def execute_agent(self):
keys = Keys()
screen = Screen()
print('Starting running Direct Imitation Learning in...')
# Countdown to start running the agent
for i in list(range(4))[::-1]:
print(i + 1)
time.sleep(1)
paused = False
while True:
if not paused:
img = screen.GrabScreenBGR()
# Preprocessing input image
converted_img = np.array(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY))
resized_img = cv2.resize(converted_img, (120, 90))
reshaped_img = resized_img.reshape(-1, 90, 120, 1)
normalized_img = reshaped_img.astype('float32') / 255.0
final_img = normalized_img
# Prediction and Entropy of the Labels
move_prediction = self.predict_move(final_img)
action_prediction = self.predict_action(final_img)
move_action = np.argmax(move_prediction[0])
control_action = np.argmax(action_prediction[0])
print(move_prediction[0], ' - ', action_prediction[0])
# Move action
self.execute_movement_action(move_action)
#self.execute_movement_action_with_threshold(move_prediction[0])
# Action action
self.execute_control_action(control_action)
#self.execute_control_action_with_threshold(action_prediction[0])
keys_pressed = keys.KeyCheck()
if 'Q' in keys_pressed:
if paused:
paused = False
print('unpaused!')
time.sleep(1)
else:
print('Pausing!')
self.release_moves()
paused = True
time.sleep(1)
def CreateTrainingData(self):
keys = Keys()
screen = Screen()
print('Starting Training in...')
# Countdown to start the training
for i in list(range(4))[::-1]:
print(i + 1)
time.sleep(1)
paused = False
last_time = time.time()
while True:
if not paused:
grabbed_screen = screen.GrabScreenBGR() # Get actual frame
new_screen = cv2.resize(
cv2.cvtColor(grabbed_screen, cv2.COLOR_BGR2GRAY),
(120, 90)) # Converted and Resized frame
#normalized_screen = new_screen.astype('float32') / 255 # Normalizing
keys_pressed = keys.KeyCheck() # Check for pressed keys
output_move = keys.KeysMovementOutput(
keys_pressed) # Verifies if one move key was pressed
output_action = keys.KeysActionOutput(
keys_pressed) # Verifies if one action key was pressed
self.training_data.append(
[new_screen, output_move,
output_action]) # Create an instance of training data
if len(self.training_data) % 1000 == 0:
print(len(self.training_data))
keys_pressed = keys.KeyCheck()
# Pausing or Unpausing training
if 'Q' in keys_pressed:
if paused:
paused = False
print('Unpausing training...')
time.sleep(2)
else:
print('Pausing training!')
paused = True
time.sleep(1)
# Saving Data
if 'P' in keys_pressed:
np.save(self.path_file, self.training_data)
def execute_active_deep_imitation_learning_movement_actions(self):
keys = Keys()
screen = Screen()
print('Starting running Active Deep Imitation Learning in...')
# Countdown to start running the agent
for i in list(range(4))[::-1]:
print(i + 1)
time.sleep(1)
paused = False
number_active_samples_to_reach = self.active_sample_size * len(
self.dataset_behavioral_cloning.training_data) / 100.0
print(len(self.active_samples.training_data))
print(number_active_samples_to_reach)
while len(self.active_samples.training_data
) < number_active_samples_to_reach:
if not paused:
img = screen.GrabScreenBGR()
# Preprocessing input image
converted_img = np.array(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY))
resized_img = cv2.resize(converted_img, (120, 90))
reshaped_img = resized_img.reshape(-1, 90, 120, 1)
normalized_img = reshaped_img.astype('float32') / 255
final_img = normalized_img
# Prediction and Entropy of the Labels
move_prediction = self.behavioral_agent.predict_move(final_img)
action_prediction = self.behavioral_agent.predict_action(
final_img)
move = np.argmax(move_prediction[0])
action = np.argmax(action_prediction[0])
entropy = self.calculate_entropy(action_prediction[0])
print(move_prediction[0], ' -> ', entropy)
if entropy < self.threshold:
## Execute the behavioral agent's action
# Move action
self.behavioral_agent.execute_movement_action(move)
# Control action
self.behavioral_agent.execute_control_action(action)
else:
## Queries a non-expert action
self.query_sound()
print('Query non-expert action: ')
self.behavioral_agent.release_moves()
# Waiting for a non-expert action
while True:
keys_pressed = keys.KeyCheck(
) # Check for pressed keys
non_expert_move = keys.KeysMovementOutput(
keys_pressed
) # Verifies if one move key was pressed
if non_expert_move != [0, 0, 0, 0]:
print(non_expert_move)
self.active_samples.training_data.append(
[resized_img, non_expert_move])
break
keys_pressed = keys.KeyCheck()
if 'Q' in keys_pressed:
if paused:
paused = False
print('unpaused!')
time.sleep(1)
else:
print('Pausing!')
self.behavioral_agent.release_moves()
paused = True
time.sleep(1)
self.active_samples.save_data()
keys = Keys()
screen = Screen()
print('Starting running in...')
# Countdown to start running the agent
for i in list(range(4))[::-1]:
print(i + 1)
time.sleep(1)
paused = False
while True:
if not paused:
new_screen = screen.GrabScreenBGR()
new_screen = cv2.resize(new_screen, (140, 80))
move_prediction = agent.predict_move(
[new_screen.reshape(1, 140, 80, 3)])
action_prediction = agent.predict_action(
[new_screen.reshape(1, 140, 80, 3)])[0]
moves = list(np.around(move_prediction[0]))
print(moves)
print(action_prediction)
if moves == [0, 0, 0, 1]:
agent.right()
elif moves == [0, 0, 1, 0]:
agent.down()
elif moves == [0, 1, 0, 0]:
