CS116 -- Artificial Intelligence
Fall, 2006
Search-based Sudoku Solver
(last updated 10/14/2006)

Design and implement an "agent" that solves a 9x9 Sudoku puzzle.  A given puzzle starts with a few of the cells of the grid filled in with numbers.  These cannot change.  To solve the puzzle you must fill in all the blank cells with numerals 1-9 such that each row and column has one and only one of each of the numerals 1-9.  In addition, each main 3x3 sub-grid must also have one and only one of each of the numbers.  

Your system should take as input a text file consisting of eleven rows of characters -- nine data rows and two filler rows.  Each data row is a space-delimitted sequence of numerals or underscores with pipe characters (vertical bar) separating 3x3 sub-grids.  The two filler rows form horizontal separation between sub-grids.

Your agent should output a completed puzzle where all underscores have been replaced with the appropriate numeral (while abiding by the constraints of the puzzle rules).  Your output must maintain the visual spacing (filler rows and pipe chars).

Here is an example input with the corresponding (correct) output.

Solution method
I want you to implement an informed, or heuristic, search method for this project.  You are free to use any of the heuristic methods that we have discussed in class.  If you want to use another search method that we have not yet covered or will not be covering, be sure to consult with me first.  Note, I want you to implement your search routine.  Please don't download an implementation.  This is an easy program to write; we went over the schema for all the search methods on a single slide in class.  I am expecting that you would spend more time creating and implementing your heurstic (i.e., thinking about what features of the problem will be useful in distinguishing between good and bad partial-solutions) than you will on the search routine itself. 

Thinking about this in terms of state-space search, the initial state is specified by the input puzzle, you transform one state to another by writing a number in an unmarked cell or by changing the number of a cell that you previously wrote (be careful not to change any cells that are specified by the input), and the goal state is a 9x9 array of numbers that satisfy the constraints of the solution.

In order to measure your search effectiveness, you should instrument your code with several metrics.  These must include (but are not limited to) (a) total number of positions placed on the closed list, (b) total number of positions placed on the open list, (c) maximum size of the open list during the course of solving the puzzle, and (d) final solution path length.  (Note, the solution path for these puzzles will always be the number of underscores that need to be replaced with numbers, but your code should be written such that if applied in another context it would report the appropriate path length.)

The goals of this assignment are:
  1. gain experience implementing a search method
  2. think creatively about heuristics that can effectively guide the search in the Sudoku context
  3. establish a comparison for a later project that will use a logic-based solution method
  4. have fun and impress your friends and neighbors with the speed at which you can solve a Sudoku puzzle
Submission and grading
Submit your completed search agents to Eureka in the usual way (zip or tar file).  Include a readme file that explains how to run your agent, describes your heuristic and identifies it as admissible or not, and any other special considerations or features that I will need to know in order to test and grade your project.  I will grade this project based on (a) output solution correctness, (b) coding style, and (c) heuristic creativity.