Satisfaction Frontier In my last article, I decided to review the world’s fastest sudoku solver and the corresponding code and see if I could learn a thing or two. It was very informative. You should read the summary as it contains amazing information that I can not possibly do a great job of summarizing myself. Some of the key insights that I took from that article was: Send as many bits to the processor as you can.

What happened? So after publishing my last article, I did some research and found some great information online about solving Sudoku (specifically, solving hard Sudoku puzzles as fast as possible). I was very happy with my solver that could solve one of the 9 million sudoku puzzles (from Kaggle) in an average of 131μs; however, this was not even close to good enough on the scale of what was possible. In terms of a computer vision application, this performance level is most likely adequate; however, after reading about the possibility of much better performance, I was nerd-sniped and decided I needed to improve my code.

Overview For my first task, I need to create a base library that understands the Sudoku game. This library will allow you to create a board with the current state, solve a board, provide a hint, generate a puzzle of a given difficulty. Once I have done those base operation, I will need to create a way to persist a board. I will use a flat file system, so I will need a serializer/de-serializer.