aicadΒΆ

aicad compiles constraint satisfaction problems into decision diagrams (MDDs) and searches for solutions with the help of those diagrams. It is a Rust library, and a Python package, pyaicad, built on it: everything that can be done from Python can be done from Rust. You describe a problem once, as a model, and then

  • compile it into a diagram, exact or relaxed, to count, sample, read solutions off it, or prove that there is none;

  • search for solutions with a local search that repairs assignments with belief propagation on the diagrams, optionally guided by a trained network;

  • train such a network from a set of problems (needs the learning feature).

The Python guide and the Rust guide go through these in the same order, with the same examples: the Python ones are run when the documentation is built, the Rust ones are compiled and run by cargo test. The Python API reference lists everything of the Python package with its signature; the Rust API is on docs.rs, and the all-items Rust API (private items included, for contributors) is published with this documentation. People who want to change the library, or to plug their own constraint, destroy operator, search or network into it, should read Contributing.