Kernel: Anaconda (Python 3)
In [29]:
import numpy as np import matplotlib.pyplot as plt %matplotlib notebook plt.style.use('ggplot') import random from random import randint import math import h5py
In [23]:
food_amount = 20 pacman_amount = 50 board_size = 1000 default_health = 1500 pacmen = [] # global array of all the pacmen food = [] # '' of the food crot = 1 #cost of rotation cloc = .0005 # cost of locamotion cbite = 1 #cost of bite turn_loss = 1000 # loss each turn (starvation factor) phi_prec = (80/255.) # params used for binary genes, this is field of view ex out of 80 degrees alpha_prec = 10/255. # not implemented, but is angular error delta_prec = 10/255. #error for depth pacman_radius = 1 # raduis for the bite to be sucessful food_health = 3000 # health gained from a sucessful bite mutation_freq = .15 # percent any 0/1 would flip for each gene generations = 500 # num of gens turns = 200 #num of turns in each gen
In [ ]:
In [24]:
class pacman: # class of pacman has everything needed, including multple properly def __init__(self): #change dynamically self.x = random.uniform(49*board_size/100,51*board_size/100) self.y = random.uniform(49*board_size/100,51*board_size/100) #this is set so the prgression will be outward self.theta = randint(0,359) self.health = default_health self.active = 1 #dna self.phi = randint(0,255) # field of view self.alpha = randint(0,255) #angle error self.delta = randint(0,255) # depth error self.phi_val = self.phi*phi_prec # converts binary to a float self.alpha_val = self.alpha*alpha_prec self.delta_val = self.delta*delta_prec #self.f = closest food #self.dist = closest food distance def __bool__(self): return True def __repr__(self): #return "{pacman, x:"+str(self.x)+", y:"+str(self.y)+", theta:"+str(self.theta)+"}" return str(self.health)+"|"+str(self.phi_val) def find_dist(self,f): return np.sqrt((self.x-f.x)**2 + (self.y-f.y)**2) def find_angle(self,f): dx = f.x - self.x dy = f.y - self.y arc = math.degrees(math.atan((dy/dx))) theta1 = 1 if dx > 0: theta1 = 180 - arc #this used to be an if, but i think arctan returns a negative when a negative is inputted. #therefor the two would be equivalent else: if dy > 0: theta1 = np.absolute(arc) #this is absolute because the angle will be a negative else: theta1 = 360 - arc #theta1 is the angle the food is from the x axis #returned is the difference in the angles return np.absolute(theta1 - self.theta) def check_angle(self,f):#check if the food is in the range diff = (self.phi_val/2)#this creates virutal bounds f_angle = self.find_angle(f) if(f_angle < diff): return True return False def find_food(self): global food closest = False closest_dist = False for i in food: dist = self.find_dist(i) if((dist < closest_dist or not closest_dist) and (self.check_angle(i))): closest = i closest_dist = dist self.f = closest self.dist = closest_dist if(closest != False): return True return False def rotate_to_food(self):# returns (in radians!!!) the angle needed to get the food with genetic error #no cost to rotation dx = self.f.x - self.x dy = self.f.y - self.y arc = math.atan((dy/dx)) error = random.uniform(self.alpha_val/-2., self.alpha_val/2.) new_theta = arc + error return new_theta def get_food(self): theta = self.rotate_to_food() #warning this is in radians... #theta is not used... error = random.uniform(self.delta_val/-2., self.delta_val/2.) new_dist = self.dist + (error*self.dist) #the error for distance is relative to the distance, ie the greater the dist the greater the error dx = self.f.x - self.x dy = self.f.y - self.y distx = new_dist*(dx/self.dist)#equivalent to cos, error is not associted with the angle yet disty = new_dist*(dy/self.dist)#equivalent to sin self.x += distx self.y += disty cost = cloc*new_dist self.health -= cost #cost of rotate def bite(self): to_food = self.find_dist(self.f) self.health -= cbite if(to_food < pacman_radius):#means the bite was sucessful self.health += food_health self.f.reset() self.check_bounds() def check_bounds(self): if(self.x > board_size or self.x < 0): self.x = self.x % board_size #if(self.x < 0): # self.x = board_size + self.x if(self.y > board_size or self.y < 0): self.y = self.y % board_size def random_rotate(self): self.theta = randint(0,359) #self.health += randint(-2,1) def step(self): self.health -= turn_loss if(self.find_food()): self.get_food() self.bite() self.random_rotate() def get_genes(self): return[self.phi,self.alpha,self.delta] def set_genes(self,genes): self.phi = genes[0] self.alpha = genes[1] self.delta = genes[2] def to_bits(self): genes = self.get_genes() split_genes = [] for i in genes: bin_str = format(i, '#010b') bin_str = bin_str[2:] split_genes.append([bin_str[:4],bin_str[4:]]) return split_genes def mutate(self, gene): #checks each 1/0 and finds a random num and if it below the threshold it mutates new_gene = '' for i in gene: rand = random.uniform(0,1) if(rand < mutation_freq): #this means the dna is flipping num = int(i) if(num == 1): num = 0 else: num = 1 new_gene += str(num) else: new_gene += str(i) return new_gene def reproduce(self,other): son1 = pacman() son2 = pacman() son1_genes = [] son2_genes = [] mybits = self.to_bits() otherbits = other.to_bits() for i,v in enumerate(mybits): son1_gene = v[0]+otherbits[i][1] son1_gene = self.mutate(son1_gene) son2_gene = otherbits[i][0]+v[1] son2_gene = self.mutate(son2_gene) son1_gene = int('0b'+son1_gene,base=2) son2_gene = int('0b'+son2_gene,base=2) son1_genes.append(son1_gene) son2_genes.append(son2_gene) son1.set_genes(son1_genes) son2.set_genes(son2_genes) return [son1,son2]
In [25]:
class dot: def __init__(self): self.x = random.uniform(0, board_size) self.y = random.uniform(0, board_size) self.active = 1 def __repr__(self): return "{dot, x:"+str(self.x)+", y:"+str(self.y)+"}" def reset(self): self.x = random.uniform(0, board_size) self.y = random.uniform(0, board_size)
In [26]:
def gen1(): global pacmen pacmen = [] for i in range(0,pacman_amount): p = pacman() pacmen.append(p) pacmen = np.array(pacmen) pac_data = [] for d,v in enumerate(pacmen): pac_data.append([]) pac_data[d].append(v.x) pac_data[d].append(v.y) pac_data[d].append(v.theta) pac_data[d].append(v.health) pac_data[d].append(v.phi_val) pac_data[d].append(v.delta_val) #pac_turn_data_first.append(pac_data) return pacmen def food1(): global food food = [] for i in range(0,food_amount): f = dot() food.append(f) food = np.array(food) return food def reproduce1(healthy_pacmen):# takes an array of pacmen and reproduces each with a neibor #inactive for now sons = [] for i,v in enumerate(healthy_pacmen): if( i % 2 == 1): temp = v.reproduce(healthy_pacmen[randint(0,len(healthy_pacmen)-1)]) sons.extend(temp) return sons def find_average_health(p,x=10): #takes an array of pacmen and finds the averge health #it will also sort the pacemen by health and return the helath of the xth pacman pacmen_ranked = [] avg_health = 0 for i in p: pacmen_ranked.append(i.health) avg_health += i.health avg_health = avg_health/len(p) pacmen_ranked = np.array(sorted(pacmen_ranked)) med_health = pacmen_ranked[x] #print(avg_health, med_health) return avg_health, med_health def reproduce2(healthy_pacmen, x):# takes an array of pacmen and reproduces randomly with other heathy_pacmen # x is how many are cut off in the evolution sons = [] total = int(len(healthy_pacmen)/2) + int(x/2) + 1 for i in range(0,total): rand1 = randint(0,len(healthy_pacmen)-1) rand2 = randint(0,len(healthy_pacmen)-1) p1 = healthy_pacmen[rand1] p2 = healthy_pacmen[rand2] temp = p1.reproduce(p2) sons.extend(temp) return sons def reproduce3(healthy_pacmen): myscale = len(pacmen)/2 sons = [] total = int(len(pacmen)/2) for i in range(0,total): rand1 = np.random.exponential(scale=myscale) rand2 = np.random.exponential(scale=myscale) while rand1 > len(pacmen) - 1: rand1 = np.random.exponential(scale=myscale) while rand2 > len(pacmen) - 1: rand2 = np.random.exponential(scale=myscale) rand1 = int(rand1/2) rand2 = int(rand2/2) #print(rand1,rand2) p1 = healthy_pacmen[rand1] p2 = healthy_pacmen[rand2] temp = p1.reproduce(p2) sons.extend(temp) return sons def average_genes(p):#phi alpha delta avg1 = 0 avg2 = 0 avg3 = 0 for i in p: avg1 += i.phi_val avg2 += i.alpha_val avg3 += i.delta_val avg1 = avg1/len(p) avg2 = avg2/len(p) avg3 = avg3/len(p) return avg1,avg2,avg3 def bubbleSort(unranked): for passnum in range(len(unranked)-1,0,-1): for i in range(passnum): if unranked[i].health>unranked[i+1].health: temp = unranked[i] unranked[i] = unranked[i+1] unranked[i+1] = temp
In [27]:
def driver(gen,life_span,generations): food = food1() dt = 1 t = [0] tracker = [] genetics = [[],[],[]] cycles = [] global pacmen for g in range(0,generations): #iterates through each generation tracker.append([]) avg1,avg2,avg3 = average_genes(pacmen) genetics[0].append(avg1) genetics[1].append(avg2) genetics[2].append(avg3) for i in pacmen: tracker[g].append([]) cycles.append([]) t.append(t[-1]+dt) for q in range(0,life_span): #runs through time with current generation for i,v in enumerate(pacmen): tracker[g][i].append([v.x,v.y,v.theta,v.health,v.phi,v.delta]) v.step() cycles[g].append([])# note cycles starts after the first gen cycles[g][q].append([]) cycles[g][q].append([]) cycles[g][q].append([]) cycles[g][q].append([]) for w in pacmen: cycles[g][q][0].append(w.x) cycles[g][q][1].append(w.y) for f in food: cycles[g][q][2].append(f.x) cycles[g][q][3].append(f.y) #makes the next generation #x = 150 # number of pacmen dieing #healthy_pacmen = [] #avg,med = find_average_health(pacmen,x) #for i,v in enumerate(pacmen): #end of a generation mean reproduce #if(v.health > med): #healthy_pacmen.append(v) #this system kills off the bottom x pacmen unranked = pacmen bubbleSort(unranked) ranked = unranked[::-1] #print(ranked) sons = reproduce3(ranked) pacmen = sons # if it is the last gen it will save the data if(g == generations - 1): genN_file = h5py.File("genN.h5","w") # the first argument is the filename, second says we're allowing writes ("w"). Read only is "r". genN_file['/genN'] = tracker[generations-1] genN_file.close() gen1_file = h5py.File("gen1.h5","w") # the first argument is the filename, second says we're allowing writes ("w"). Read only is "r". gen1_file['/gen1'] = tracker[0] gen1_file.close() return t,tracker,genetics,cycles
In [28]:
t,tracker,genetics,cycles = driver(gen1(),turns,generations)#turns then generations
In [14]:
#remember to check both params out plt.figure() plt.plot(genetics[0])# should increase #plt.plot(genetics[1])#should decrease #plt.plot(genetics[2])#should decrease # error is disabled
<IPython.core.display.Javascript object>
[<matplotlib.lines.Line2D at 0x7f77024d6c88>]
In [ ]:
tracker = np.array(tracker) figs = 8 gen = 0 f, axarr = plt.subplots(figs, sharex=True,sharey=True) for i in range(0,figs): axarr[i].scatter(cycles[gen][i][0],cycles[gen][i][1]) axarr[i].scatter(cycles[gen][i][2],cycles[gen][i][3], marker='^',c='red')
In [ ]:
infile = h5py.File("gen1.h5", "r") data_from_disk = infile['/gen1'][:] infile.close() print(data_from_disk) print('next!!!!') infile = h5py.File("genN.h5", "r") data_from_disk = infile['/genN'][:] infile.close() print(data_from_disk)
In [ ]:
In [ ]: