DualPerspective 0.1.4

Python interface for DualPerspective.jl - A Julia package for obtaining maximum-entropy solutions to underdetermined linear systems.

DualPerspective Python Package

Python interface for DualPerspective.jl, a Julia package for solving large-scale KL divergence problems.

Installation

pip install DualPerspective

The package automatically:

1. Installs Julia if not already installed (via juliacall)
2. Installs the DualPerspective.jl Julia package from the official registry

TODO:

• Precompiles dependencies for fast performance from the first run

Basic Usage

import numpy as np
from DualPerspective import DPModel, solve, regularize

# Generate sample data
np.random.seed(42)
n = 200  # dimension of solution
m = 100  # number of measurements
x0 = np.pi * (tmp := np.random.rand(n)) / np.sum(tmp)
A = np.random.rand(m, n)
b = A @ x0  # measurements

# Create and solve the problem
model = DPModel(A, b)
regularize(model, 1e-4)  # Optional: set regularization parameter
solution = solve(model)

print(f"Sum of solution: {np.sum(solution):.6f} (should be ≈ {np.pi:.6f})")
print(f"Optimal solution shape: {solution.shape}")

Features

• Python interface for DualPerspective.jl
• Automatic installation of Julia dependencies
• Integration with NumPy arrays

Performance Considerations

This package uses Julia's precompilation to ensure good performance. The first import may take slightly longer as it sets up the Julia environment, but subsequent operations should be fast.

Advanced Usage

Updating DualPerspective.jl

To reinstall or update the Julia package:

uv pip install --force-reinstall DualPerspective

Local Development

By default, the Python interface uses the version of DualPerspective.jl from the Julia registry. For local development, use the environment variable DUALPERSPECTIVE_USE_LOCAL:

export DUALPERSPECTIVE_USE_LOCAL=true

Or, to set it temporarily for a single command:

DUALPERSPECTIVE_USE_LOCAL=true python [command]

Building and Publishing

Prerequisites

Install the necessary tools:

uv pip install --upgrade build twine

Building the Package

Build both wheel and source distributions:

python -m build

This creates distribution files in the dist/ directory.

Testing Locally

Before publishing, test the package locally:

# Install in development mode
source venv/bin/activate
uv pip install -e .

# Or install the built wheel
uv pip install dist/DualPerspective-0.1.1-py3-none-any.whl

Publishing to TestPyPI (Optional)

To test the publishing process:

twine upload --repository-url https://test.pypi.org/legacy/ dist/*

# Install from TestPyPI
uv pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ DualPerspective

Publishing to PyPI

Once tested and ready:

twine upload dist/*

Updating for New Releases

1. Update the version number in pyproject.toml
2. Make code changes
3. Rebuild: python -m build
4. Upload: twine upload dist/*

Requirements

• Python 3.7+
• NumPy
• juliacall

License

This project is licensed under the MIT License.