grandmasters typically play a small number of (recorded) games per year, would it be possible to train a neural network on lots of games to recognize what were likely moves of a player from a small number of games - so that you could have a computer to work with that would prepare you for a match with a grandmaster?