def one_game():
pygame.event.pump()
game_over = False
lead_x = 150
lead_y = 150
snake = Snake(showScreen, screenx, screeny, snakeImg, lead_x, lead_y)
apple = Apple(showScreen, screenx, screeny, b_size, appleImg, snake.snake_list)
while not game_over:
x, y = apple.get_apple_pos()
snake.update_snake_list(x, y)
if snake.is_alive() is False: game_over = True
showScreen.fill(white)
if snake.eaten is True: apple.update_apple_pos(snake.snake_list)
apple.display()
snake.eaten = False
snake.display()
snake.display_score()
pygame.display.update()
a_x, a_y = apple.get_apple_pos()
s_x, s_y = snake.get_snake_head()
visited = snake.snake_list.copy()
visited.remove([s_x, s_y])
result = BreathFirstSearch(screenx, screeny, b_size, visited, [a_x, a_y], [s_x, s_y])
next_cell = result[1]
x_diff = next_cell[0] - s_x
y_diff = next_cell[1] - s_y
if x_diff > 0: snake.direction = "right"
elif x_diff < 0: snake.direction = "left"
elif y_diff > 0: snake.direction = "down"
elif y_diff < 0: snake.direction = "up"
time.tick(FPS)
def one_game():
pygame.event.pump()
game_over = False
# snake will start in the middle of the game window
lead_x = 70
lead_y = 70
# snake default direction is right
snake = Snake(gameDisplay, display_width, display_height, img, lead_x,
lead_y)
apple = Apple(gameDisplay, display_width, display_height, block_size, img2,
snake.snake_list)
while not game_over:
# based on the direction, we can work out the x, y changes to update the snake
x, y = apple.get_apple_pos()
snake.update_snake_list(x, y)
# check if snake dies
if snake.is_alive() is False:
game_over = True
gameDisplay.fill(white)
# if snake eats the apple, make a random new apple
if snake.eaten is True:
apple.update_apple_pos(snake.snake_list)
apple.display()
snake.eaten = False
snake.display()
snake.display_score()
pygame.display.update()
# this part is using the snake position and apple
# position to use the A* method to get the path
a_x, a_y = apple.get_apple_pos()
s_x, s_y = snake.get_snake_head()
visited = snake.snake_list.copy()
visited.remove([s_x, s_y])
result = A_star(display_width, display_height, block_size, visited,
(a_x, a_y), (s_x, s_y))
# since the path starts from snake position, the second element will
# be next move
next_cell = result[1]
# update the snake position based on the next move position
x_diff = next_cell[0] - s_x
y_diff = next_cell[1] - s_y
if x_diff > 0:
snake.direction = "right"
elif x_diff < 0:
snake.direction = "left"
elif y_diff > 0:
snake.direction = "down"
elif y_diff < 0:
snake.direction = "up"
clock.tick(FPS)
def training_game(times=10):
# s list will store the score of each game
s = []
for i in range(times):
# print out the game number
print(i)
pygame.event.pump()
game_over = False
# Will be the leader of the #1 block of the snake
lead_x = 70
lead_y = 70
# snake default direction is right
snake = RL_Snake(gameDisplay, display_width, display_height, img,
lead_x, lead_y)
apple = Apple(gameDisplay, display_width, display_height, block_size,
img2, snake.snake_list)
a_x, a_y = apple.get_apple_pos()
# get the initial state, and action will be "up" starting
old_state = snake.get_state([a_x, a_y])
old_action = "up"
while not game_over:
# based on the direction, we can work out the x, y changes to update the snake
a_x, a_y = apple.get_apple_pos()
snake.update_snake_list(a_x, a_y)
# snake not die or eats the apple, reward will be -10
# this is negative so that it will "encourage" the snake to
# move towards to the apple, since that is the only positive award
reward = -10
# check if snake dies
if snake.is_alive() is False:
game_over = True
# if snake dies, award is -100
reward = -100
s.append(snake.snake_length - 1)
gameDisplay.fill(white)
# if snake eats the apple, make a random new apple
if snake.eaten is True:
apple.update_apple_pos(snake.snake_list)
# if snake eats the apple, reward is 500
reward = 500
#############################################
# get he new state and new action, then we can update the Q table
state = snake.get_state([a_x, a_y])
action = snake_agent.getA(tuple(state))
snake_agent.updateQ(tuple(old_state), old_action, tuple(state),
action, reward)
old_action = action
# training will take a lot of time, so archive the Q table
snake_agent.saveQ()
# this part is using the snake position and apple
# position to use the Sarsa method to get the action
a_x, a_y = apple.get_apple_pos()
old_state = snake.get_state([a_x, a_y])
snake.set_direction_by_action(action)
#############################################
apple.display()
snake.eaten = False
snake.display()
snake.display_score()
pygame.display.update()
clock.tick(FPS)
# after traning is done, print out the average score
print("Average score is: {}".format(sum(s) / len(s)))
def training_game(times=100):
# s list will store the score of each game
s = []
for i in range(times):
# print out the game number
print(i)
pygame.event.pump()
game_over = False
# Will be the leader of the #1 block of the snake
lead_x = 70
lead_y = 70
# snake default direction is right
snake = DQN_Snake(gameDisplay, display_width, display_height, img,
lead_x, lead_y)
apple = Apple(gameDisplay, display_width, display_height, block_size,
img2, snake.snake_list)
a_x, a_y = apple.get_apple_pos()
# get the initial state, and action will be "up" starting
action = "up"
old_state = snake.get_state([a_x, a_y])
while not game_over:
# based on the direction, we can work out the x, y changes to update the snake
a_x, a_y = apple.get_apple_pos()
snake.update_snake_list(a_x, a_y)
# get the new state
state = snake.get_state([a_x, a_y])
# snake not die or eats the apple, reward will be -10
# this is negative so that it will "encourage" the snake to
# move towards to the apple, since that is the only positive award
reward = -10
# check if snake dies
if snake.is_alive() is False:
game_over = True
# copy the weights from q_model to target_model
agent.copy_weights()
# if snake dies, award is -100
reward = -100
s.append(snake.snake_length - 1)
gameDisplay.fill(white)
# if snake eats the apple, make a random new apple
if snake.eaten is True:
apple.update_apple_pos(snake.snake_list)
# if snake eats the apple, reward is 100
reward = 100
#############################################
# store the train_data to the memory
agent.store_train_data(np.reshape(old_state, [1, 5]),
look_up[action], reward,
np.reshape(state, [1, 5]), game_over)
# if the memory size is larger than the batch_size, start training
if len(agent.memory) > batch_size:
agent.train(batch_size)
# push state to old state
a_x, a_y = apple.get_apple_pos()
old_state = snake.get_state([a_x, a_y])
# get the action from the DQN model
action = actions[agent.get_action(np.reshape(old_state, [1, 5]))]
snake.set_direction_by_action(action)
#############################################
apple.display()
snake.eaten = False
snake.display()
snake.display_score()
pygame.display.update()
clock.tick(FPS)
# when the training is finised, sae the model
agent.save_model()
# after traning is done, print out the average score
print("Average score is: {}".format(sum(s) / len(s)))
def training_game(times=40000):
# we need to store the training data first
# then use the data to train the NN
train_data = []
# f = open("train_data.txt", "rb")
# train_data = pickle.load(f)
# f.close()
for i in range(times):
# print out the game number
print(i)
pygame.event.pump()
game_over = False
# Will be the leader of the #1 block of the snake
lead_x = 70
lead_y = 70
# snake default direction is right
snake = NN_Snake(gameDisplay, display_width, display_height, img,
lead_x, lead_y)
apple = Apple(gameDisplay, display_width, display_height, block_size,
img2, snake.snake_list)
a_x, a_y = apple.get_apple_pos()
# get the initial state, and action will be "up" starting
action = "up"
state = snake.get_state([a_x, a_y], action)
old_distance = snake.get_distance([a_x, a_y])
while not game_over:
# based on the direction, we can work out the x, y changes to update the snake
a_x, a_y = apple.get_apple_pos()
snake.update_snake_list(a_x, a_y)
# after snake moves, get the new distance
distance = snake.get_distance([a_x, a_y])
# default reward is 0
reward = 0
# check if snake dies
if snake.is_alive() is False:
game_over = True
# if snake dies, award is -1
reward = -1
gameDisplay.fill(white)
# if snake eats the apple, make a random new apple
if snake.eaten is True:
apple.update_apple_pos(snake.snake_list)
# if snake eats the apple, or moved closer to apple, reward is 1
if snake.eaten is True or distance < old_distance:
reward = 1
#############################################
# collect the training data for NN
train_data.append([np.array(state), reward])
# this part is using random method to move the snake
action = random.choice(actions)
a_x, a_y = apple.get_apple_pos()
old_distance = snake.get_distance([a_x, a_y])
state = snake.get_state([a_x, a_y], action)
snake.set_direction_by_action(action)
#############################################
apple.display()
snake.eaten = False
snake.display()
snake.display_score()
pygame.display.update()
clock.tick(FPS)
# store the training data to txt
print(len(train_data))
f = open("train_data.txt", "wb")
pickle.dump(train_data, f)
f.close()
def testing_game(times=10):
# load the trained NN model
model = load_model('my_model.h5')
# s list will store the score of each game
s = []
for i in range(times):
# print out the game number
print(i)
pygame.event.pump()
game_over = False
# Will be the leader of the #1 block of the snake
lead_x = 70
lead_y = 70
# snake default direction is right
snake = NN_Snake(gameDisplay, display_width, display_height, img,
lead_x, lead_y)
apple = Apple(gameDisplay, display_width, display_height, block_size,
img2, snake.snake_list)
a_x, a_y = apple.get_apple_pos()
# get the initial state, and action will be "up" starting
action = "up"
state = snake.get_state([a_x, a_y], action)
old_distance = snake.get_distance([a_x, a_y])
while not game_over:
# based on the direction, we can work out the x, y changes to update the snake
a_x, a_y = apple.get_apple_pos()
snake.update_snake_list(a_x, a_y)
# check if snake dies
if snake.is_alive() is False:
game_over = True
s.append(snake.snake_length)
gameDisplay.fill(white)
# if snake eats the apple, make a random new apple
if snake.eaten is True:
apple.update_apple_pos(snake.snake_list)
#############################################
# get the position of the apple
a_x, a_y = apple.get_apple_pos()
# use NN model to get the action with max Q
precictions = {}
for action in actions:
state = snake.get_state([a_x, a_y], action)
precictions[action] = model.predict(
np.array(state).reshape(-1, 5))[0][0]
action = max(precictions, key=precictions.get)
# set the direction of snake using the chosen action
snake.set_direction_by_action(action)
#############################################
apple.display()
snake.eaten = False
snake.display()
snake.display_score()
pygame.display.update()
clock.tick(FPS)
# after traning is done, print out the average score
print("Average score is: {}".format(sum(s) / len(s)))
