def eval_genome(g, config):
net = neat.nn.FeedForwardNetwork.create(g, config)
cycles_per_net = 5
fitnesses = [0] * cycles_per_net
for cycle in range(cycles_per_net):
x = random.randrange(SNAKE_WIDTH + SNAKE_WIDTH / 2,
WIN_WIDTH - (SNAKE_WIDTH + SNAKE_WIDTH / 2),
SNAKE_WIDTH)
y = random.randrange(SNAKE_WIDTH, WIN_WIDTH - (SNAKE_WIDTH),
SNAKE_WIDTH)
h = Head(WIN_WIDTH, SNAKE_WIDTH, x, y)
b1 = Body(h.x, h.y + SNAKE_WIDTH, h, SNAKE_WIDTH)
b2 = Body(b1.x, b1.y + SNAKE_WIDTH, b1, SNAKE_WIDTH)
bodies = [b1, b2]
a = Apple(WIN_WIDTH, SNAKE_WIDTH, h, bodies)
run = True
score = 0
time = 200
while (run):
time -= 1
fitnesses[cycle] += 0.01
if (time <= 0):
run = False
break
# Control Code
inputs = game.determine_position(h, bodies, a)
output = net.activate(
(inputs
)) #distance_up, distance_down, distance_left, distance_right
i = output.index(max(output))
if i == 0:
h.moveForward()
elif i == 1:
h.moveLeft()
elif i == 2:
h.moveRight()
game.update_positions(h, bodies)
if (h.check_collisions(bodies)):
run = False
break
if (h.check_apple(a)):
score += 1
fitnesses[cycle] += 10 + time * 0.1
time = 200
end = bodies[-1]
bodies.append(Body(end.p_x, end.p_y, end, SNAKE_WIDTH))
a.reset(h, bodies)
return [statistics.median(fitnesses), net]
def single_game(manual, distances=False, net=None, generation=None):
win = pygame.display.set_mode((WIN_WIDTH, WIN_WIDTH))
win.fill((0, 0, 0))
clock = pygame.time.Clock()
# generating head of snake
x = random.randrange(SNAKE_WIDTH + SNAKE_WIDTH / 2,
WIN_WIDTH - (SNAKE_WIDTH + SNAKE_WIDTH / 2),
SNAKE_WIDTH)
y = random.randrange(SNAKE_WIDTH, WIN_WIDTH - (SNAKE_WIDTH), SNAKE_WIDTH)
h = Head(WIN_WIDTH, SNAKE_WIDTH, x, y)
# generating body of snake
b1 = Body(h.x, h.y + SNAKE_WIDTH, h, SNAKE_WIDTH)
b2 = Body(b1.x, b1.y + SNAKE_WIDTH, b1, SNAKE_WIDTH)
bodies = [b1, b2]
# generating apple for snake to eat
a = Apple(WIN_WIDTH, SNAKE_WIDTH, h, bodies)
run = True
score = 0
time = 200
while (run):
# setting clock speed. Slower if the game is being manually run
if manual:
clock.tick(MANUAL_SNAKE_SPEED)
else:
clock.tick(SNAKE_SPEED)
# Counting down time and cause the game to quit if an apple hasnt been eaten within the time frame
time -= 1
if (time <= 0):
run = False
print('Time Out!')
break
# Automated Control
# inputs to neural network:
# distance forward, distance left, distance right, apple right, apple top
inputs = determine_position(h, bodies, a)
# Previous attempts
# head x, head y, boundaries, apple x, apple y # Failed to train after 100 generatations of 100 population -> has no sense of direction
# distance front, distance down, distance left, distance right, left distance to apple, upwards distance to apple # Failed to train after 100 generatations of 100 population -> has no sense of direction
# if the user is not playing the game manually, run the game with automated controls
if not (manual):
# run inputs into neural network
output = net.activate((inputs))
# determing direction indicated by NN
i = output.index(max(output))
if i == 0:
h.moveForward()
elif i == 1:
h.moveLeft()
elif i == 2:
h.moveRight()
for event in pygame.event.get():
# checking if exit button is pushed on game
if (event.type == pygame.QUIT):
run = False
break
# manual control - checking for key pushes
if manual:
if event.type == pygame.KEYDOWN:
if event.key == pygame.K_UP:
h.KeyMoveUp()
break # breaking prevents multiple keys being processed at the same time
elif event.key == pygame.K_DOWN:
h.KeyMoveDown()
break
elif event.key == pygame.K_LEFT:
h.KeyMoveLeft()
break
elif event.key == pygame.K_RIGHT:
h.KeyMoveRight()
break
# update the position of the head and bodies
update_positions(h, bodies)
# checking for collisions
if (h.check_collisions(bodies)):
run = False
print('Collision!')
break
# check if head ate apple
if (h.check_apple(a)):
score += 1
time = 200
end = bodies[-1]
bodies.append(Body(end.p_x, end.p_y, end, SNAKE_WIDTH))
a.reset(h, bodies)
# if (score >= 200):
# run = False
# break
# outputing distances for debugging
distance_list = []
if (distances):
distance_list = [inputs[0], inputs[1], inputs[2]]
# drawing game state onto screen
draw_screen(win, h, bodies, a, distance_list, score, time, generation)
