def __init__(self, args, env):
self.args = args
self.env = env
self.evolver = utils_ne.SSNE(self.args)
self.best_r = 0
self.best_state = []
#Init population
self.pop = []
for _ in range(args.pop_size):
self.pop.append(ddpg.Actor(args))
#Turn off gradients and put in eval mode
for actor in self.pop:
actor.eval()
#Init RL Agent
self.rl_agent = ddpg.DDPG(args)
self.replay_buffer = replay_memory.ReplayMemory(args.buffer_size)
self.ounoise = ddpg.OUNoise(args.action_dim)
#Trackers
self.num_games = 0
self.num_frames = 0
self.gen_frames = None
def __init__(self, args, env):
self.args = args
self.evolver = utils_ne.SSNE(self.args)
# self.replay_buffer = replay_memory.ReplayMemory(args.buffer_size)
self.pop = []
for _ in range(args.pop_size):
self.pop.append(ddpg.Actor(args))
for actor in self.pop: actor.eval()
# self.workers = [Worker.remote(args) for _ in range(self.args.pop_size+1)]
# args.is_cuda = True; args.is_memory_cuda = True
self.rl_agent = ddpg.DDPG(args)
# self.rl_agent.share_memory()
self.ounoise = ddpg.OUNoise(args.action_dim)
# self.replay_queue = mp.Manager().Queue() # mp.Manager().list()
# self.replay_queue = mp.Queue()
# self.replay_memory = mp.Manager().list()
# self.replay_memory = mp.Array()
# self.replay_memory = mp.Queue()
self.workers = self.pop.append(self.rl_agent.actor)
# for key in range(self.args.pop_size):
# self.replay_memory[key] = replay_memory.ReplayMemory(self.args.buffer_size)
# self.learner = LearnerThread(self.replay_memory, self.rl_agent)
# self.learner.start()
# Stats
# self.timers = {
# k: TimerStat()
# for k in [
# "put_weights", "get_samples", "sample_processing",
# "replay_processing", "update_priorities", "train", "sample"
# ]
# }
self.num_games = 0; self.num_frames = 0; self.gen_frames = 0; self.len_replay = 0
# print()
if not self.replay_memory.empty():
print(self.replay_memory.qsize())
time.sleep(1)
print(self.replay_memory.get_nowait())
if __name__ == "__main__":
args = Parameters()
env = utils.NormalizedActions(gym.make(env_tag))
args.action_dim = env.action_space.shape[0]
args.state_dim = env.observation_space.shape[0]
pop = []
for _ in range(1):
pop.append(ddpg.Actor(args))
for actor in pop:
actor.eval()
mp.set_start_method('spawn')
replay_memory = mp.Queue()
learner = LearnerThread(replay_memory)
learner.start()
# dict_all_returns = mp.Manager().dict()
processes = []
time_start = time.time()
for key, pop in enumerate(pop):
pop.share_memory()