#action-space

Question 1: Single choice

A policy network for Connect Four takes the board as input and has 7 output nodes, one probability per column the agent could play. You build the same kind of network for chess. How many output nodes should it have?

4672, one for each move in a fixed list of possible moves1
2, one for each of the two players in a game2
16, one for each of the pieces a player starts the game with3
64, one for each square of the chess board4

Timed test