Deep Reinforcement Learning sudoku solver
An attempt to adapt the Google DeepMind approach to learning to play Atari games to the task of solving sudokus. Currently seems to make some progress in that sudokus are solved in gradually fewer steps, but some form of "catastrophic forgetting" appear to kick in after a while, so WIP for the time being...
Much inspiration taken from https://github.com/devsisters/DQN-tensorflow