Last active
May 11, 2022 17:27
-
-
Save NicolleLouis/d4f88d5bd566298d4279bcb69934f51d to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#!/usr/bin/python3.5 | |
# -*-coding:Utf-8 -* | |
import random | |
import operator | |
import time | |
import matplotlib.pyplot as plt | |
temps1 = time.time() | |
#genetic algorithm function | |
def fitness (password, test_word): | |
score = 0 | |
i = 0 | |
while (i < len(password)): | |
if (password[i] == test_word[i]): | |
score+=1 | |
i+=1 | |
return score * 100 / len(password) | |
def generateAWord (length): | |
i = 0 | |
result = "" | |
while i < length: | |
letter = chr(97 + int(26 * random.random())) | |
result += letter | |
i +=1 | |
return result | |
def generateFirstPopulation(sizePopulation, password): | |
population = [] | |
i = 0 | |
while i < sizePopulation: | |
population.append(generateAWord(len(password))) | |
i+=1 | |
return population | |
def computePerfPopulation(population, password): | |
populationPerf = {} | |
for individual in population: | |
populationPerf[individual] = fitness(password, individual) | |
return sorted(populationPerf.items(), key = operator.itemgetter(1), reverse=True) | |
def selectFromPopulation(populationSorted, best_sample, lucky_few): | |
nextGeneration = [] | |
for i in range(best_sample): | |
nextGeneration.append(populationSorted[i][0]) | |
for i in range(lucky_few): | |
nextGeneration.append(random.choice(populationSorted)[0]) | |
random.shuffle(nextGeneration) | |
return nextGeneration | |
def createChild(individual1, individual2): | |
child = "" | |
for i in range(len(individual1)): | |
if (int(100 * random.random()) < 50): | |
child += individual1[i] | |
else: | |
child += individual2[i] | |
return child | |
def createChildren(breeders, number_of_child): | |
nextPopulation = [] | |
for i in range(len(breeders)/2): | |
for j in range(number_of_child): | |
nextPopulation.append(createChild(breeders[i], breeders[len(breeders) -1 -i])) | |
return nextPopulation | |
def mutateWord(word): | |
index_modification = int(random.random() * len(word)) | |
if (index_modification == 0): | |
word = chr(97 + int(26 * random.random())) + word[1:] | |
else: | |
word = word[:index_modification] + chr(97 + int(26 * random.random())) + word[index_modification+1:] | |
return word | |
def mutatePopulation(population, chance_of_mutation): | |
for i in range(len(population)): | |
if random.random() * 100 < chance_of_mutation: | |
population[i] = mutateWord(population[i]) | |
return population | |
def nextGeneration (firstGeneration, password, best_sample, lucky_few, number_of_child, chance_of_mutation): | |
populationSorted = computePerfPopulation(firstGeneration, password) | |
nextBreeders = selectFromPopulation(populationSorted, best_sample, lucky_few) | |
nextPopulation = createChildren(nextBreeders, number_of_child) | |
nextGeneration = mutatePopulation(nextPopulation, chance_of_mutation) | |
return nextGeneration | |
def multipleGeneration(number_of_generation, password, size_population, best_sample, lucky_few, number_of_child, chance_of_mutation): | |
historic = [] | |
historic.append(generateFirstPopulation(size_population, password)) | |
for i in range (number_of_generation): | |
historic.append(nextGeneration(historic[i], password, best_sample, lucky_few, number_of_child, chance_of_mutation)) | |
return historic | |
#print result: | |
def printSimpleResult(historic, password, number_of_generation): #bestSolution in historic. Caution not the last | |
result = getListBestIndividualFromHistorique(historic, password)[number_of_generation-1] | |
print ("solution: \"" + result[0] + "\" de fitness: " + str(result[1])) | |
#analysis tools | |
def getBestIndividualFromPopulation (population, password): | |
return computePerfPopulation(population, password)[0] | |
def getListBestIndividualFromHistorique (historic, password): | |
bestIndividuals = [] | |
for population in historic: | |
bestIndividuals.append(getBestIndividualFromPopulation(population, password)) | |
return bestIndividuals | |
#graph | |
def evolutionBestFitness(historic, password): | |
plt.axis([0,len(historic),0,105]) | |
plt.title(password) | |
evolutionFitness = [] | |
for population in historic: | |
evolutionFitness.append(getBestIndividualFromPopulation(population, password)[1]) | |
plt.plot(evolutionFitness) | |
plt.ylabel('fitness best individual') | |
plt.xlabel('generation') | |
plt.show() | |
def evolutionAverageFitness(historic, password, size_population): | |
plt.axis([0,len(historic),0,105]) | |
plt.title(password) | |
evolutionFitness = [] | |
for population in historic: | |
populationPerf = computePerfPopulation(population, password) | |
averageFitness = 0 | |
for individual in populationPerf: | |
averageFitness += individual[1] | |
evolutionFitness.append(averageFitness/size_population) | |
plt.plot(evolutionFitness) | |
plt.ylabel('Average fitness') | |
plt.xlabel('generation') | |
plt.show() | |
#variables | |
password = "banana" | |
size_population = 100 | |
best_sample = 20 | |
lucky_few = 20 | |
number_of_child = 5 | |
number_of_generation = 50 | |
chance_of_mutation = 5 | |
#program | |
if ((best_sample + lucky_few) / 2 * number_of_child != size_population): | |
print ("population size not stable") | |
else: | |
historic = multipleGeneration(number_of_generation, password, size_population, best_sample, lucky_few, number_of_child, chance_of_mutation) | |
printSimpleResult(historic, password, number_of_generation) | |
evolutionBestFitness(historic, password) | |
evolutionAverageFitness(historic, password, size_population) | |
print time.time() - temps1 |
@MarlieChiller
For that issue, you simply need to change that "for i in int(range(len(breeders) / 2)):" to "for i in int(range(len(breeders) // 2)):" -> since range function itself cannot take float type, the error occurs. "//" operator leaves only quotients, dropping remainders.
I've tried it and the code worked. I hope it will be helpful.
@danelee2601
It doesn't work.
for i in range(int(len(breeders)/2)): <--it works.
print time.time() - temps1 gives syntax error
divyagangadhar, if you're using python 3, try
print (time.time() - temps1)
that should fix it
hey, can you just help me by explaining the 2 output graphs. i.e., fitness best individual vs generation and average fitness vs generation
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
for i in int(range(len(breeders) / 2)):
(Line 70) produces this error:
TypeError: 'float' object cannot be interpreted as an integer - when you try and run it out of the box