PARASOL 1.X
=================================

Parallel CP/MIP Portfolio Solver

Parasol is a multi-threaded, low-overhead universal portfolio executor written in Rust that allows solving Constraint (Satisfaction/Optimisation) Problems defined in the MiniZinc language by using a portfolio approach. It is built to support a wide range of state-of-the-art constraint solvers, coordinating them dynamically via a machine-learning-based portfolio selector.

In a nutshell, Parasol relies on two sequential steps:

1. FEATURE EXTRACTION & STATIC PRE-SOLVING: Consists of the parallel execution
of a 7-core static schedule concurrently with the extraction of the problem
features from the FlatZinc representation using fzn2feat and model
compilation on the remaining core, ensuring that cores are not left
idle during the start-up and inference phase.
2. DYNAMIC SOLVING: Consists of the parallel execution of a predicted
portfolio (utilising the full 8-core budget) selected by a machine learning
model based on the 95 extracted features. The model is an ensemble of 30
Support Vector Classifiers with RBF kernels utilizing a bagged margin-weighted
approach (BagSVC-MW). It selects whether to run CP-SAT on all 8 cores or
to deploy a custom portfolio of constituent solvers with optional
bound-sharing intervals.

This version is the submission to the MiniZinc Challenge 2026, configured with:  OR-Tools CP-SAT (1 core), Chuffed (1 core), Gecode (2 cores), Gurobi (2 cores), Picat (1 core), Yuck (1 core).

Parasol is available on GitHub: https://github.com/Portfolio-Solver-Platform/parasol

AUTHORS
=======

Parasol is developed as part of a master's thesis project by Jonas Bork Jørgensen and Sofus Hesseldahl Laubel
(University of Southern Denmark), under the supervision of Professor Jacopo Mauro and in collaboration with Alessio Pellegrino.
For any questions or information, please contact the authors at:
alessio at imada.sdu.dk

mauro at imada.sdu.dk

jonjo21 at student.sdu.dk

slaub21 at student.sdu.dk