Skip to content

Installation

The project targets Python 3.12 or newer and packages its source under src/imu_activity_pipeline/. Use an editable install so root entry points, experiment scripts, notebooks, and direct Python imports all resolve the same package.

Requirements

  • Python ≥ 3.12
  • Conda, or Python venv + pip
  • Git
  • A CUDA-capable environment for normal model training
  • JDK 17 + Android SDK only when building the Android app

Note

The public smoke test is intentionally CPU-safe and uses tiny temporary files. Full training needs the documented data; inference downloads missing public weights as described in Data & model assets.

Conda

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 .

The environment pins the numerical and machine-learning stack used by the repository, including PyTorch 2.5.1.

pip

python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt
python -m pip install -e .

py -3.12 -m venv .venv
.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements.txt
python -m pip install -e .

For a CUDA-specific PyTorch build, choose the wheel appropriate for the host driver using the official PyTorch installer, then install the remaining project dependencies. The pip requirements use PyTorch 2.8.0 for Hugging Face ZeroGPU compatibility; the reproducibility Conda environment retains the original PyTorch 2.5.1 research stack.

Verify

python -c "import imu_activity_pipeline; print(imu_activity_pipeline.__version__)"
python tests/smoke_test.py

Expected package version:

0.1.0

The smoke test checks:

  1. package imports and canonical paths;
  2. a tiny temporary tab-separated signal stream;
  3. annotation parsing; and
  4. prediction workbook writing.

Preview

Documentation dependencies are isolated from the research environment:

python -m pip install -r requirements-docs.txt
mkdocs serve

Open http://127.0.0.1:8000/. A strict production build is:

mkdocs build --strict

Troubleshooting

The package cannot be imported

Run python -m pip install -e . from the repository root and confirm that the active interpreter is the one shown by python -m pip --version.

PyTorch cannot see the GPU

Verify the driver and installed PyTorch wheel independently. The project requirements pin the framework version, but CUDA wheel selection is host-specific.

The smoke test passes but inference cannot start

That usually means the code installation is healthy but required local signal files are missing, or the public model download failed. Continue with the quick start and asset map.