Installation¶

Requires Python 3.10 or newer. PyTorch is optional.

pip install rddac

Verify the install:

import rddac
print(rddac.__version__)

PyTorch¶

RDDACDataset and the rest of rddac.pytorch require PyTorch.

pip install rddac[torch]

This pulls the default PyTorch wheel from PyPI (CPU on Windows and macOS, CUDA 12.x on Linux). For a different build (CUDA, ROCm, MPS) install PyTorch first from the PyTorch selector, then pip install rddac without the extra. The package does not pin PyTorch, so an existing wheel is preserved.

Preprocessing¶

The pointcloud stage of rddac preprocess (geometric cleaning, ICP alignment, random-forest fin classifier) needs scipy, scikit-learn and joblib; the force, sheet and oil stages run without them.

pip install rddac[preprocessing]

Advanced¶

Documentation build¶

pip install rddac[docs]

Then mkdocs serve from the repository root for a live preview at http://127.0.0.1:8000.

Development install¶

git clone https://github.com/BaumSebastian/RDDAC.git
cd RDDAC
pip install -e .[dev,torch]
pre-commit install

[dev] adds pytest, black, ruff, bumpver, pre-commit, and the [preprocessing] extra so the whole test suite collects.

All runtime dependencies install automatically. This includes the ddacs package, which provides the shared dataset machinery and the download of the matching simulations.