CPMpy/cpmpy

Constraint Programming and Modeling library in Python, based on numpy, with direct solver access.

What it solves

CPMpy tackles combinatorial problems involving Boolean and integer variables, such as scheduling, resource allocation, and optimization. It provides high-level constraints that are automatically decomposed when a solver does not support them natively, making it easy to express complex problems.

How it works

You define decision variables as numpy arrays, add constraints and an objective, then call solve(). CPMpy translates your model to the chosen solver (CP, ILP, SMT, PB, SAT, etc.) and can compare different solvers. It supports incremental solving, assumption variables, and adding constraints or updating objectives after solving.

Who it's for

Researchers and practitioners in constraint programming, operations research, and machine learning who need a lightweight, well-documented library that integrates with the scientific Python ecosystem (pandas, plotly, Jupyter). It is also used in industry and academic research.

Highlights

  • Solver-agnostic: works with CP, ILP, SMT, PB, and SAT solvers
  • ML-friendly: decision variables are numpy arrays with vectorized operations
  • Incremental solving: assumption variables, adding constraints, updating objectives
  • Extensively tested with a large test suite and fuzz testing
  • Tools for parameter tuning, debugging, explanation generation, and XCSP3 benchmarking
  • Open-source (Apache 2.0) with an active community and contributions welcome

Related

  • Project
    pyscf/gpu4pyscf

    GPU4PySCF is a plugin that adds GPU acceleration to the PySCF quantum chemistry package, enabling fast Hartree-Fock and DFT calculations on NVIDIA GPUs. It supports molecular and periodic systems, excited states, and geometry optimization, with reported speedups of 10–30 times over a 32-core CPU baseline.

  • Project
    AllenDowney/BayesMadeSimple

    A beginner-friendly tutorial that teaches Bayesian statistics through interactive Python notebooks, using grid algorithms to solve practical probability and decision-making problems.

  • Project
    Trustworthy-AI-Group/Adversarial_Examples_Papers

    A curated, continuously updated repository of papers on adversarial examples, created to replace a crashing original list. It aggregates recent research and includes a snapshot up to 2023-09-01, plus a separate list on transfer-based attacks.

  • Project
    MinghuiChen43/awesome-trustworthy-deep-learning

    An awesome list of research papers and resources on trustworthy deep learning, covering topics like out-of-distribution generalisation, adversarial attacks, privacy, fairness, and interpretability. It is updated daily from arXiv and includes paper digests, links to full texts, and pointers to related tools and events.