LyndonArmitage
commented
9 years ago The process would probably go something like this:
- Run a basic logic solver over the initial Sudoku to get it to a stage that requires some form of guessing.
- Take the Sudoku and create an initial population with random values in the questionable cells.
- Score the population based on validity
- Breed the population
- Repeat until a valid solution is found
Creating a global list of available numbers for the initial random stage could help reduce the number of invalid Sudokus created initially.
An exhaustive search might be just as quick as any GA algorithm and will need investigating at some point.
By coming up with some scoring mechanism; produce a Genetic Algorithm based Sudoku solver.
To do this will need to be able to score a solution based on how valid it is. This could be done by counting the amount of valid rows and columns.