dyln / hillclimb-sudokusolver

Tries to solve a 9x9 Sudoku using local search hill climb

Sudoku Solver

This is a Scala implementation of a Sudoku solver for a 9x9 puzzle that uses a local search Hill Climb to find a solution. It will choose a better fitness cost with a probability of .995

Objective function: Number of repeated numbers in all the rows plus number of repeated numbers in each quadrant. All columns get initialized to already satisfy the property of containing numbers 1 through 9.

Transformation function: Implemented by doing a swap of numbers in the same column yet in distinct rows. A random column number is picked, then two random rows are selected with the only requirement that they must be different.

Stopping Criteria: If the number of iterations exceeds 5 million then the Hill Climb will terminate even though a solution may not have been found on the given input for the problem.

Requirements

java 1.6 or higher

sbt 0.12.2

Building

Change into main directory and run:

1. sbt reload update
2. sbt compile
3. sbt run <filename> where <filename> is any custom file in the repository. Look at *.in files for examples on the format

Testing

Run tests from main directory:

sbt test

Benchmarks

Below are some benchmarks on the number of total iterations to 
convergence by varying the probability of choosing a 'worse' move 
for easy inputs (easy*.in) and hard inputs (difficult*.in)
  
Convergence

Easy Input
.001  => 1,608,918 ; 
.005  => 647,143 ; 211,313 ; 126,902 ; 434,617 ; 335,546 ; 627,381
.006  => 1,338,593
.0003 => 399,604 ; 5,131,079 ; nc (Over 5 min) 
.0004 => 54,692 ; 2,340,466
.0005 => 1,078,154
.0006 => 2,088,491

Difficult Input
.01   => nc (Over 10 min) ; nc (Over 10 min)
.001  => 70,373 ; nc
.003  => 1,342,784
.004  => 2,318,363
.005  => 797,134 ; 1,025,887 ; 3,394,429 ; 2,826,606
.0006 => 193,541
.0008 => 1,722,199

* nc - no convergence