHA; | |
| `main` and other moving revisions are not valid for scientific runs. | |
| ## Citation and attribution | |
| Cite the original BCI Competition IV 2a dataset and the upstream Kaggle source | |
| when using this derived tabular copy. Do not represent this repository as the | |
| original competition distribution. | |
| ## Release status | |
| `gate_pass: false` for the current scientific evaluation. This publication | |
| provides versioned data access only; it does not certify model quality or | |
| clinical/deployment readiness. | |