Skip to content

Reproduce

The public repository supports two distinct workflows: a data-free package check and full experiments with authorized recordings. Keeping them separate avoids implying that participant data are bundled with the code.

Public verification

git clone https://github.com/rudykon/Wearable-IMU-Activity-Segmentation-Pipeline.git
cd Wearable-IMU-Activity-Segmentation-Pipeline
conda env create -f environment.yml
conda activate imu-activity-pipeline
python -m pip install -e .
python tests/smoke_test.py

The smoke test is CPU-safe and requires neither participant recordings nor model checkpoints.

Workflow map

Step Documentation Main command
Install Environment and verification python tests/smoke_test.py
Place authorized data Dataset and asset boundaries —
Train posterior models Training protocol python train.py
Generate records Inference and TRL python run_inference.py
Score records Segment evaluation python evaluate.py --split external_test
Build the mobile prototype Android ./gradlew assembleDebug

Python and Android weights are hosted in the public Hugging Face model repository. Missing assets are downloaded into their expected paths and checked against model-assets.json.

Full experiment wrapper

After placing the required local data and fixed assets, run:

bash run_reproducibility_experiments.sh

The wrapper executes the saved-model external evaluation, internal robustness and policy-selection checks, representative timeline generation, an external unlabeled-cohort stress test, and summary-figure generation. Outputs remain under:

experiments/results/
experiments/figures/
experiments/logs/

Use a specific interpreter when needed:

PYTHON_BIN=/absolute/path/to/python bash run_reproducibility_experiments.sh

Interfaces

  • Quickstart — minimum authorized-data path.
  • API — package imports and output schema.
  • Models and licenses — filenames, hashes, and distribution boundaries.
  • Citation — versioned software citation.

Data boundary

Participant recordings are not stored in the public repository. Preserve the documented access, privacy, split, and licensing conditions when reproducing or extending the study.