def main():
pygame.mixer.pre_init(44100, -16, 2, 4096)
pygame.init()
screen = pygame.display.set_mode((640, 480))
max_frame_rate = 60
dashboard = Dashboard("./img/font.png", 8, screen)
sound = Sound()
level = Level(screen, sound, dashboard)
menu = Menu(screen, dashboard, level, sound)
while not menu.start:
menu.update()
mario = Mario(0, 0, level, screen, dashboard, sound)
clock = pygame.time.Clock()
while not mario.restart:
pygame.display.set_caption(
str(
round(float("{:.2f} FPS".format(clock.get_fps()).split(" ")
[0]))) + " FPS")
level.drawLevel(mario.camera)
dashboard.update()
mario.update()
pygame.display.update()
clock.tick(max_frame_rate)
main()
def main():
pygame.mixer.pre_init(44100, -16, 2, 4096)
pygame.init()
screen = pygame.display.set_mode(windowSize)
max_frame_rate = 60
dashboard = Dashboard("./img/font.png", 8, screen)
sound = Sound()
level = Level(screen, sound, dashboard)
menu = Menu(screen, dashboard, level, sound)
while not menu.start:
menu.update()
mario = Mario(0, 0, level, screen, dashboard, sound)
clock = pygame.time.Clock()
while not mario.restart:
pygame.display.set_caption("Super Mario running with {:d} FPS".format(int(clock.get_fps())))
if mario.pause:
mario.pauseObj.update()
else:
level.drawLevel(mario.camera)
dashboard.update()
mario.update()
pygame.display.update()
clock.tick(max_frame_rate)
return 'restart'
def main():
pygame.mixer.pre_init(44100, -16, 2, 2048)
pygame.init()
screen = pygame.display.set_mode((640, 480))
max_frame_rate = 60
level = Level(screen)
dashboard = Dashboard("./img/font.png", 8, screen)
mario = Mario(0, 0, level, screen, dashboard)
input = Input(mario)
clock = pygame.time.Clock()
while (not mario.restart):
pygame.display.set_caption("{:.2f} FPS".format(clock.get_fps()))
level.drawLevel(mario.camera)
dashboard.update()
input.checkForInput()
mario.update()
pygame.display.update()
clock.tick(max_frame_rate)
mario.sound.shutdown()
main()
def init_game(self, y_position=0, coins=0, points=0, time=0, clock=0):
if not self.headless:
pygame.mixer.pre_init(44100, -16, 2, 4096)
pygame.init()
self.screen = pygame.display.set_mode((640, 480))
self.dashboard = Dashboard("./img/font.png",
8,
self.screen,
coins=0,
points=0,
time=0)
self.sound = Sound()
self.level = Level(self.screen, self.sound, self.dashboard,
self.levelname)
self.menu = Menu(self.screen, self.dashboard, self.level,
self.sound)
self.menu.update()
self.mario = Mario(0, y_position / 32, self.level, self.screen,
self.dashboard, self.sound)
if self.experiment == 'FD':
self.mario2 = Mario(0, 2, self.level, self.screen,
self.dashboard, self.sound)
self.clock = pygame.time.Clock()
self.menu.dashboard.state = "play"
self.menu.dashboard.time = 0
self.menu.start = True
pygame.display.update()
else:
self.level = LevelHeadless(self.levelname)
if self.experiment == 'FD':
self.mario2 = MarioHeadless(0, 2, self.level)
self.mario = MarioHeadless(0, 0, self.level)
self.clock = clock
def main(self):
# Initializing the game, graphics and sound
pygame.mixer.pre_init(44100, -16, 2, 4096)
pygame.init()
screen = pygame.display.set_mode(self.windowSize)
max_frame_rate = 9999
dashboard = Dashboard("./img/font.png", 8, screen)
sound = Sound()
maxReward = 0
timesMaxedTheReward = 0
winStreak = 0
maxWinStreak = 0
differentialFramesCount = 10
dxStartFrame = 0
diffXOverFrames = 0
dxConstant = 0
lastX = 0
epsCoef = 0
for episodeCounter in range(0, self.experimentsCount):
print("=============================================")
print("============WELCOME TO EPISODE {0}============".format(
episodeCounter))
print("=============================================")
framesCounter = 0
stillLearning = episodeCounter <= self.learningPeriod
if stillLearning and episodeCounter / self.experimentsCount >= self.globalStep / self.globalStepMax:
self.decreaseLearningRate()
self.printLearningRate()
self.level = Level(screen, sound, dashboard)
self.level.loadLevel('Level1-1')
dashboard.coins = 0
dashboard.points = 0
self.mario = Mario(0, 0, self.level, screen, dashboard, sound)
self.mario.setPos(0, 384)
clock = pygame.time.Clock()
max_x = 0
obs = np.append(
np.append(self.mario.getObservation(), framesCounter),
self.level.getClosestEntityDistance(self.mario))
finishedLevel = False
while framesCounter < self.maxFrames and not finishedLevel:
#print(framesCounter)
pygame.display.set_caption(
"Super Mario running with {:d} FPS".format(
int(clock.get_fps())))
if self.mario.pause:
self.mario.pauseObj.update()
else:
self.level.drawLevel(self.mario.camera)
dashboard.update()
self.mario.update()
#print(self.mario.rect.x, self.mario.rect.y)
prevObs = obs
obs = np.append(
np.append(self.mario.getObservation(), framesCounter),
self.level.getClosestEntityDistance(self.mario))
reward = 0
actionIdentifier = self.player.agent.sample_action(
obs, self.eps)
if max_x < obs[0]:
max_x = obs[0]
reward += 1
# diffXOverFrames = self.mario.rect.x - dxConstant
if self.mario.restart:
print("I DIED!")
finishedLevel = True
reward -= 400
winStreak = 0
else:
if max_x >= self.level.max_X and not finishedLevel:
finishedLevel = True
timesMaxedTheReward += 1
winStreak += 1
reward += 200
else:
if self.mario.rect.x > max_x:
reward += 5
if framesCounter - dxStartFrame >= differentialFramesCount:
dxStartFrame = framesCounter
dxConstant = self.mario.rect.x
if stillLearning:
next = self.player.agent.predict(obs)
assert (len(next.shape) == 1)
G = reward + self.gamma * np.max(next)
self.player.agent.update(prevObs, actionIdentifier, G)
self.doAction(actionIdentifier)
pygame.display.update()
clock.tick(max_frame_rate)
framesCounter += 1
if timesMaxedTheReward > 0:
print("I've maxed the reward {0} times!".format(
timesMaxedTheReward))
if maxWinStreak < winStreak:
maxWinStreak = winStreak
print("winStreak {0}, maxWinStreak: {1}".format(
winStreak, maxWinStreak))
if stillLearning:
print("Still learning!")
self.running_avg[episodeCounter] = self.totalrewards[
0:episodeCounter + 1].mean()
self.eps -= 1 / (
(epsCoef + 2) * (epsCoef + 1)
) # eps=1/(n+1) for n=epsCoef ~~ episodeCounter
epsCoef += 1
if self.running_avg[episodeCounter] < self.running_avg[int(
np.ceil(np.sqrt(episodeCounter)))]:
self.eps = 1 / (np.ceil(np.sqrt(episodeCounter)) + 1)
epsCoef = np.ceil(np.sqrt(episodeCounter))
print("Reverting epsilon**: {0}".format(self.eps))
elif episodeCounter >= self.episodesPerStep and self.running_avg[
episodeCounter] < self.running_avg[
episodeCounter - self.episodesPerStep]:
self.eps = 1 / (
(episodeCounter - self.episodesPerStep) + 1)
epsCoef = episodeCounter
print("Reverting epsilon: {0}".format(self.eps))
else:
self.running_avg[episodeCounter] = self.totalrewards[
learningPeriod:episodeCounter + 1].mean()
totalreward = max_x
if totalreward > maxReward:
maxReward = totalreward
self.totalrewards[episodeCounter] = totalreward
print("Episode: {0}, maxReward: {1} eps: {2} totalReward: {3}".
format(episodeCounter, maxReward, self.eps, totalreward))
print("Avg: {0}".format(self.running_avg[episodeCounter] /
self.level.max_X))
for reward in self.totalrewards:
print(reward)
class MainWindow:
def __init__(self):
# Initializing ML parameters
self.techniqueIdentifier = 0
self.startingLearningRate = 0.1
self.param1 = 0.9
self.param2 = 0.99
self.learningPeriod = 300
self.gamma = 0.97
self.eps = 1.0
self.experimentsCount = 20000
self.maxFrames = 3000
self.windowSize = (640, 480)
self.mario = None
self.episodesPerStep = 10
self.globalStepMax = int(self.experimentsCount / self.episodesPerStep)
self.globalStep = 1
self.decayRate = 0.99
self.learningRate = tf.train.exponential_decay(
self.startingLearningRate,
self.globalStep,
1,
self.decayRate,
staircase=True)
self.player = MLPlayer(self.techniqueIdentifier, self.maxFrames, 1888,
self.learningRate, self.globalStep, self.param1,
self.param2)
self.totalrewards = np.empty(self.experimentsCount)
self.running_avg = np.empty(self.experimentsCount)
#(lambda: os.system('cls'))() # Clear console
def decreaseLearningRate(self):
self.globalStep += 1
def calculateLearningRate(self):
return self.startingLearningRate * (self.decayRate
**(self.globalStep / 1))
def printLearningRate(self):
print("Learning rate is:{0}".format(self.calculateLearningRate()))
def main(self):
# Initializing the game, graphics and sound
pygame.mixer.pre_init(44100, -16, 2, 4096)
pygame.init()
screen = pygame.display.set_mode(self.windowSize)
max_frame_rate = 9999
dashboard = Dashboard("./img/font.png", 8, screen)
sound = Sound()
maxReward = 0
timesMaxedTheReward = 0
winStreak = 0
maxWinStreak = 0
differentialFramesCount = 10
dxStartFrame = 0
diffXOverFrames = 0
dxConstant = 0
lastX = 0
epsCoef = 0
for episodeCounter in range(0, self.experimentsCount):
print("=============================================")
print("============WELCOME TO EPISODE {0}============".format(
episodeCounter))
print("=============================================")
framesCounter = 0
stillLearning = episodeCounter <= self.learningPeriod
if stillLearning and episodeCounter / self.experimentsCount >= self.globalStep / self.globalStepMax:
self.decreaseLearningRate()
self.printLearningRate()
self.level = Level(screen, sound, dashboard)
self.level.loadLevel('Level1-1')
dashboard.coins = 0
dashboard.points = 0
self.mario = Mario(0, 0, self.level, screen, dashboard, sound)
self.mario.setPos(0, 384)
clock = pygame.time.Clock()
max_x = 0
obs = np.append(
np.append(self.mario.getObservation(), framesCounter),
self.level.getClosestEntityDistance(self.mario))
finishedLevel = False
while framesCounter < self.maxFrames and not finishedLevel:
#print(framesCounter)
pygame.display.set_caption(
"Super Mario running with {:d} FPS".format(
int(clock.get_fps())))
if self.mario.pause:
self.mario.pauseObj.update()
else:
self.level.drawLevel(self.mario.camera)
dashboard.update()
self.mario.update()
#print(self.mario.rect.x, self.mario.rect.y)
prevObs = obs
obs = np.append(
np.append(self.mario.getObservation(), framesCounter),
self.level.getClosestEntityDistance(self.mario))
reward = 0
actionIdentifier = self.player.agent.sample_action(
obs, self.eps)
if max_x < obs[0]:
max_x = obs[0]
reward += 1
# diffXOverFrames = self.mario.rect.x - dxConstant
if self.mario.restart:
print("I DIED!")
finishedLevel = True
reward -= 400
winStreak = 0
else:
if max_x >= self.level.max_X and not finishedLevel:
finishedLevel = True
timesMaxedTheReward += 1
winStreak += 1
reward += 200
else:
if self.mario.rect.x > max_x:
reward += 5
if framesCounter - dxStartFrame >= differentialFramesCount:
dxStartFrame = framesCounter
dxConstant = self.mario.rect.x
if stillLearning:
next = self.player.agent.predict(obs)
assert (len(next.shape) == 1)
G = reward + self.gamma * np.max(next)
self.player.agent.update(prevObs, actionIdentifier, G)
self.doAction(actionIdentifier)
pygame.display.update()
clock.tick(max_frame_rate)
framesCounter += 1
if timesMaxedTheReward > 0:
print("I've maxed the reward {0} times!".format(
timesMaxedTheReward))
if maxWinStreak < winStreak:
maxWinStreak = winStreak
print("winStreak {0}, maxWinStreak: {1}".format(
winStreak, maxWinStreak))
if stillLearning:
print("Still learning!")
self.running_avg[episodeCounter] = self.totalrewards[
0:episodeCounter + 1].mean()
self.eps -= 1 / (
(epsCoef + 2) * (epsCoef + 1)
) # eps=1/(n+1) for n=epsCoef ~~ episodeCounter
epsCoef += 1
if self.running_avg[episodeCounter] < self.running_avg[int(
np.ceil(np.sqrt(episodeCounter)))]:
self.eps = 1 / (np.ceil(np.sqrt(episodeCounter)) + 1)
epsCoef = np.ceil(np.sqrt(episodeCounter))
print("Reverting epsilon**: {0}".format(self.eps))
elif episodeCounter >= self.episodesPerStep and self.running_avg[
episodeCounter] < self.running_avg[
episodeCounter - self.episodesPerStep]:
self.eps = 1 / (
(episodeCounter - self.episodesPerStep) + 1)
epsCoef = episodeCounter
print("Reverting epsilon: {0}".format(self.eps))
else:
self.running_avg[episodeCounter] = self.totalrewards[
learningPeriod:episodeCounter + 1].mean()
totalreward = max_x
if totalreward > maxReward:
maxReward = totalreward
self.totalrewards[episodeCounter] = totalreward
print("Episode: {0}, maxReward: {1} eps: {2} totalReward: {3}".
format(episodeCounter, maxReward, self.eps, totalreward))
print("Avg: {0}".format(self.running_avg[episodeCounter] /
self.level.max_X))
for reward in self.totalrewards:
print(reward)
# Controls
def doAction(self, actionIdentifier):
if (actionIdentifier == 0):
# Left
self.mario.traits['jumpTrait'].jump(True)
self.mario.traits["goTrait"].direction = -1
elif (actionIdentifier == 1):
# Right
self.mario.traits['jumpTrait'].jump(True)
self.mario.traits["goTrait"].direction = 1
if (actionIdentifier == 2):
# Left
self.mario.traits['jumpTrait'].jump(False)
self.mario.traits["goTrait"].direction = -1
elif (actionIdentifier == 3):
# Right
self.mario.traits['jumpTrait'].jump(False)
self.mario.traits["goTrait"].direction = 1
screen = pygame.display.set_mode(window_size)
bg = pygame.transform.scale(screen, (w, h))
sound_player = SoundPlayer()
while True:
menu = Menu(screen, sound_player)
while not menu.pause:
menu.update()
pygame.display.update()
fps_clock.tick(FPS / 6)
if sound_player.allow_sound:
if not sound_player.bg_sound.is_playing():
sound_player.bg_sound.play_sound(True)
else:
sound_player.bg_sound.stop_sound()
level = Level("levels/" + menu.level_name, screen)
background = Background(0, 0, screen, level)
mario = Mario(0, 0, Mario.DIRECTION_RIGHT, 0, Mario.IN_AIR, screen,
background, level, sound_player)
background.set_character(mario)
while not mario.restart:
if mario.pause:
mario.pause_obj.update()
else:
background.update()
mario.update()
pygame.display.update()
fps_clock.tick(FPS)