Modelling Kakuro as CSP and Solving Kakuro Puzzles

Kakuro or Kakkuro or Kakoro is a kind of logic puzzle that is often referred to as a mathematical transliteration of the crossword.

We model Kakuro as a CSP and solve Kakuro puzzles with various algorithms. Our goal is to show how different algorithms for CSP behave during the solution of these puzzles.

Algorithms for CSP

The algorithms tested are combinations of:

BT (Backtracking-Search)
FC (Forward-Checking)
MAC (Maintaining-Arc-Consistency)
MRV (Minimum-Remaining-Values)
LCV (Least-Constraining-Value)

Usage

$ python3 kakuro.py

Implementation

The csp.py, search.py and utils.py files are modified files of aimacode/aima-python.

The kakuro.py file is written by me and it's the modelization of Kakuro as a CSP and the main function.

Kakuro Puzzles

The Kakuro puzzles are taken from here and here.

Runtime and Number of Assignments

Easy, 4x3 Puzzle	Average Runtime (seconds)	Average Number of Assignments
BT + FC + MRV	0.00099	20
BT + MAC + MRV	0.00099	15
BT + FC + MRV + LCV	0.00096	15
BT + MAC + MRV + LCV	0.00401	33

Easy, 6x6 Puzzle	Average Runtime (seconds)	Average Number of Assignments
BT + FC + MRV	0.00697	72
BT + MAC + MRV	0.00797	46
BT + FC + MRV + LCV	0.00697	74
BT + MAC + MRV + LCV	0.01395	46

Easy, 14x14 Puzzle	Average Runtime (seconds)	Average Number of Assignments
BT + FC + MRV	2.02458	10027
BT + MAC + MRV	0.20046	214
BT + FC + MRV + LCV	1.33246	4949
BT + MAC + MRV + LCV	0.19248	215

Intermediate, 5x7 Puzzle	Average Runtime (seconds)	Average Number of Assignments
BT + FC + MRV	0.41786	1053
BT + MAC + MRV	0.50165	45
BT + FC + MRV + LCV	0.38500	862
BT + MAC + MRV + LCV	0.43480	45