python 中 "smart dots" 的遗传算法不起作用
Genetic algorithm for "smart dots" in python doesn't work
在过去的几天里，我一直在尝试实现所谓的“智能点”游戏。我第一次看到它是在 Code Bullet youtube 频道：https://www.youtube.com/watch?v=BOZfhUcNiqk。不幸的是，它是用 Processing 语言编码的，而我几乎不知道的唯一语言是 Python。我完成了游戏的 python 版本，但出现了一些错误。
问题在于，在第二代中，被选为最佳点的点几乎立即停止移动。我认为这与我不擅长 OOP 和复制 Brain class 错误有关。步骤（我用于移动）在主循环的第一次或第二次循环中从零（在开始时设置）跳到最大值（200）。但问题还不止于此。在下一代，当我尝试将 brain step 设置为零时，它会中断：
AttributeError: 'NoneType' object has no attribute 'brain'
我尝试手动设置新大脑，但我仍然遇到同样的错误。如果有人已经做了这个或者有时间可以帮助我解决这个错误甚至项目，我将不胜感激。
我知道代码中有很多未使用的东西，但这只是我尝试修复它的结果
:(
注释掉的代码是我用过的一些旧代码。
main2.py（主循环）：
</p>
<pre><code>import pygame
import klase2
pygame.init()
def main():
win = pygame.display.set_mode((klase2.WIN_W, klase2.WIN_H))
clock = pygame.time.Clock()
population = klase2.Population()
dots = population.return_dots(1000)
goal = klase2.Goal()
run = True
while run:
clock.tick(30)
for event in pygame.event.get():
if event.type == pygame.QUIT:
run = False
win.fill((255, 255, 255))
goal.draw_goal(win)
for dot in dots:
dot.draw_dot(win)
dot.update_dot()
if population.all_dots_dead():
# natural selection
population.natural_selection()
# mutation
dots = population.mutate_dots()
population.gen += 1
print(population.gen)
pygame.display.update()
main()
kase2（处理所有功能和 classes）：
</p>
<pre><code>import pygame
import numpy as np
from pygame import gfxdraw
import math
import random
pygame.init()
WIN_W = 500
WIN_H = 500
class Brain:
def __init__(self, size):
self.step = 0
self.size = size
self.directionsx = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.size / 2)))
self.directionsy = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.size / 2)))
def clone(self):
self.size = self.size
self.directionsx = self.directionsx
self.directionsy = self.directionsy
self.step = 0
class Goal:
def __init__(self):
self.x = WIN_W / 2
self.y = 10
self.color = (255, 20, 20)
self.r = 5
def draw_goal(self, win):
pygame.gfxdraw.aacircle(win, int(self.x), int(self.y), self.r, self.color)
pygame.gfxdraw.filled_circle(win, int(self.x), int(self.y), self.r, self.color)
class Dot:
goal = Goal()
def __init__(self):
self.tick = 0
self.goal = Goal()
self.brain = Brain(400)
self.velx = 0
self.vely = 0
self.accx = 0
self.accy = 0
self.x = WIN_W / 2
self.y = WIN_H - 10
self.r = 3
self.color = (0, 0, 0)
self.alive = True
self.velLimit = 5
self.fitness = 0
def draw_dot(self, win):
pygame.gfxdraw.aacircle(win, int(self.x), int(self.y), self.r, self.color)
pygame.gfxdraw.filled_circle(win, int(self.x), int(self.y), self.r, self.color)
def move_dot(self):
if self.brain.size / 2 > self.brain.step:
self.accx = self.brain.directionsx[self.brain.step]
self.accy = self.brain.directionsy[self.brain.step]
self.brain.step += 1
else:
self.alive = False
self.velx += self.accx
self.vely += self.accy
if self.velx > self.velLimit:
self.velx = self.velLimit
elif self.velx < -self.velLimit:
self.velx = -self.velLimit
if self.vely > self.velLimit:
self.vely = self.velLimit
elif self.vely < -self.velLimit:
self.vely = -self.velLimit
self.x += self.velx
self.y += self.vely
def update_dot(self):
if not self.reached_goal():
self.tick += 1
if self.alive:
self.move_dot()
if self.x < 0 + self.r or self.x > WIN_W - self.r or self.y < 0 + self.r or self.y > WIN_H - self.r or self.reached_goal():
self.alive = False
def distance_to_goal(self):
a = abs(self.x - self.goal.x)
b = abs(self.y - self.goal.y)
return math.sqrt(a**2 + b**2)
def reached_goal(self):
if self.distance_to_goal() <= self.r + self.goal.r:
return True
return False
def fitness_dot(self):
if self.reached_goal():
self.fitness = 1 / (self.brain.step)
else:
self.fitness = 1 / (self.distance_to_goal()**2)
return self.fitness
class Population:
def __init__(self):
self.dots = []
self.newDots = []
self.gen = 0
self.mutateChance = 800
self.size = 0
self.fitness_sum = 0
def return_dots(self, size):
self.size = size
for _ in range(size):
self.dots.append(Dot())
return self.dots
def all_dots_dead(self):
for i in range(len(self.dots)):
if self.dots[i].alive:
return False
return True
def sort_dots(self):
self.dots = sorted(self.dots, key=lambda dot: dot.fitness, reverse=True)
def sum_fitness(self):
for dot in self.dots:
self.fitness_sum += dot.fitness_dot()
return self.fitness_sum
def get_parent(self):
rand = random.uniform(0, self.fitness_sum)
running_sum = 0
for dot in self.dots:
running_sum += dot.fitness
if running_sum >= rand:
return dot
def natural_selection(self):
for dot in self.dots:
dot.fitness_dot()
self.sort_dots()
self.newDots.append(self.dots[0])
self.sum_fitness()
for i in range(1, len(self.dots)):
parent = self.get_parent()
self.newDots.append(Dot())
self.newDots[i].brain = parent.brain
self.newDots[i].brain.step = 0
self.dots = self.newDots
def mutate_dots(self):
for i in range(1, len(self.dots)):
rand = random.randint(0, 1000)
if rand > self.mutateChance:
self.dots[i].brain.directionsx = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.dots[i].brain.size / 2)))
self.dots[i].brain.directionsy = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.dots[i].brain.size / 2)))
return self.dots
# def natural_selection(self):
# self.selectedDots = []
# for dot in self.dots:
# dot.fitness_dot()
# self.sort_dots()
# for i in range(0, int(len(self.dots) / 3)):
# self.selectedDots.append(self.dots[i])
# self.selectedDots[i].tick = 0
# self.selectedDots[i].velx = 0
# self.selectedDots[i].vely = 0
# self.selectedDots[i].accx = 0
# self.selectedDots[i].accy = 0
# self.selectedDots[i].x = WIN_W / 2
# self.selectedDots[i].y = WIN_H - 10
# self.selectedDots[i].alive = True
# self.selectedDots[i].fitness = 0
# self.selectedDots[i].brain.step = 0
# self.selectedDots[i].goal = Goal()
#
# def new_dots(self):
# for i in range(len(self.selectedDots), len(self.dots)):
# self.selectedDots.append(Dot())
# self.dots = self.selectedDots
#
# def mutate_dots(self):
# for i, dot in enumerate(self.dots):
# isMutating = random.randint(0, 1000)
# if self.mutateChance > isMutating and i > int(len(self.dots) / 3) and i < (2 * int((len(self.dots) / 3))):
# for j in range(len(dot.brain.directionsx)):
# isMutatingDir = random.randint(0, 1000)
# if isMutatingDir >= 800:
# dot.brain.directionsx[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
# for j in range(len(dot.brain.directionsy)):
# isMutatingDir = random.randint(0, 1000)
# if isMutatingDir >= 800:
# dot.brain.directionsy[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
# return self.dots
'''
def natural_selection(self):
self.selectedDots = []
for dot in self.dots:
dot.fitness_dot()
self.sort_dots()
self.selectedDots = self.dots[0:int(0.3 * len(self.dots))]
def new_dots(self):
for i in range(len(self.dots) - int(0.3 * len(self.dots))):
self.selectedDots.append(self.dots[i])
self.dots = []
def mutate_dots(self):
for i, selectedDot in enumerate(self.selectedDots):
self.tick = 0
self.x = WIN_W / 2
self.y = WIN_H - 10
self.r = 3
self.alive = True
self.velLimit = 5
self.fitness = 0
self.dots = self.selectedDots
return self.dots
'''
'''
def mutate_dots(self):
for i, selectedDot in enumerate(self.selectedDots):
selectedDot.alive = True
if i >= 1:
isMutating = random.randint(0, 1000)
if isMutating <= self.mutateChance:
for j in range(len(selectedDot.brain.directionsx)):
isMutatingDir = random.randint(0, 1000)
if isMutatingDir >= 800:
selectedDot.brain.directionsx[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
for j in range(len(selectedDot.brain.directionsy)):
isMutatingDir = random.randint(0, 1000)
if isMutatingDir >= 800:
selectedDot.brain.directionsy[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
elif isMutating <= 800:
selectedDot.brain.directionsx = np.array(np.random.uniform(low=-2.5, high=2.5, size=200))
selectedDot.brain.directionsy = np.array(np.random.uniform(low=-2.5, high=2.5, size=200))
self.newDots.append(selectedDot)
return self.newDots
'''
None类型错误是由get_parent方法引起的。它搜索子点，但如果搜索失败则没有 return 值（与 return None 效果相同）。此代码将克服该错误
def get_parent(self):
rand = random.uniform(0, self.fitness_sum)
running_sum = 0
for dot in self.dots:
running_sum += dot.fitness
if running_sum >= rand:
return dot
return self.dots[0] # search failed, return 1st dot
在过去的几天里，我一直在尝试实现所谓的“智能点”游戏。我第一次看到它是在 Code Bullet youtube 频道：https://www.youtube.com/watch?v=BOZfhUcNiqk。不幸的是，它是用 Processing 语言编码的，而我几乎不知道的唯一语言是 Python。我完成了游戏的 python 版本，但出现了一些错误。
问题在于，在第二代中，被选为最佳点的点几乎立即停止移动。我认为这与我不擅长 OOP 和复制 Brain class 错误有关。步骤（我用于移动）在主循环的第一次或第二次循环中从零（在开始时设置）跳到最大值（200）。但问题还不止于此。在下一代，当我尝试将 brain step 设置为零时，它会中断：
AttributeError: 'NoneType' object has no attribute 'brain'
我尝试手动设置新大脑，但我仍然遇到同样的错误。如果有人已经做了这个或者有时间可以帮助我解决这个错误甚至项目，我将不胜感激。
我知道代码中有很多未使用的东西，但这只是我尝试修复它的结果 :(
注释掉的代码是我用过的一些旧代码。
main2.py（主循环）：
</p>
<pre><code>import pygame
import klase2
pygame.init()
def main():
win = pygame.display.set_mode((klase2.WIN_W, klase2.WIN_H))
clock = pygame.time.Clock()
population = klase2.Population()
dots = population.return_dots(1000)
goal = klase2.Goal()
run = True
while run:
clock.tick(30)
for event in pygame.event.get():
if event.type == pygame.QUIT:
run = False
win.fill((255, 255, 255))
goal.draw_goal(win)
for dot in dots:
dot.draw_dot(win)
dot.update_dot()
if population.all_dots_dead():
# natural selection
population.natural_selection()
# mutation
dots = population.mutate_dots()
population.gen += 1
print(population.gen)
pygame.display.update()
main()
kase2（处理所有功能和 classes）：
</p>
<pre><code>import pygame
import numpy as np
from pygame import gfxdraw
import math
import random
pygame.init()
WIN_W = 500
WIN_H = 500
class Brain:
def __init__(self, size):
self.step = 0
self.size = size
self.directionsx = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.size / 2)))
self.directionsy = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.size / 2)))
def clone(self):
self.size = self.size
self.directionsx = self.directionsx
self.directionsy = self.directionsy
self.step = 0
class Goal:
def __init__(self):
self.x = WIN_W / 2
self.y = 10
self.color = (255, 20, 20)
self.r = 5
def draw_goal(self, win):
pygame.gfxdraw.aacircle(win, int(self.x), int(self.y), self.r, self.color)
pygame.gfxdraw.filled_circle(win, int(self.x), int(self.y), self.r, self.color)
class Dot:
goal = Goal()
def __init__(self):
self.tick = 0
self.goal = Goal()
self.brain = Brain(400)
self.velx = 0
self.vely = 0
self.accx = 0
self.accy = 0
self.x = WIN_W / 2
self.y = WIN_H - 10
self.r = 3
self.color = (0, 0, 0)
self.alive = True
self.velLimit = 5
self.fitness = 0
def draw_dot(self, win):
pygame.gfxdraw.aacircle(win, int(self.x), int(self.y), self.r, self.color)
pygame.gfxdraw.filled_circle(win, int(self.x), int(self.y), self.r, self.color)
def move_dot(self):
if self.brain.size / 2 > self.brain.step:
self.accx = self.brain.directionsx[self.brain.step]
self.accy = self.brain.directionsy[self.brain.step]
self.brain.step += 1
else:
self.alive = False
self.velx += self.accx
self.vely += self.accy
if self.velx > self.velLimit:
self.velx = self.velLimit
elif self.velx < -self.velLimit:
self.velx = -self.velLimit
if self.vely > self.velLimit:
self.vely = self.velLimit
elif self.vely < -self.velLimit:
self.vely = -self.velLimit
self.x += self.velx
self.y += self.vely
def update_dot(self):
if not self.reached_goal():
self.tick += 1
if self.alive:
self.move_dot()
if self.x < 0 + self.r or self.x > WIN_W - self.r or self.y < 0 + self.r or self.y > WIN_H - self.r or self.reached_goal():
self.alive = False
def distance_to_goal(self):
a = abs(self.x - self.goal.x)
b = abs(self.y - self.goal.y)
return math.sqrt(a**2 + b**2)
def reached_goal(self):
if self.distance_to_goal() <= self.r + self.goal.r:
return True
return False
def fitness_dot(self):
if self.reached_goal():
self.fitness = 1 / (self.brain.step)
else:
self.fitness = 1 / (self.distance_to_goal()**2)
return self.fitness
class Population:
def __init__(self):
self.dots = []
self.newDots = []
self.gen = 0
self.mutateChance = 800
self.size = 0
self.fitness_sum = 0
def return_dots(self, size):
self.size = size
for _ in range(size):
self.dots.append(Dot())
return self.dots
def all_dots_dead(self):
for i in range(len(self.dots)):
if self.dots[i].alive:
return False
return True
def sort_dots(self):
self.dots = sorted(self.dots, key=lambda dot: dot.fitness, reverse=True)
def sum_fitness(self):
for dot in self.dots:
self.fitness_sum += dot.fitness_dot()
return self.fitness_sum
def get_parent(self):
rand = random.uniform(0, self.fitness_sum)
running_sum = 0
for dot in self.dots:
running_sum += dot.fitness
if running_sum >= rand:
return dot
def natural_selection(self):
for dot in self.dots:
dot.fitness_dot()
self.sort_dots()
self.newDots.append(self.dots[0])
self.sum_fitness()
for i in range(1, len(self.dots)):
parent = self.get_parent()
self.newDots.append(Dot())
self.newDots[i].brain = parent.brain
self.newDots[i].brain.step = 0
self.dots = self.newDots
def mutate_dots(self):
for i in range(1, len(self.dots)):
rand = random.randint(0, 1000)
if rand > self.mutateChance:
self.dots[i].brain.directionsx = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.dots[i].brain.size / 2)))
self.dots[i].brain.directionsy = np.array(np.random.uniform(low=-2.5, high=2.5, size=int(self.dots[i].brain.size / 2)))
return self.dots
# def natural_selection(self):
# self.selectedDots = []
# for dot in self.dots:
# dot.fitness_dot()
# self.sort_dots()
# for i in range(0, int(len(self.dots) / 3)):
# self.selectedDots.append(self.dots[i])
# self.selectedDots[i].tick = 0
# self.selectedDots[i].velx = 0
# self.selectedDots[i].vely = 0
# self.selectedDots[i].accx = 0
# self.selectedDots[i].accy = 0
# self.selectedDots[i].x = WIN_W / 2
# self.selectedDots[i].y = WIN_H - 10
# self.selectedDots[i].alive = True
# self.selectedDots[i].fitness = 0
# self.selectedDots[i].brain.step = 0
# self.selectedDots[i].goal = Goal()
#
# def new_dots(self):
# for i in range(len(self.selectedDots), len(self.dots)):
# self.selectedDots.append(Dot())
# self.dots = self.selectedDots
#
# def mutate_dots(self):
# for i, dot in enumerate(self.dots):
# isMutating = random.randint(0, 1000)
# if self.mutateChance > isMutating and i > int(len(self.dots) / 3) and i < (2 * int((len(self.dots) / 3))):
# for j in range(len(dot.brain.directionsx)):
# isMutatingDir = random.randint(0, 1000)
# if isMutatingDir >= 800:
# dot.brain.directionsx[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
# for j in range(len(dot.brain.directionsy)):
# isMutatingDir = random.randint(0, 1000)
# if isMutatingDir >= 800:
# dot.brain.directionsy[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
# return self.dots
'''
def natural_selection(self):
self.selectedDots = []
for dot in self.dots:
dot.fitness_dot()
self.sort_dots()
self.selectedDots = self.dots[0:int(0.3 * len(self.dots))]
def new_dots(self):
for i in range(len(self.dots) - int(0.3 * len(self.dots))):
self.selectedDots.append(self.dots[i])
self.dots = []
def mutate_dots(self):
for i, selectedDot in enumerate(self.selectedDots):
self.tick = 0
self.x = WIN_W / 2
self.y = WIN_H - 10
self.r = 3
self.alive = True
self.velLimit = 5
self.fitness = 0
self.dots = self.selectedDots
return self.dots
'''
'''
def mutate_dots(self):
for i, selectedDot in enumerate(self.selectedDots):
selectedDot.alive = True
if i >= 1:
isMutating = random.randint(0, 1000)
if isMutating <= self.mutateChance:
for j in range(len(selectedDot.brain.directionsx)):
isMutatingDir = random.randint(0, 1000)
if isMutatingDir >= 800:
selectedDot.brain.directionsx[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
for j in range(len(selectedDot.brain.directionsy)):
isMutatingDir = random.randint(0, 1000)
if isMutatingDir >= 800:
selectedDot.brain.directionsy[j] = np.random.uniform(low=-2.5, high=2.5, size=1)
elif isMutating <= 800:
selectedDot.brain.directionsx = np.array(np.random.uniform(low=-2.5, high=2.5, size=200))
selectedDot.brain.directionsy = np.array(np.random.uniform(low=-2.5, high=2.5, size=200))
self.newDots.append(selectedDot)
return self.newDots
'''
None类型错误是由get_parent方法引起的。它搜索子点，但如果搜索失败则没有 return 值（与 return None 效果相同）。此代码将克服该错误
def get_parent(self):
rand = random.uniform(0, self.fitness_sum)
running_sum = 0
for dot in self.dots:
running_sum += dot.fitness
if running_sum >= rand:
return dot
return self.dots[0] # search failed, return 1st dot