#Creating a request for initializing a map, initial position, the initial energy, and the maximum number of steps of the DQN agent
request = ("map" + str(mapID) + "," + str(posID_x) + "," +
str(posID_y) + ",50,100")
#Send the request to the game environment (GAME_SOCKET_DUMMY.py)
minerEnv.send_map_info(request)
# Getting the initial state
minerEnv.reset() #Initialize the game environment
s = minerEnv.get_state() #Get the state after reseting.
#This function (get_state()) is an example of creating a state for the DQN model
total_reward = 0 #The amount of rewards for the entire episode
terminate = False #The variable indicates that the episode ends
maxStep = minerEnv.state.mapInfo.maxStep #Get the maximum number of steps for each episode in training
#Start an episde for training
for step in range(0, maxStep):
action = DQNAgent.act(
s) # Getting an action from the DQN model from the state (s)
minerEnv.step(
str(action)
) # Performing the action in order to obtain the new state
s_next = minerEnv.get_state() # Getting a new state
reward = minerEnv.get_reward() # Getting a reward
terminate = minerEnv.check_terminate(
) # Checking the end status of the episode
# Add this transition to the memory batch
memory.push(s, action, reward, terminate, s_next)
# Sample batch memory to train network
if (memory.length > INITIAL_REPLAY_SIZE):
#If there are INITIAL_REPLAY_SIZE experiences in the memory batch
#then start replaying
total_reward = 0 #The amount of rewards for the entire episode
terminate = False #The variable indicates that the episode ends
maxStep = minerEnv.state.mapInfo.maxStep #Get the maximum number of steps for each episode in training
average_loss = 0
#Here: maxStep =100 line 64 GAME_SOCKET_DUMMY
#Start an episde for training
for step in range(0, maxStep):
#s.shape =(MAP_MAX_X, MAP_MAX_Y) moi phan tu la gia tri score VD 0, -1, 100, 250, -3
if args.load_model != "":
action = np.argmax(DQNAgent.model.predict(s.reshape(1,len(s))))
else:
# _id = np.random.randint(1,5)
# if _id == 1:
# action = DQNAgent.act(s) # Getting an action from the DQN model from the state (s)
# else: action = bots[_id-2].act_sample(s)
action = DQNAgent.act(s)
#action int, VD action =3
minerEnv.step(str(action)) # Performing the action in order to obtain the new state
s_next = minerEnv.get_state2(limit) # Getting a new state
#s_next tuong tu s
reward = minerEnv.get_reward() # Getting a reward
#reward int, VD 0
terminate = minerEnv.check_terminate() # Checking the end status of the episode
#terminal True or False
# Add this transition to the memory batch
memory.push(s, action, reward, terminate, s_next)
# Sample batch memory to train network
if (memory.length > INITIAL_REPLAY_SIZE):
#If there are INITIAL_REPLAY_SIZE experiences in the memory batch
#then start replaying
