HAR-CoT
85,482 triaxial accel windows + GPT-4o chain-of-thought (OpenTSLM / TimeNet opentslm/har-cot)
Companion CoT resource from OpenTSLM (arxiv 2510.02410). Authors merged windows from DaLiAc, DOMINO, HHAR, PAMAP2, RealWorld, and Shoaib et al., keeping only activity classes present in all sources: sitting, standing, lying, walking, running, biking, walking upstairs, walking downstairs. Each window is plotted and passed to GPT-4o (gpt-4o-2024-08-06, temperature 0.3, seed 42) with the correct label and one dissimilar label to write a rationale.
- License
MIT (OpenTSLM release of windows + rationales; upstream source dataset licences still apply to underlying signals)
- Type
- Dataset
- DOI
- Not reported
- Access
- Not reported
- Subjects
- Not reported
- Signals
- IMU
- PPG sites
- None
- Setting
- Not reported
- Wrist
- Yes
- Raw waveforms
- No
- Tasks
- Not reported
- Updated
- Nov 10, 2026

Key numbers
How it works
No model in the hub reports results on this dataset yet.
- Sampling
- 50 Hz windows × 2.56 s (128 samples/axis); sources resampled into this grid
- Reference labels
- activity
Cite
No citation metadata yet.
Cite the ChatPPG record
Cite the dated snapshot 2026-10 for a fixed, reproducible version. Please also cite the original publications listed above.
APA
ChatPPG. (2026). HAR-CoT: dataset record (Version 2026-10) [Data set]. ChatPPG Research Hub. https://chatppg.com/research/datasets/har-cot
BibTeX
@misc{chatppg_dataset_har_cot,
author = {{ChatPPG}},
title = {{HAR-CoT: dataset record}},
year = {2026},
month = nov,
version = {2026-10},
howpublished = {\url{https://chatppg.com/research/datasets/har-cot}},
note = {ChatPPG Research Hub, Snapshot 2026-10, https://chatppg.com/research/snapshots/2026-10}
}CITATION.cff
cff-version: 1.2.0 message: "If you use this resource, please cite it as below." type: dataset title: "HAR-CoT: dataset record" authors: - name: "ChatPPG" url: "https://chatppg.com/research/datasets/har-cot" repository-artifact: "https://chatppg.com/research/snapshots/2026-10" version: "2026-10" date-released: "2026-11-10"
History
Record updated
Papers
Lapras: Latent Reasoning for Time Series Language ModelsSame task: Activity recognitionarXiv 2610.11111 · 2026- Accelerometry-Based Machine Learning Estimation of Energy Expenditure and Physical Activity Intensity Classification in Daily TasksSame task: Activity recognitionSensors 26 · 2026
ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learningSame task: Activity recognitionScientific Reports · 2026
