Sample Data
A small subset of the Multisite PPG Dataset for quick inspection, tutorial work, and pipeline testing without pulling the full ~18.6 GB.
Contents
This subset contains two participants — P7 and P8 — laid out exactly like the main dataset:
sample_data/
├── raw_data/
│ ├── P7/ # Earring / Necklace / Ring / Watch raw recordings
│ │ # + Polar H10 ECG + Polar H10 accelerometer
│ │ # + activity log
│ └── P8/ # Same structure as P7
└── ppg_windowed_data/
├── P7/ # Pre-windowed PPG segments (8 s windows, 1 s stride)
└── P8/ # Same structure as P7
Total size: ~1.42 GB.
How it was obtained
The files in this folder are bit-for-bit copies of raw_data/P7/, raw_data/P8/, ppg_windowed_data/P7/, and ppg_windowed_data/P8/ from the main dataset — no resampling, filtering, downsampling, or other modification was applied. File schemas, dtypes, sampling rates, units, and gap-handling conventions are identical to the full dataset.
P7 and P8 were chosen as representative samples because both participants have:
- Data from all four wearable sites (Earring, Necklace, Ring, Watch),
- Both Polar H10 reference signals (ECG and 3-axis accelerometer),
- Non-empty self-reported activity logs spanning multiple labelled activities.
Download just the sample data
If you only want to try the format on a small footprint, fetch this folder alone instead of the whole repository:
python -c '
from huggingface_hub import snapshot_download
snapshot_download(
repo_id="anonymous-ppg-dataset/multisite-ppg-submission",
repo_type="dataset",
local_dir="anonymous-ppg-dataset/multisite-ppg-submission",
allow_patterns="sample_data/*",
)
'
This pulls roughly 1.42 GB instead of 18.6 GB. Files land under anonymous-ppg-dataset/multisite-ppg-submission/sample_data/, mirroring the on-Hub structure.
Further documentation
For file-level schemas (keys, shapes, dtypes), sampling-rate details, activity-log format, intended uses, limitations, ethics, and license, see the main dataset README. Everything documented there applies identically to this subset.