def main(env_id, embedding_size):
env = wrap_deepmind(make_atari(env_id), scale=True)
embedding_model = DQN(embedding_size)
agent = NECAgent(env, embedding_model)
# subprocess.Popen(["tensorboard", "--logdir", "runs"])
configure("runs/pong-run")
for t in count():
if t == 0:
reward = agent.warmup()
else:
reward = agent.episode()
print("Episode {}\nTotal Reward: {}".format(t, reward))
log_value('score', reward, t)
# trajectory is finished
next_state = np.zeros(len(self.cur_state))
done = 1
self.cur_state=next_state
return next_state,reward,done,action
env=Pseudo_env(df)
embedding_model = Embed(len(feature_fields),32)
agent = NECAgent(env, embedding_model,batch_size=32,sgd_lr=1e-5)
for t in count():
if t < 100:
reward = agent.warmup()
else:
reward = agent.episode()
print("Episode {}\nTotal Reward: {}".format(t, reward))
test_df = pd.read_csv('HFpEF data/aim3data_test_set.csv')
a = test_df.copy()
num = np.size(a,0)
patient_num = np.size(pd.unique(a['EMPI']))
from torch import Tensor
from torch.autograd import Variable
embedding_model.eval()
import pickle
filename = 'evaluate data/aim3state_test.data'