The sign of virtual loss is reversed

[mokemokechicken]

virtual_loss_for_w must be decided before env.step(action).

reversi-alpha-zero/src/reversi_zero/agent/player.py

Lines 180 to 184 in a09d90b

	action_t = self.select_action_q_and_u(env, is_root_node)
	_, _ = env.step(action_t)
	virtual_loss = self.config.play.virtual_loss
	virtual_loss_for_w = virtual_loss if env.next_player == Player.black else -virtual_loss

This is a critical bug which makes MCTS search more same moves.