def get_env(name):
if name == 'cartpole':
env = gym.make('CartPole-v0')
env._max_episode_steps = 500
return env
elif name == 'pong':
return pong.Pong(800, 600, int(400/2), int(200/2))
def __init__(self, width, height, frame_rate):
# title of the application is ""
game_mouse.Game.__init__(self, "Pong", width, height, frame_rate)
# create a Picture instance
self.mPong = pong.Pong(width, height)
return
def __init__( self, title, width, height, frame_rate ):
game.Game.__init__( self, title, width, height, frame_rate )
# create a game instance
# YOU SHOULD CHANGE THIS TO IMPORT YOUR GAME MODULE
self.mGame = pong.Pong( width, height )
return
button = digitalio.DigitalInOut(board.A3)
ui.Screen.setEncoder(encoder, button)
ui.Screen.setDisplay(d)
screen = ui.Screen(True)
button = ui.Button(screen, 38, 32, 38, 16, "Pong", func)
framerate = ui.Text(screen, 0, 8, int(1 / loopTime), max_glyphs=3)
num1 = ui.SingleDigitNumberSelector(screen, 116 - 24, 32)
num2 = ui.SingleDigitNumberSelector(screen, 116, 32)
num3 = ui.SingleDigitNumberSelector(screen, 140, 32)
num4 = ui.SingleDigitNumberSelector(screen, 140 + 24, 32)
pong = ui.Screen()
# back = ui.Button(pong, 128, 32, 38, 16, 11, "Back", func2)
p = game.Pong(d, encoder)
pPress = False
pmenu = ui.Screen()
resume = ui.Button(pmenu, 48, 32, 68, 16, "Resume", pResume)
restart = ui.Button(pmenu, 128, 32, 68, 16, "Restart", pRestart)
exit = ui.Button(pmenu, 256 - 48, 32, 68, 16, "Exit", pExit)
# Main Loop
while True:
# d.setFont(11)
framerate.label.text = int(1 / loopTime)
ui.update()
if ui.Screen.current == pong:
p.update()
if ui.Screen.button.value == 0:
screen = turtle.Screen()
screen.setup(width, height)
screen.title("Pong")
screen.bgcolor("black")
rectCors = ((-70, 10), (70, 10), (70, -10), (-70, -10))
screen.register_shape("rectangle", rectCors)
return screen
screen = init_screen(
WIDTH, HEIGHT
) # Init the Screen, must be before initing panels (pongs) becuase pongs may not create a rectangular shape
screen.tracer(0)
pong1 = pong.Pong(0)
pong2 = pong.Pong(1)
p1_score = score.ScoreBoard(5, (-100, 270))
p2_score = score.ScoreBoard(5, (100, 270))
direction = random.choice(directions)
ball = ball.Ball(direction)
pong1_up = ball.ycor() + 35
pong_1_down = ball.ycor() - 35
pong2_up = ball.ycor() + 35
pong2_down = ball.ycor() - 35
force = ball.return_force(MAGNITUDE, direction)
des_point = force.desired_point
"--server_port",
dest="port",
type=int,
default=1080,
help="Port on which host server for camera is running")
args = parser.parse_args()
grabber = utility.WebcamVideoStream((args.host, args.port))
if args.debug: print("Camera Resolution:", grabber.cam_size)
grabber.start()
# Create our signal handler and connect it
handler = SignalHandler(grabber)
signal.signal(signal.SIGINT, handler)
game = pong.Pong()
# Kernal's for image processing
kernel_morp = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7, 7))
kernel_erod = np.ones((3, 3), np.uint8)
# Parameters for image processing
threshold = 60 # BINARY threshold
blurValue = 51 # GaussianBlur parameter
bgSubThreshold = 100
historyL, historyR = 0, 0
learningRate = 0
# Camera crop parameters
cropLY, cropLX = (0, grabber.cam_size[0]), (0, 100)
cropRY, cropRX = (0, grabber.cam_size[0]), (540, grabber.cam_size[1])
import pong
import sys
commandLineError = "Please enter a command line argument as follows:\n" \
+ "'user' if you want to play\n" \
+ "'nn' if you want a neural network to play\n" \
+ "'generator' if you want a perfect non-ml based bot to play\n"
generatorError = "Note: If you choose generator, enter a second argument specifying:\n" \
+ "'True' if you want it to generate a dataset and 'False' otherwise\n"
if (len(sys.argv) < 2):
print(commandLineError + generatorError)
else:
if (sys.argv[1] == 'user'):
game = pong.Pong('user')
game.runLoop()
elif (sys.argv[1] == 'nn'):
game = pong.Pong('nn')
game.runLoop()
elif (sys.argv[1] == 'generator'):
if (len(sys.argv) < 3):
print(generatorError)
elif (sys.argv[2] == 'True' or sys.argv[2] == 'False'):
game = pong.Pong('generator', sys.argv[2])
game.runLoop()
else:
print(generatorError)
# Console output thread
thread_console_output = console.ConsoleOutput(shared.q_console_output)
thread_console_output.start()
# Command interpreter thread
thread_command_interpreter = command_interpreter.CommandInterpreter(
shared.q_command_interpreter)
thread_command_interpreter.start()
# Console input thread
thread_console_input = console.ConsoleInput()
thread_console_input.daemon = True
thread_console_input.start()
# Pong thread
thread_pong = pong.Pong(shared.q_pong)
thread_pong.start()
# AI thread
thread_ai = ai.AI(shared.q_ai)
thread_ai.start()
# Player thread
thread_player = player.Player(shared.q_player)
thread_player.start()
# PROGRAM RUNS HERE TILL EXIT
# wait for all threads to be stopped with their various exit signals
# (except for threads which are daemonic and will be killed by sys.exit)
thread_console_output.join()
__author__ = 'Matt Geiger'
import pong
if __name__ == '__main__':
app = pong.Pong()
app.run()
def main():
if not training_mode:
agent = Agent(optimizer='rmsprop',
mode=learning_mode,
load_trained_model=True)
exploration_rate = 0
else:
agent = Agent(optimizer='rmsprop',
mode=learning_mode,
load_trained_model=False)
exploration_rate = initial_exploration_rate
print(agent.action_network.summary())
episode = 1
counter = 0
while episode <= max_episode:
# learning_mode indicate whether it learns from low_dims or high_dims
game = pong.Pong(mode=learning_mode)
done = False
training_history = []
if learning_mode == 'low_dims':
state_1 = np.expand_dims(game.GetPresentFrame(normalize=True),
axis=0)
elif learning_mode == 'high_dims':
frame = game.GetPresentFrame()
frame = cv2.cvtColor(
cv2.resize(frame, (input_dims[0], input_dims[1])),
cv2.COLOR_BGR2GRAY)
ret, frame = cv2.threshold(frame, 1, 255, cv2.THRESH_BINARY)
# stack frames, that is our input tensor
state_1 = np.stack((frame, frame, frame, frame), axis=2)
state_1 = state_1.reshape((1, input_dims[0], input_dims[1], 4))
while not done:
for event in pygame.event.get():
if event.type == QUIT:
sys.exit()
Q = agent.action_network.predict(state_1)
print('Q value is {}'.format(np.max(Q)))
if np.random.random_sample() > exploration_rate:
action = np.argmax(Q)
# print(Q)
else:
action = np.random.randint(3)
# get next state and reward
if learning_mode == 'low_dims':
reward, state_2 = game.GetNextFrame(action, normalize=True)
state_2 = np.expand_dims(state_2, axis=0)
if game.total_score < -20:
done = True
elif learning_mode == 'high_dims':
reward, frame = game.GetNextFrame(action, normalize=True)
frame = cv2.cvtColor(
cv2.resize(frame, (input_dims[0], input_dims[1])),
cv2.COLOR_BGR2GRAY)
ret, frame = cv2.threshold(frame, 1, 255, cv2.THRESH_BINARY)
frame = np.reshape(frame, (1, input_dims[0], input_dims[1], 1))
state_2 = np.append(frame, state_1[:, :, :, 0:3], axis=3)
if game.total_score < -20:
done = True
if training_mode:
agent.remember(state_1, action, reward, state_2, done)
if counter > observation_steps:
if counter == observation_steps + 1:
print('learning start!')
history = agent.experience_replay()
if counter % 100 == 0:
training_history.append([episode, history, np.max(Q)])
pd.DataFrame(
training_history,
columns=['episode', 'loss', 'Q']).to_csv(
'playing_pong-master/training_history.csv')
print('Episode {} : {}th frame: loss {}, Q {}'.format(
episode, counter - observation_steps,
history * 1000, np.max(Q)))
if counter % K == 0:
agent.train_target(tau)
if counter % 1000 == 0:
agent.save_model()
print('Made a copy of model')
# decay exploration rate
if exploration_rate >= final_exploration_rate:
exploration_rate -= exploration_step
# update state and counter
counter += 1
state_1 = state_2
print('Episode {} : total score is {}'.format(episode,
game.total_score))
episode += 1
def main():
spyral.director.push(pong.Pong())
