Analysis of Hyper-Parameters for Alphazero-Like Deep Reinforcement Learning