def main():
# try:
parse_cmd_args()
sess = tf.Session()
K.set_session(sess)
db = Database()
env = Environment(db, argus)
actor_critic = ActorCritic(env, sess, learning_rate=argus['learning_rate'], train_min_size=argus['train_min_size'],
size_mem=argus['maxlen_mem'], size_predict_mem=argus['maxlen_predict_mem'])
num_trials = argus['num_trial'] # ?
# trial_len = 500 # ?
# ntp
env.preheat()
# First iteration
cur_state = env._get_obs() # np.array (inner_metric + sql)
cur_state = cur_state.reshape((1, env.state.shape[0]))
# action = env.action_space.sample()
action = env.fetch_action() # np.array
action_2 = action.reshape((1, env.action_space.shape[0])) # for memory
new_state, reward, done, _ = env.step(action, 0, 1) # apply the action -> to steady state -> return the reward
new_state = new_state.reshape((1, env.state.shape[0]))
reward_np = np.array([reward])
print("0-shape-")
print(new_state.shape)
actor_critic.remember(cur_state, action_2, reward_np, new_state, done)
actor_critic.train() # len<32, useless
cur_state = new_state
for i in range(num_trials):
# env.render()
cur_state = cur_state.reshape((1, env.state.shape[0]))
action, isPredicted = actor_critic.act(cur_state)
print(action)
action_2 = action.reshape((1, env.action_space.shape[0])) # for memory
# action.tolist() # to execute
new_state, reward, done, _ = env.step(action, isPredicted, i + 1)
new_state = new_state.reshape((1, env.state.shape[0]))
reward_np = np.array([reward])
print("%d-shape-" % i)
print(new_state.shape)
actor_critic.remember(cur_state, action_2, reward_np, new_state, done)
actor_critic.train()
cur_state = new_state
'''
sess = tf.Session()
K.set_session(sess)
db = Database(argus) # connector knobs metric
env = Environment(db, argus)
actor_critic = ActorCritic(env, sess, learning_rate=argus['learning_rate'], train_min_size=argus['train_min_size'],
size_mem=argus['maxlen_mem'], size_predict_mem=argus['maxlen_predict_mem'])
num_trials = argus['num_trial'] # ?
# trial_len = 500 # ?
# ntp
env.preheat()
# First iteration
cur_state = env._get_obs() # np.array (inner_metric + sql)
cur_state = cur_state.reshape((1, env.state.shape[0]))
# action = env.action_space.sample()
action = env.fetch_action() # np.array
action_2 = action.reshape((1, env.action_space.shape[0])) # for memory
new_state, reward, done, socre, _ = env.step(action, 0, 1) # apply the action -> to steady state -> return the reward
new_state = new_state.reshape((1, env.state.shape[0]))
reward_np = np.array([reward])
print("0-shape")
print(new_state.shape)
actor_critic.remember(cur_state, action_2, reward_np, new_state, done)
actor_critic.train() # len<32, useless
cur_state = new_state
predicted_rewardList = []
