def __init__(self, agent, env=None):
self.lock = th.Lock()
self.agent = agent
# one and only
self.para_list = self.get_parameter_list()
self.farmer = farmer_class(self.para_list)
self.ep_num = 0
self.total_steps = 0
self.history_reward = []
self.ep_value = {}
self.value_init()
self.relative_time = 0
self.average_len_of_episode = self.agent.args.max_pathlength
self.num_rollouts = int(self.agent.args.timesteps_per_batch /
self.average_len_of_episode)
self.rollout_count = 0
self.rollout_paths = []
self.iteration = 0
self.log_scalar_name_list = [
'reward', 'kl_div', 'entropy', 'surrogate_loss', 'value_loss'
]
self.log_scalar_type_list = [
tf.float32, tf.float32, tf.float32, tf.float32, tf.float32
]
self.logger = Logger(self.agent.session,
self.agent.args.log_path + 'train',
self.log_scalar_name_list,
self.log_scalar_type_list)
self.write_log = self.logger.create_scalar_log_method()
self.start_time = time.time()
def __init__(self, agent, env=None):
self.lock = th.Lock()
self.agent = agent
# one and only
self.para_list = self.get_parameter_list()
self.farmer = farmer_class(self.para_list)
self.ep_num = 0
self.total_steps = 0
self.history_reward = []
self.ep_value = {}
self.value_init()
self.relative_time = 0
self.average_steps = self.agent.args.max_pathlength
self.start_time = time.time()
def __init__(self, agent, env=None):
self.lock = th.Lock()
self.agent = agent
# one and only
self.farmer = farmer_class(self.agent.para_list)
self.ep_num = 0
self.total_steps = 0
self.history_reward = []
self.ep_value = {}
self.value_init()
self.relative_time = 0
self.average_steps = self.agent.para_list["max_pathlength"]
self.log_scalar_name_list = [
'mean_reward', 'actor_loss', 'critic_loss'
]
self.log_scalar_type_list = [tf.float32, tf.float32, tf.float32]
self.logger = Logger(self.agent.session,
self.agent.para_list["log_path"] + 'train',
self.log_scalar_name_list,
self.log_scalar_type_list)
self.write_log = self.logger.create_scalar_log_method()
self.start_time = time.time()
def refarm(self): # most time no use
del self.farmer
self.farmer = farmer_class()