pyunlocbox 0.6.1

Convex Optimization in Python using Proximal Splitting

The PyUNLocBoX is a Python package which uses proximal splitting methods to solve non-differentiable convex optimization problems. The documentation is available on Read the Docs and development takes place on GitHub. A (mostly unmaintained) Matlab version exists.

The package is designed to be easy to use while allowing any advanced tasks. It is not meant to be a black-box optimization tool. You’ll have to carefully design your solver. In exchange you’ll get full control of what the package does for you, without the pain of rewriting the proximity operators and the solvers and with the added benefit of tested algorithms. With this package, you can focus on your problem and the best way to solve it rather that the details of the algorithms.

Content

The following solvers are included:

• Gradient descent

• Forward-backward proximal splitting (FISTA and ISTA)

• Generalized forward-backward proximal splitting

• Douglas-Rachford proximal splitting

• Monotone+Lipschitz forward-backward-forward primal-dual

• Projection-based primal-dual

The following acceleration schemes are included:

• Backtracking acceleration based on a quadratic approximation of the objective

• FISTA acceleration for forward-backward solvers

• FISTA acceleration with backtracking for forward-backward solvers

• Regularized nonlinear acceleration (RNA) for gradient descent

To compose your objective, the following functions are included:

• L1-norm (eval, prox)

• L2-norm (eval, prox, grad)

• Nuclear-norm (eval, prox)

• TV-norm (eval, prox)

• Projection on the positive octant (eval, prox)

• Projection on the L2-ball (eval, prox)

• Structured sparsity (eval, prox)

Alternatively, you can easily define a custom function by implementing an evaluation method and a proximal operator or gradient method:

>>> from pyunlocbox import functions
>>> class myfunc(functions.func):
...     def _eval(self, x):
...         return 0  # Function evaluated at x.
...     def _grad(self, x):
...         return x  # Gradient evaluated at x, if available.
...     def _prox(self, x, T):
...         return x  # Proximal operator evaluated at x, if available.

Likewise, custom solvers are defined by inheriting from solvers.solver and implementing _pre, _algo, and _post. Custom acceleration schemes are defined by inheriting from acceleration.accel and implementing _pre, _update_step, _update_sol, and _post.

Usage

Following is a typical usage example that solves an optimization problem composed by the sum of two convex functions. The functions and solver objects are first instantiated with the desired parameters. The problem is then solved by a call to the solving function.

>>> from pyunlocbox import functions, solvers
>>> f1 = functions.norm_l2(y=[4, 5, 6, 7])
>>> f2 = functions.dummy()
>>> solver = solvers.forward_backward()
>>> ret = solvers.solve([f1, f2], [0., 0, 0, 0], solver, atol=1e-5)
Solution found after 9 iterations:
    objective function f(sol) = 6.714385e-08
    stopping criterion: ATOL
>>> ret['sol']
array([3.99990766, 4.99988458, 5.99986149, 6.99983841])

You can try it online, look at the tutorials to learn how to use it, or look at the reference guide for an exhaustive documentation of the API. Enjoy!

Installation

UV (Recommended)

For the fastest installation and dependency management, use UV:

$ uv add pyunlocbox

Or install directly:

$ uv pip install pyunlocbox

To set up a development environment with UV:

$ git clone https://github.com/epfl-lts2/pyunlocbox.git
$ cd pyunlocbox
$ uv sync --dev

Pip

The PyUNLocBoX is available on PyPI:

$ pip install pyunlocbox

Conda

The PyUNLocBoX is available on conda-forge:

$ conda install -c conda-forge pyunlocbox

Contributing

See the guidelines for contributing in CONTRIBUTING.rst.

For development, we recommend using UV for fast dependency management:

$ git clone https://github.com/epfl-lts2/pyunlocbox.git
$ cd pyunlocbox
$ uv sync --dev
$ source .venv/bin/activate  # On Windows: .venv\Scripts\activate

Quick setup with just (modern command runner):

$ just setup  # Installs dependencies and sets up pre-commit hooks

Set up pre-commit hooks (recommended):

$ uv run pre-commit install

This will automatically run code formatting and linting checks before each commit.

Pre-commit hooks include:

• Code formatting: black for consistent Python code style

• Import sorting: isort for organized imports

• Linting: flake8 for code quality and style checks

• Security scanning: bandit for common security issues

• Modern Python: pyupgrade for upgrading syntax to newer Python versions

• General checks: trailing whitespace, file endings, YAML/TOML validation

Run tests with pytest:

$ uv run pytest

Run tests with coverage:

$ uv run pytest --cov=pyunlocbox --cov-report=html

Run specific test files:

$ uv run pytest pyunlocbox/tests/test_acceleration.py

Note: All tests have been successfully migrated from unittest to pytest! The complete test suite now uses modern pytest conventions: test_acceleration.py, test_functions.py, test_operators.py, test_solvers.py, and test_docstrings.py.

Run pre-commit checks manually:

$ uv run pre-commit run --all-files

Run linting:

$ uv run flake8 --doctests --exclude=doc,.venv

Format code:

$ uv run black .
$ uv run isort .

Build documentation:

$ uv run sphinx-build -b html doc/ doc/_build/

Note: This project uses GitHub Actions for continuous integration and deployment.

Similar libraries

Other proximal based algorithms and operators can be found in:

• http://proximity-operator.net/

• https://gitlab.gwdg.de/nam/ProxPython

Furthermore, many proximal operators are availlable in the proxop python library.

Acknowledgments

The PyUNLocBoX was started in 2014 as an academic open-source project for research purpose at the EPFL LTS2 laboratory.

It is released under the terms of the BSD 3-Clause license.

If you are using the library for your research, for the sake of reproducibility, please cite the version you used as indexed by Zenodo. Or cite the generic concept as:

@misc{pyunlocbox,
  title = {PyUNLocBoX: Optimization by Proximal Splitting},
  author = {Defferrard, Micha\"el and Pena, Rodrigo and Perraudin, Nathana\"el},
  doi = {10.5281/zenodo.1199081},
  url = {https://github.com/epfl-lts2/pyunlocbox/},
}