torchvinecopulib

A Python library for fitting and sampling vine copulas, using PyTorch.

• C/D/R-Vine full-sampling/ quantile-regression/ conditional-sampling, all in one package
  • Flexible sampling order for experienced users
• Vectorized tensor computation with GPU (device='cuda') support
• Shorter runtimes for higher dimension simulations
• Pure Python library, inspired by pyvinecopulib on Windows, Linux, MacOS
• IO and visualization support

Citation

If you use torchvinecopulib in your work, please cite:

Cheng, Tuoyuan, Thibault Vatter, Thomas Nagler, and Kan Chen. "Vine Copulas as Differentiable Computational Graphs." arXiv preprint arXiv:2506.13318 (2025).

@article{cheng2025vine,
  title={Vine Copulas as Differentiable Computational Graphs},
  author={Cheng, Tuoyuan and Vatter, Thibault and Nagler, Thomas and Chen, Kan},
  journal={arXiv preprint arXiv:2506.13318},
  year={2025},
  url={https://arxiv.org/abs/2506.13318},
}

Examples

Visit the ./examples/ folder for .ipynb Jupyter notebooks.

Installation

• By pip from PyPI (see the dependencies and uv sections below for CUDA support):

pip install torchvinecopulib torch

• Or pip from ./dist/*.whl or ./dist/*.tar.gz in this repo. Need to use proper file name.

# inside project root folder
pip install ./dist/torchvinecopulib-1.1.0-py3-none-any.whl
# or
pip install ./dist/torchvinecopulib-1.1.0.tar.gz

(Recommended) uv for Dependency Management and Packaging

After git clone https://github.com/TY-Cheng/torchvinecopulib.git, cd into the project root where pyproject.toml exists,

# From inside the project root folder
# Create and activate local virtual environment
uv venv .venv
source .venv/bin/activate

# Sync dependencies with CPU support (default)
uv sync --extra cpu

# Or for CUDA 12.6 or 12.8 support (depends on your CUDA version)
uv sync --extra cu126

# Additionally, to install additional dependencies for the examples
uv sync --extra examples

Dependencies

# inside the `./pyproject.toml` file;
fastkde = "*"
numpy = "*"
pyvinecopulib = "*"
python = ">=3.11"
scipy = "*"
# optional to facilitate customization
torch = [
    { index = "torch-cpu", extra = "cpu" },
    { index = "torch-cu126", extra = "cu126" },
    { index = "torch-cu128", extra = "cu128" },
]

For PyTorch with cuda:

pip install torch --index-url https://download.pytorch.org/whl/cu126 --force-reinstall
# check cuda availability
python -c "import torch; print(torch.cuda.is_available())"

Tip

macOS users should set device='cpu' at this stage, for using device='mps' won't support dtype=torch.float64.

Documentation

• Visit GitHub Pages website

• Or build by yourself (need Sphinx, theme furo and the GNU make)

# inside project root folder
sphinx-apidoc -o ./docs ./torchvinecopulib && cd ./docs && make html && cd ..
# if using uv
uv run sphinx-apidoc -o docs torchvinecopulib/ --separate
uv run sphinx-build docs docs/_build/html

Tests

# inside project root folder
python -m pytest ./tests
# coverage report
coverage run -m pytest ./tests && coverage html
# if using uv
uv run coverage run --source=torchvinecopulib -m pytest ./tests
uv run coverage report -m

TODO

• VineCop.rosenblatt
• replace dict with torch.Tensor using some mod
• vectorized union-find
• flatten _visit logic
• examples/someapplications.ipynb
• flatten dynamic nested dicts into tensors
• fastkde.pdf onto torch.Tensor

Contributing

We welcome contributions, whether it's a bug report, feature suggestion, code contribution, or documentation improvement.

• If you encounter any issues with the project or have ideas for new features, please open an issue on GitHub or privately email us. Make sure to include detailed information about the problem or feature request, including steps to reproduce for bugs.

Code Contributions

1. Fork the repository and create a new branch from the main branch.
2. Make your changes and ensure they adhere to the project's coding style and conventions.
3. Write tests for any new functionality and ensure existing tests pass.
4. Commit your changes with clear and descriptive commit messages.
5. Push your changes to your fork and submit a pull request to the main branch of the original repository.

Pull Request Guidelines

• Keep pull requests focused on addressing a single issue or feature.
• Include a clear and descriptive title and description for your pull request.
• Make sure all tests pass before submitting the pull request.
• If your pull request addresses an open issue, reference the issue number in the description using the syntax #issue_number.
• in-place ops can be slower
• torch.jit.script can be slower

License

This project is released under the MIT License (© 2024- Tuoyuan Cheng, Kan Chen).
See LICENSE for the full text, including our own grant of rights and disclaimer.

Third-Party Dependencies

See the “Third-Party Dependencies” section in LICENSE for details on the PyTorch, FastKDE, and pyvinecopulib licenses that govern those components.