Learning Attentional Communication with a Common Network for Multiagent Reinforcement Learning

Table 1

Training parameters of the MAACCN algorithm.

Parameters	Value	Descriptions
Lr	0.0005	The learning rate
Epsilon	1	Probability of exploration
Min_epsilon	0.05	Minimum probability of exploration
Anneal_steps	50000	The annealing steps of exploration
T_max	2000000	The total step size of training
N_episodes	1	The number of episodes sampled at an epoch
Evaluate_cycle	100	The interval of the evaluation cycle
Evaluate_epoch	32	Frequency of evaluation
Batch_size	32	The batch data size for training
Buffer_size	5000	The size of the buffer
Target_update_cycle	200	The update interval of the target network
hidden_dim	64	The dimension of a hidden layer
Head	8	The number of the multihead