neural network that gets a game turn and game state as input

Question

I have seen this video: https://youtu.be/v9M2Ho9I9Qo?t=49

It’s about creating AI for “GO”

at 0:50 he talks about feeding a neural network a game state of go and a possible move of a player.

My question: what are the best way to feed the game state and the move. I know I can just feed the neural net all the game states after each possible move. But in the video, he said he feeds the move with the board state before the move.

how should it be done?

Answer 1

I think you need to understand policy-based methods. In policy-based methods we are trying to learn directly the policy function that maps state to action.(policy gradients and actor-critic method)