Sudoku-solver it's a sandbox where I try to improve my Pythonic programming, using NumPy.
You need Python 3.6 and all the modules into requirements.txt.
To run the script you need a CSV file like this:
quizzes,solutions
004300209005009001070060043006002087190007400050083000600000105003508690042910300,864371259325849761971265843436192587198657432257483916689734125713528694542916378
where quizzes is the flatten version of a 9x9 sudoku where 0s are the missing cells and solutions is the complete sodoku.
Nowadays not all the sudoku can be solved by the program, so you can find a CSV style output on standard error containing the wrong row.
python -m sudoku.benchmark /tmp/sudoku.csv 123 2> /tmp/sudoku_ko.csv
I used 1 milion sudoku found at Kaggle to benchmark the program.
The machine is and old Intel Core Duo, 4 GB of RAM, CentOS.
The best result is: 2:51:06 h, 124 errors.