def __init__(self, *args, **kwargs):
super(A3CSpliterEnv, self).__init__(*args, **kwargs)
random.seed(time.time())
self.game_no = 0
self.out_classes = 9
self.a3c_model = ActorCritic(5, self.out_classes, 64)
self.optimizer = optim.SGD(self.a3c_model.parameters(), lr=0.1)
self.reset()
self.gamma = 0.99
self.tau = 1.0
self.entropy_coef = 0.01
self.batch_size = 256
self.buffer_size = 1000
self.i = 0
self.nll_loss_fn = nn.NLLLoss()
self.sw = SpliterWarpper(5, 128)
def __init__(self, *args, **kwargs):
super(A3CSpliterEnv, self).__init__(*args, **kwargs)
random.seed(time.time())
self.game_no = 0
self.out_classes = 9
self.a3c_model = ActorCritic(5, self.out_classes, 128)
self.a3c_model.load_state_dict(
torch.load('F:\\Dota2BotStepByStep\\tmp\\model_3023_24392'))
self.optimizer = optim.SGD(self.a3c_model.parameters(), lr=0.01)
self.reset()
self.gamma = 0.99
self.tau = 1.0
self.entropy_coef = 0.01
self.batch_size = 256
self.buffer_size = 1000
self.i = 0
self.nll_loss_fn = nn.NLLLoss()
self.sw = SpliterWarpper(5, 128)
def __init__(self, *args, **kwargs):
super(DoubleA3CPPOEnv, self).__init__(*args, **kwargs)
random.seed(os.getpid())
torch.random.manual_seed(os.getpid())
self.game_no = 0
self.out_classes = 9
self.a3c_model = ActorCritic(5, self.out_classes, 64)
self.optimizer = optim.SGD(self.a3c_model.parameters(), lr=0.1)
self.reset()
self.gamma = 0.99
self.tau = 1.0
self.entropy_coef = 0.01
self.epsilon = 0.2
self.batch_size = 256
self.buffer_size = 1000
self.i = 0
self.update_steps = 3
self.nll_loss_fn = nn.NLLLoss()
self.writer = SummaryWriter(comment='_%d' % os.getpid())
self.rank = -1
v.env = env
v.env.canvas = v.canvas
v.visualize()
gen = env.run(True)
gen.send(None)
while True:
try:
gen.send((None, ))
except StopIteration:
#barrier.wait()
torch.save(env.a3c_model.state_dict(),
"./tmp/model_%d_%d" % (env.game_no, os.getpid()))
#barrier.wait()
gen = env.run(True)
gen.send(None)
if __name__ == '__main__':
model = ActorCritic(5, 9, 64)
import torch.nn.init as weight_init
for name, param in model.named_parameters():
weight_init.normal(param)
model.share_memory()
barrier = mp.Barrier(2)
parallel.start_parallel(DoubleA3CPPOEnv,
model,
np=2,
func=test_without_gui,
args=barrier)
env = clazz()
env.set_model(model)
env.set_rank(rank)
gen = env.run(True)
gen.send(None)
while True:
try:
gen.send((None, ))
except StopIteration:
if rank == 0:
torch.save(env.a3c_model.state_dict(),
"./tmp/model_%d_%d" % (env.game_no, os.getpid()))
gen = env.run(True)
gen.send(None)
if __name__ == '__main__':
model = ActorCritic(5, 9, num_hidden)
import torch.nn.init as weight_init
for name, param in model.named_parameters():
weight_init.normal(param)
model.share_memory()
#test(model)
parallel.start_parallel(ParallelA3CPPOEnv,
model,
np=os.cpu_count(),
func=test_without_gui,
args=None)