def load_model(filename):
model = DQN()
checkpoint = torch.load(filename)
model.load_state_dict(checkpoint['state_dict'])
return model
"""
def __init__(self, size):
self.size = size
self.memory = deque(maxlen=self.size)
def update(self, SARS):
self.memory.append(SARS)
def sample(self, batch_size):
return zip(*random.sample(self.memory, batch_size))
r_memory = ReplayMemory(memory_size)
agent = DQN(12, 12, 16)
target = DQN(12, 12, 16)
target.load_state_dict(agent.state_dict())
optimizer = Adam(agent.parameters())
def update_target():
if len(r_memory.memory) < batch_size:
return
observation, action, reward, observation_next, done = r_memory.sample(
batch_size)
observations = torch.cat(observation)
observation_next = torch.cat(observation_next)
actions = index_action(torch.LongTensor(action))
rewards = torch.LongTensor(reward)
done = torch.FloatTensor(done)
q_values = agent(observations)
p_q_values_next = agent(observation_next)