Smart Snake and Ladder

In the normal snakes and ladders, the game is pure luck (the agent does not choose an action, just follows the dice). So I will require a modification to the rules where the player can choose how many states to advance, limited by the dice roll. In other words, if they roll a 3, then they can choose to advance 0, 1, 2, or 3 states. Instead I’d recommend depth-limited expectimax, using an evaluation function (similar to a heuristic) to estimate the risk/reward of states at the depth limit. Lastly, you can use the NN to predict risk/reward as its output.