return np.tanh(x);
def act_prime(x):
return 1-np.tanh(x)**2;
# loss function and its derivative
def loss(y_true, y_pred):
return np.mean(np.power(y_true-y_pred, 2));
def loss_prime(y_true, y_pred):
return 2*(y_pred-y_true)/y_true.size;
# training data
x_train = np.array([[[0,0]], [[0,1]], [[1,0]], [[1,1]]]);
y_train = np.array([[[0]], [[1]], [[1]], [[0]]]);
# network
net = Network();
net.add(FCLayer((1,2), (1,3)));
net.add(ActivationLayer((1,3), act, act_prime));
net.add(FCLayer((1,3), (1,1)));
net.add(ActivationLayer((1,1), act, act_prime));
# train
net.use(loss, loss_prime);
net.fit(x_train, y_train, epochs=1000, learning_rate=0.1);
# test
out = net.predict(x_train);
print(out);
class Individual:
def __init__(self):
self.NN = Network(5, 1)
self.alive = True
self.reached_pilars = 0
self.score = 0
self.rect = pygame.Rect((int(WIDTH / 2), int(HEIGHT / 2)), (40, 40))
self.movement_y = 0
self.bool = True
self.bool2 = True
def draw(self, display):
pygame.draw.rect(display, (0, 255, 0), self.rect, 1)
def update(self):
if self.alive:
self.score+=1
self.predict()
self.movement_y += gravity
self.rect.centery += self.movement_y
self.check()
VARIABLES.best_round_score = self.reached_pilars
def check(self):
global pipe_rect_list
if self.rect.collidelist(pipe_rect_list) != -1:
self.alive = False
if pipe_rect_list[0].centerx - 10 < self.rect.centerx < pipe_rect_list[0].centerx + 10 and self.bool:
self.bool = False
self.reached_pilars += 1
elif not pipe_rect_list[0].centerx - 10 < self.rect.centerx < pipe_rect_list[0].centerx + 10:
self.bool = True
if pipe_rect_list[2].centerx - 10 < self.rect.centerx < pipe_rect_list[1].centerx + 10 and self.bool2:
self.bool2 = False
self.reached_pilars += 1
elif not pipe_rect_list[2].centerx - 10 < self.rect.centerx < pipe_rect_list[1].centerx + 10:
self.bool2 = True
if self.rect.bottom>HEIGHT or self.rect.top < 0:
self.alive = False
def jump(self):
self.movement_y = -8
def get_state(self):
global pipe_list
if pipe_list[0].rect_upper.right < self.rect.left and pipe_list[0].rect_upper.right < pipe_list[1].rect_upper.right:
pipe = pipe_list[1]
elif pipe_list[1].rect_upper.right < self.rect.left and pipe_list[1].rect_upper.right < pipe_list[0].rect_upper.right:
pipe = pipe_list[0]
else:
pipe = pipe_list[0]
return [self.rect.centery/HEIGHT, pipe.rect_upper.bottom/HEIGHT,
pipe.rect_lower.top/HEIGHT, pipe.rect_lower.centerx/WIDTH,
self.movement_y/20]
def pair(self, other):
new_weight1 = self.NN.get_flatted()
new_weight2 = other.NN.get_flatted()
gene = random.randint(0, len(new_weight1) - 1)
new_weight1[gene:], new_weight2[:gene] = new_weight2[gene:], new_weight1[:gene]
return new_weight1, new_weight2
def predict(self):
output = self.NN.predict(self.get_state())
if output < 0.5:
self.jump()
def reset(self):
self.rect.centery = int(HEIGHT / 2)
self.alive = True
self.reached_pilars = 0
self.score = 0