def search():
#start with N random genetic codes
codes = [ GeneticCode() for _ in range( 0 , N ) ]
for gen in range( 0 , T+1 ):
sumFitness = sum( [ x.get_fitness() for x in codes ] )
#pick K pairs of states and cross them.
for childNum in range( 0 , K ):
#select first parent proportional to fitness
key1 = uniform( 0 , sumFitness )
currSum = 0
for code in codes:
currSum += code.get_fitness()
if ( currSum > key1 ):
parent1 = code
break
#select second parent proportional to fitness
key2 = uniform( 0 , sumFitness )
currSum = 0
for code in codes:
currSum += code.get_fitness()
if ( currSum > key2 ):
parent2 = code
break
child = GeneticCode.cross( parent1 , parent2 )
child.mutate()
codes.append( child )
#save the image every 1000 children
if ( (K * gen + childNum) % 1000 == 0 ):
bestCode = max( codes , key = lambda p : p.get_fitness() )
print "Processing generation " + str( gen ) + "; best fitness = " + str( bestCode.get_fitness() )
Search.save_code_as_image( bestCode , gen )
#keep N best states
codes.sort( Search.fitness_comparator )
codes = codes[ 0:N ]
#repeat T times
#save the last image
#print "Processing generation " + str( T-1 )
#Search.save_code_as_image( codes[ 0 ] , T-1 )
return codes
