# This function generates a i+1th generation on passing the ith generation as the parameter. We have sorted the # vectors on basis seighted sum of training and validation error and taken the top six vectors of the ith generation # as the parents for the next generations. Two appraches have been followed to generate the next generation. # 1) This approach assigned weighted probablity to the parents for being selected # 2) This apprach gives equal probablity to each of the parents for being selected. def create_next_gen(population): return_population=np.zeros((population_size, MAX_DEGREE+2)) unmodulated_population=np.zeros((population_size, MAX_DEGREE+2)) for i in range(population_size): if parent_selection_type: par1 = random.randint(0, parent_range-1) par2=random.randint(0,parent_range-1) else: par1 = random.randint(1, 30) if par1<=5: par1=1 elif par1<=10: par1=2 elif par1<=15: par1=3 elif par1<=20: par1=4 elif par1<=25: par1=5 else: par1=6 par1-=1 par2=random.randint(1,30) if par2<=5: par2=1 elif par2<=10: par2=2 elif par2<=15: par2=3 elif par2<=20: par2=4 elif par2<=25: par2=5 else: par2=6 par2-=1 # boss+=1 # if ch[11]<=650000000000 and ch[12]<=650000000000: # To make sure we only store the good vectors (used for case of high modulation) # return_population.append(ch) # cnt+=1 # if boss>=150: # break ch=cross_over(population[par1], population[par2]) unmodulated_population[i]=ch[1] ch=ch[0] ch=get_score(ch) return_population[i]=(ch) return (return_population, unmodulated_population)