Solving Picross using Constraint-Programming tools

• Creation of a generic picross instance, loaded with ad-hoc .px files.

• Creation of a picross-solver class, taking the .px file location as an input

• First version consisted in enumeration of each solution for each row and each column.

• Current versions tries to capture reasoning done by humans.

• 🐖 : PIG (Picross Instance Generator), small python module creating .px instances.

• 🧠 : Writing a checker, ensuring every solution is correct

• ⚔️ : First version of the solver works well for middle-size grids, but fails to load whenever too many tuples are possible.
• 🤔 : Current version does not (yet) capture every human reasoning, but bigger instances can be loaded. (Roughly every 15x15 grid can be solved)

• ↪️ : Propagation of the constraints, looking for the level of consistency of our model.
• 🚅 : Benchmarking : How many nodes are used ? How many tuples were enumerated ?

• ⁉️ : Is there a model with the same consistencies, but using less tuples ?
• 1. This can be done using DFAs (cf. Choco solver, however there is no DFA / regular on CPLEX).
• ✍️ : Writing a small report on the whole process (cf. Overleaf ).

• Algebraic formulation for a picross instance