def __init__(self):
self._cpu_collector = CpuCollector()
self._disk_collector = DiskCollector()
self._memory_collector = MemoryCollector()
self._network_collector = NetworkCollector()
# define model
model = torch.hub.load('andompesta/ppo2',
'ppo2',
reset_param=True,
force_reload=True,
input_dim=obs_size,
hidden_dim=args.hidden_dim,
action_space=action_space,
dropout=0)
model.to(device)
# setup training function
train_fn, optm = step_setup(args, model, device)
# create memory collector for different episode. Used for batch training
memory_collector = MemoryCollector(env, model, args.n_step, args.gamma,
args.lam, device)
ep_info_buf = deque(maxlen=100)
n_env = 1
n_batch = n_env * args.n_step
n_updates = args.total_timesteps // args.batch_size
n_batch_train = n_batch // args.mini_batchs
for update in range(1, n_updates + 1):
assert n_batch % args.mini_batchs == 0
# Start timer
frac = 1.0 - (update - 1.0) / n_updates
if update % args.log_every == 0:
print('Stepping environment...')
