14.2.4. 設定 hyper parameters 超參數

本教程是使用 DQN 方法來學習的。DQN 是一種強化學習方法，通過近似動作值函數（Q值）來選擇深度神經網絡。Agent 會遵照在 /turtlebot3_machine_learning/turtlebot3_dqn /nodes/turtlebot3_dqn_stage＃中的 hyper parameters 超參數。

Hyper parameter	預設值	說明
episode_step	6000	The time step of one episode.
target_update	2000	Update rate of target network.
discount_factor	0.99	Represents how much future events lose their value according to how far away.
learning_rate	0.00025	Learning speed. 如果該值太大，學習效果不好，如果太小，學習時間就會很長。
epsilon	1.0	The probability of choosing a random action.
epsilon_decay	0.99	Reduction rate of epsilon. When one episode ends, the epsilon reduce.
epsilon_min	0.05	The minimum of epsilon.
batch_size	64	Size of a group of training samples.
train_start	64	Start training if the replay memory size is greater than 64.
memory	1000000	The size of replay memory.