#monte-carlo-rollouts

Question 1: Single choice

An Othello player plans with Monte Carlo rollouts, using a trained policy network and a value network. How does it choose a move?

For each of the few likeliest moves, it samples short continuations with the policy, scores their ends with the value net, and takes the best average1
It plays the move with the highest policy probability and uses one rollout only to check that the move is legal2
It searches every legal move exhaustively to the end of the game and plays the guaranteed winner3
It keeps training the policy network during each game, using the rollout outcomes as new labels4