configure.conf

[ENV]
env_name = Catcher
env_num = 50
capacity = 10000

[MONITOR]
wandb = 0
tensorboard = 0

[NETWORK]
layers = 3
hidden_dims = 128,128

[DQN]
# the max iter of training
max_train_steps = 500
# the max steps in one game
max_eposide_step = 500
# the evaluation frequency
evaluate_freq = 10
# learning freqecny in one game
learn_freq = 10
# the evaluation times
evaluate_times = 3
# learning rate
dqn_lr = 0.001
# discounted parameter
gamma = 0.9
epsilon = 0.4
epsilon_min = 0.001
epsilon_decay = 0.0001
# batch size
batch_size = 10000
mini_batch_size = 256
# use lr decay trick
use_lr_decay = True
# the frequency to update target network
update_target = 200

[PPO]
max_train_steps = 500  
per_batch_steps = 1000  
evaluate_freq = 20  
save_freq = 20  
batch_size = 10000
mini_batch_size = 512   
lr_a = 0.005  
lr_c = 0.0002  
gamma = 0.98  
lamda = 0.95  
epsilon = 0.2     
use_gae = True  
grad_clip_param = 0.5  
use_adv_norm = True  
use_state_norm = False
use_reward_norm = False  
use_reward_scaling = False 
entropy_coef = 0.05  
use_lr_decay = True  
use_grad_clip = True  
use_orthogonal_init = True  
use_ppo_clip = True 


[DDPG]
# the max iter of training
max_train_steps = 500
# the max steps in one game
max_eposide_step = 500
# the evaluation frequency
evaluate_freq = 10
# learning freqecny in one game
learn_freq = 10
# the evaluation times
evaluate_times = 3
# learning rate
dqn_lr = 0.001
# discounted parameter
gamma = 0.9
tau = 0.5
# batch size
batch_size = 10000
mini_batch_size = 256
# use lr decay trick
use_lr_decay = True
# the frequency to update target network
update_target = 200