'clip_range': (0.05, 0.3), # range of the adapted penalty factor
'scale_constant': 1.2,
},
},
'replay': {
# 'replay_class': 'FIFOReplay',
'batch_size': 64,
'memory_size': 96,
'sampling_start_size': 64,
'replay_shards': 1,
},
'parameter_publish': {
'exp_interval': 4096,
},
})
PPO_DEFAULT_LEARNER_CONFIG.extend(BASE_LEARNER_CONFIG)
PPO_DEFAULT_ENV_CONFIG = Config({
'env_name': '',
'action_repeat': 1,
'pixel_input': False,
'use_grayscale': False,
'use_depth': False,
'frame_stacks': 1,
'sleep_time': 0,
'video': {
'record_video': False,
'save_folder': None,
'max_videos': 500,
'record_every': 5,
},
},
'replay': {
'batch_size': 512,
'memory_size':
int(1000000 /
3), # The total replay size is memory_size * replay_shards
'sampling_start_size': 3000,
'replay_shards': 3,
},
'parameter_publish': {
# Minimum amount of time (seconds) between two parameter publish
'min_publish_interval': 3,
},
})
DDPG_DEFAULT_LEARNER_CONFIG.extend(BASE_LEARNER_CONFIG)
DDPG_DEFAULT_ENV_CONFIG = Config({
'env_name': '_str_',
'num_agents': '_int_',
'demonstration': None,
'use_depth': False,
'render': False,
'use_demonstration': False,
# If true, DDPG will expect an image at obs['pixel']['camera0']
'pixel_input': False,
'use_grayscale': False,
# Stacks previous image frames together to provide history information
'frame_stacks': 3,
# Each action will be played this number of times. The reward of the consecutive actions will be the the reward
# of the last action in the sequence