Welcome, guest | Sign In | My Account | Store | Cart

A simple genetic algorithm program. I followed this tutorial to make the program http://www.ai-junkie.com/ga/intro/gat1.html.

The objective of the code is to evolve a mathematical expression which calculates a user-defined target integer.


KEY:

chromosome = binary list (this is translated/decoded into a protein in the format number --> operator --> number etc, any genes (chromosome is read in blocks of four) which do not conform to this are ignored.

protein = mathematical expression (this is evaluated from left to right in number + operator blocks of two)

output = output of protein (mathematical expression)

error = inverse of difference between output and target

fitness score = a fraction of sum of of total errors


OTHER:

One-point crossover is used.

I have incorporated elitism in my code, which somewhat deviates from the tutorial but made my code more efficient (top ~7% of population are carried through to next generation)

Python, 273 lines
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
from operator import itemgetter, attrgetter
import random
import sys
import os
import math
import re


# GLOBAL VARIABLES

genetic_code = {
  '0000':'0',
  '0001':'1',
  '0010':'2',
  '0011':'3',
  '0100':'4',
  '0101':'5',
  '0110':'6',
  '0111':'7',
  '1000':'8',
  '1001':'9',
  '1010':'+',
  '1011':'-',
  '1100':'*',
  '1101':'/'
  }

solution_found = False
popN = 100 # n number of chromos per population
genesPerCh = 75
max_iterations = 1000
target = 1111.0
crossover_rate = 0.7
mutation_rate = 0.05

"""Generates random population of chromos"""
def generatePop ():
  chromos, chromo = [], []
  for eachChromo in range(popN):
    chromo = []
    for bit in range(genesPerCh * 4):
      chromo.append(random.randint(0,1))
    chromos.append(chromo)
  return chromos

"""Takes a binary list (chromo) and returns a protein (mathematical expression in string)"""
def translate (chromo):
  protein, chromo_string = '',''
  need_int = True
  a, b = 0, 4 # ie from point a to point b (start to stop point in string)
  for bit in chromo:
    chromo_string += str(bit)
  for gene in range(genesPerCh):
    if chromo_string[a:b] == '1111' or chromo_string[a:b] == '1110': 
      continue
    elif chromo_string[a:b] != '1010' and chromo_string[a:b] != '1011' and chromo_string[a:b] != '1100' and chromo_string[a:b] != '1101':
      if need_int == True:
        protein += genetic_code[chromo_string[a:b]]
        need_int = False
        a += 4
        b += 4
        continue
      else:
        a += 4
        b += 4
        continue
    else:
      if need_int == False:
        protein += genetic_code[chromo_string[a:b]]
        need_int = True
        a += 4
        b += 4
        continue
      else:
        a += 4
        b += 4
        continue
  if len(protein) %2 == 0:
    protein = protein[:-1]
  return protein
  
"""Evaluates the mathematical expressions in number + operator blocks of two"""
def evaluate(protein):
  a = 3
  b = 5
  output = -1
  lenprotein = len(protein) # i imagine this is quicker than calling len everytime?
  if lenprotein == 0:
    output = 0
  if lenprotein == 1:
    output = int(protein)
  if lenprotein >= 3:
    try :
      output = eval(protein[0:3])
    except ZeroDivisionError:
      output = 0
    if lenprotein > 4:
      while b != lenprotein+2:
        try :
          output = eval(str(output)+protein[a:b])
        except ZeroDivisionError:
          output = 0
        a+=2
        b+=2  
  return output

"""Calulates fitness as a fraction of the total fitness"""
def calcFitness (errors):
  fitnessScores = []
  totalError = sum(errors)
  i = 0
  # fitness scores are a fraction of the total error
  for error in errors:
    fitnessScores.append (float(errors[i])/float(totalError))
    i += 1
  return fitnessScores

def displayFit (error):
  bestFitDisplay = 100
  dashesN = int(error * bestFitDisplay)
  dashes = ''
  for j in range(bestFitDisplay-dashesN):
    dashes+=' '
  for i in range(dashesN):
    dashes+='+'
  return dashes


"""Takes a population of chromosomes and returns a list of tuples where each chromo is paired to its fitness scores and ranked accroding to its fitness"""
def rankPop (chromos):
  proteins, outputs, errors = [], [], []
  i = 1
  # translate each chromo into mathematical expression (protein), evaluate the output of the expression,
  # calculate the inverse error of the output
  print '%s: %s\t=%s \t%s %s' %('n'.rjust(5), 'PROTEIN'.rjust(30), 'OUTPUT'.rjust(10), 'INVERSE ERROR'.rjust(17), 'GRAPHICAL INVERSE ERROR'.rjust(105))
  for chromo in chromos: 
    protein = translate(chromo)
    proteins.append(protein)
    
    output = evaluate(protein)
    outputs.append(output)
    
    try:
      error = 1/math.fabs(target-output)
    except ZeroDivisionError:
      global solution_found
      solution_found = True
      error = 0
      print '\nSOLUTION FOUND' 
      print '%s: %s \t=%s %s' %(str(i).rjust(5), protein.rjust(30), str(output).rjust(10), displayFit(1.3).rjust(130))
      break
    else:
      #error = 1/math.fabs(target-output)
      errors.append(error)
    print '%s: %s \t=%s \t%s %s' %(str(i).rjust(5), protein.rjust(30), str(output).rjust(10), str(error).rjust(17), displayFit(error).rjust(105))
    i+=1  
  fitnessScores = calcFitness (errors) # calc fitness scores from the erros calculated
  pairedPop = zip ( chromos, proteins, outputs, fitnessScores) # pair each chromo with its protein, ouput and fitness score
  rankedPop = sorted ( pairedPop,key = itemgetter(-1), reverse = True ) # sort the paired pop by ascending fitness score
  return rankedPop

""" taking a ranked population selects two of the fittest members using roulette method"""
def selectFittest (fitnessScores, rankedChromos):
  while 1 == 1: # ensure that the chromosomes selected for breeding are have different indexes in the population
    index1 = roulette (fitnessScores)
    index2 = roulette (fitnessScores)
    if index1 == index2:
      continue
    else:
      break

  
  ch1 = rankedChromos[index1] # select  and return chromosomes for breeding 
  ch2 = rankedChromos[index2]
  return ch1, ch2

"""Fitness scores are fractions, their sum = 1. Fitter chromosomes have a larger fraction.  """
def roulette (fitnessScores):
  index = 0
  cumalativeFitness = 0.0
  r = random.random()
  
  for i in range(len(fitnessScores)): # for each chromosome's fitness score
    cumalativeFitness += fitnessScores[i] # add each chromosome's fitness score to cumalative fitness

    if cumalativeFitness > r: # in the event of cumalative fitness becoming greater than r, return index of that chromo
      return i


def crossover (ch1, ch2):
  # at a random chiasma
  r = random.randint(0,genesPerCh*4)
  return ch1[:r]+ch2[r:], ch2[:r]+ch1[r:]


def mutate (ch):
  mutatedCh = []
  for i in ch:
    if random.random() < mutation_rate:
      if i == 1:
        mutatedCh.append(0)
      else:
        mutatedCh.append(1)
    else:
      mutatedCh.append(i)
  #assert mutatedCh != ch
  return mutatedCh
      
"""Using breed and mutate it generates two new chromos from the selected pair"""
def breed (ch1, ch2):
  
  newCh1, newCh2 = [], []
  if random.random() < crossover_rate: # rate dependent crossover of selected chromosomes
    newCh1, newCh2 = crossover(ch1, ch2)
  else:
    newCh1, newCh2 = ch1, ch2
  newnewCh1 = mutate (newCh1) # mutate crossovered chromos
  newnewCh2 = mutate (newCh2)
  
  return newnewCh1, newnewCh2

""" Taking a ranked population return a new population by breeding the ranked one"""
def iteratePop (rankedPop):
  fitnessScores = [ item[-1] for item in rankedPop ] # extract fitness scores from ranked population
  rankedChromos = [ item[0] for item in rankedPop ] # extract chromosomes from ranked population

  newpop = []
  newpop.extend(rankedChromos[:popN/15]) # known as elitism, conserve the best solutions to new population
  
  while len(newpop) != popN:
    ch1, ch2 = [], []
    ch1, ch2 = selectFittest (fitnessScores, rankedChromos) # select two of the fittest chromos
        
    ch1, ch2 = breed (ch1, ch2) # breed them to create two new chromosomes 
    newpop.append(ch1) # and append to new population
    newpop.append(ch2)
  return newpop
  
      
def configureSettings ():
  configure = raw_input ('T - Enter Target Number \tD - Default settings: ')
  match1 = re.search( 't',configure, re.IGNORECASE )
  if match1:
    global target
    target = input('Target int: ' )

def main(): 
  configureSettings ()
  chromos = generatePop() #generate new population of random chromosomes
  iterations = 0

  while iterations != max_iterations and solution_found != True:
    # take the pop of random chromos and rank them based on their fitness score/proximity to target output
    rankedPop = rankPop(chromos) 
    
    print '\nCurrent iterations:', iterations
    
    if solution_found != True:
      # if solution is not found iterate a new population from previous ranked population
      chromos = []
      chromos = iteratePop(rankedPop)
            
      iterations += 1
    else:
      break

  
  
  
  
    
if __name__ == "__main__":
    main()

I am happy to accept any criticism or comments for improvements.

6 comments

David Adler (author) 11 years, 11 months ago  # | flag

Sorry that it is a bit littered with tests!

Darren Stanney 11 years, 1 month ago  # | flag

Hi David,

in the selectFitness() function I don't see the need to ensure that the two chromo are different; in fact I feel that this actually hinders the GA from converging + introduces an unneeded bottleneck to the code.

If a self cross is weak the chromo will get weeded out naturally; and if the chromo is strong it will proliferate more quickly than if you didn't allow a self cross.

If you try the following it may improve your results:

def selectFittest (fitnessScores, rankedChromos):
  #while 1 == 1: # ensure that the chromosomes selected for breeding are have different indexes in the population
  index1 = roulette (fitnessScores)
  index2 = roulette (fitnessScores)
  #if index1 == index2:
  #continue
  #else:
  #break


  ch1 = rankedChromos[index1] # select  and return chromosomes for breeding 
  ch2 = rankedChromos[index2]
  return ch1, ch2

Regards Darren.

David Adler (author) 11 years ago  # | flag

hi Darren,

Fair point, I haven't tested which one is more efficient but these are the reasons i did it the way i did:

  1. from a biological perspective you can't cross the same gene with itself
  2. we use elitism anyway so perhaps less concern about eliminating good ones and more conern about introducing variation
  3. I originally introduced it for debugging
deep 10 years, 2 months ago  # | flag

Hii David

How to implement Genetic Algorithm for classifying Biological database which contain DNA string??

samuel ilesanmi 7 years, 8 months ago  # | flag

HI david, can you help on "python implementation of genetic algorithm for student performance system in lets say computer science department

its a for a final year project, i'd appreciate if you can help out. Thanks

Dillan Campbell 7 years, 6 months ago  # | flag

how would oi go about making it so i can visually see the process instead of having it close as soon as it is done?