class DQNAgent(Agent):
def __init__(self, config, save_name=None):
super().__init__(config, save_name)
def load_model(self, model_path):
self.model = load_model(model_path)
def init_env(self):
self.env = gym.make(self.env_name,
csv_list=self.csv_list,
trading_cost=self.trading_cost,
time_cost=self.time_cost,
market=self.market)
self.env = MyWrapper(self.env)
np.random.seed(self.seed)
self.env.seed(self.seed)
self.env.action_space.seed(self.seed)
random.seed(self.seed)
tf.random.set_random_seed(self.seed)
self.steps_per_episode = int(self.env.processed_array.shape[0] *
self.split_ratio)
self.nb_actions = self.env.action_space.n
def create_agent(self):
session_conf = tf.ConfigProto(intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1)
sess = tf.Session(graph=tf.get_default_graph(), config=session_conf)
K.set_session(sess)
self.model = Model.simple(self.env)
memory = SequentialMemory(limit=self.memory_len, window_length=1)
policy = EpsGreedyQPolicy(eps=self.eps)
self.agent = DQNAgent(model=self.model,
nb_actions=self.nb_actions,
memory=memory,
nb_steps_warmup=self.steps_per_episode,
target_model_update=self.target_model_update,
policy=policy)
self.agent.compile(Adam(lr=self.lr))
def predict(self, df):
data = self.env.fe.transform(df.iloc[:]).values
val = np.argmax(self.agent.select_action(data[-1:], do_train=False))
self.signal_transform(val)
def signal_transform(self, val):
if val == 0:
print('Buy')
elif val == 1:
print('Sell')
else:
print('Close/Hold')
