Sudoku Solver

This is a simple project about making an automatic Sudoku solver. It can be run as a CLI tool, by running:

dotnet run --project SudokuSolver --configuration Release -- --board {BOARD} --solver {SOLVER} --solution {SOLUTIONFILE}

Where:

BOARD has to be a string of rows in the sudoku (total of 81 numbers), where 0 represents a blank space
SOLVER is the solver you want. The CLI tool will tell you what options are available
SOLUTION is optional. It describes the file you want the solved board to be outputted to. If none is given, no file will be made.

Solvers

There are four core solvers in this projects, three brute force (B) and one logical (L). There are also presets that combine some of these solvers, to simplify the job for the following solver.

Random Backtrack (B)

This solver is a sort of a "baseline" solver. It is pretty bad, since it selects what cell to try and fill out in random order. This leads to a lot of weird and bad branches that simply requires more time and a lot more calls.

Sequential Backtrack (B)

Sequential backtrack selects cells sequentially, column by row. It can also be visualised in the following example, where the sequence is from A until the end of the board (thanks to SudokuWiki.org for the image)

Where it will try and assign all legal values to the A cell first, then to the B cell, etc. If a cell becomes impossible to assign to, we backtrack to the previous cell.

Cardinality Backtrack (B)

This is a backtrack solver that selects cells based on the cardinality of possibilities for rows. It sorts its possible cells to select first by fewest possibilities for the entire row and then sort it by the cell in that row that have the fewest assignment possibilities itself. As an example, one can look at the following image (the small numbers mean the numbers that can legally be assigned to that cell):

It can be seen that it is row F that have the fewest total possible assignments, so that row takes priority. When it comes to selecting a cell in that row, we take the cell in the row that have the fewest possibilities too. So in this case we will try F1 and F6 before we try F9.

Logical (L)

This is perhaps the most interesting solver of them all. It attempts to solve the board by means of logical deduction. The set of logical strategies that this solver currently implements are:

Many of the boards cant be solved by logical deduction (at least with the current strategies) but paired with any of the backtrack solvers makes a powerful solver.

Performance

Benchmark is run on 1971 different Sudoku boards with a 2s time limit. The results are ordered by solved instances, then by lowest average search time and finally by average calls.

Solver	Solved	Avg Search (ms)	Avg Calls	Max Search (ms)	Min Search (ms)	Max Calls	Min Calls
LogicalWithCardinalityBacktrack	1971	3.92	28287.67	145.72	0.2	1156304	2
LogicalWithSequentialBacktrack	1971	5.3	43340.56	210.07	0.19	1903421	2
CardinalityBacktrack	1971	11.6	203805.84	1889.73	0.02	32884160	42
SequentialBacktrack	1967	22.48	421545.84	1834.6	0.02	34058393	43
LogicalWithRandomBacktrack	1383	61.89	17299889.9	1999.05	0.16	63162086	2
Logical	1043	1.27	3.23	14.63	0.18	19	1
RandomBacktrack	390	460.34	41259946.14	2001.87	0.02	62526698	73

Credits

I would like to give a special thanks to Andrew Stuart, the owner of the SudokuWiki.org website, for giving a bunch of great description of the different strategies that are involved in solving sudokus!