README.md

metadata

title: MoveTSA Dataset Builder
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
hf_oauth: true
short_description: Generate parametrised MoveTSA windows, no raw access

MoveTSA dataset builder

A gated Gradio Space that generates a parametrised MoveTSA windows parquet (HRV / BSI / ECG-derived respiration / simulator aggregates / DATEX / subjective labels) on demand — without giving users access to the raw recordings.

How it works

• The raw recordings live in the private dataset MoveTSA/movetsa-raw and are downloaded into the Space at startup. They never leave the Space.
• The pipeline code is pulled from MoveTSA/MoveTSA and imported as the MoveTSA package.
• Authorised users sign in with Hugging Face, pick window_size / overlap / normalize / baselines / familiarization, and download only the generated parquet.

Access control (gated)

The Space is public, but generation requires:

1. signing in with Hugging Face (hf_oauth), and
2. being listed in the ALLOWLIST Space variable (comma-separated usernames).

Non-listed users are told to request access from @thbndi. Add a username by editing the ALLOWLIST variable in Settings → Variables and secrets — no redeploy of code needed.

Configuration (Settings → Variables and secrets)

Name	Kind	Purpose
HF_TOKEN	secret	Fine-grained read token with access to MoveTSA/movetsa-raw and MoveTSA/MoveTSA.
ALLOWLIST	variable	Comma-separated HF usernames allowed to generate (default: thbndi).
RAW_REPO	variable (optional)	Raw dataset id (default MoveTSA/movetsa-raw).
CODE_REPO	variable (optional)	Pipeline-code dataset id (default MoveTSA/MoveTSA).

First load downloads ~13 GB of raw SIMU logs; expect a slow cold start. Enable persistent storage on the Space to avoid re-downloading on each restart.