Genetski algoritmi (GA) so prilagodljivi hevristični iskalni algoritmi, ki sodijo v večji del evolucijskih algoritmov. Genetski algoritmi temeljijo na idejah naravne selekcije in genetike. To je inteligentno izkoriščanje naključnih iskanj, opremljenih s preteklimi podatki, za usmerjanje iskanja v regijo boljše zmogljivosti v prostoru rešitev. Običajno se uporabljajo za ustvarjanje visokokakovostnih rešitev za težave z optimizacijo in iskanjem.
Genetski algoritmi simulirajo proces naravne selekcije kar pomeni, da lahko tiste vrste, ki se lahko prilagodijo spremembam v okolju, preživijo in se razmnožujejo ter preidejo v naslednjo generacijo. Preprosto povedano, simulirajo preživetje najmočnejših med posamezniki zaporednih generacij za rešitev problema. Vsaka generacija je sestavljena iz populacije osebkov in vsak posameznik predstavlja točko v iskalnem prostoru in možni rešitvi. Vsak posameznik je predstavljen kot niz znakov/celih števil/float/bitov. Ta niz je analogen kromosomu.
Osnova genetskih algoritmov
Genetski algoritmi temeljijo na analogiji z genetsko strukturo in obnašanjem kromosomov populacije. Sledi osnova GA, ki temelji na tej analogiji –
- Posamezniki v populaciji tekmujejo za vire in partnerja
- Tisti posamezniki, ki so uspešni (najbolj primerni), se nato parijo in ustvarijo več potomcev kot drugi
- Geni najsposobnejšega starša se širijo skozi generacijo, kar pomeni, da starši včasih ustvarijo potomce, ki so boljši od katerega koli od staršev.
- Tako je vsaka naslednja generacija bolj primerna za svoje okolje.
Iskanje prostora
Populacija posameznikov se ohranja v iskalnem prostoru. Vsak posameznik predstavlja rešitev v iskalnem prostoru za dano težavo. Vsak posameznik je kodiran kot vektor končne dolžine (analogno kromosomu) komponent. Te spremenljive komponente so analogne genom. Tako je kromosom (posameznik) sestavljen iz več genov (spremenljivih komponent).
Ocena telesne pripravljenosti
Vsakemu posamezniku se dodeli ocena telesne pripravljenosti, ki kaže tekmovalno sposobnost posameznika . Išče se posameznik z optimalnim rezultatom telesne pripravljenosti (ali skoraj optimalnim).
GA vzdržuje populacijo n posameznikov (kromosomov/raztopin) skupaj z njihovimi rezultati telesne pripravljenosti. Posamezniki z boljšimi rezultati telesne pripravljenosti imajo več možnosti za razmnoževanje kot drugi. Izbrani so posamezniki z boljšimi rezultati telesne pripravljenosti, ki se parijo in proizvajajo boljši podmladek z združevanjem kromosomov staršev. Velikost prebivalstva je statična, zato je treba ustvariti prostor za nove prišleke. Tako nekateri posamezniki umrejo in jih nadomestijo novi prišleki, ki na koncu ustvarijo novo generacijo, ko so izčrpane vse možnosti parjenja stare populacije. Upati je, da bodo v naslednjih generacijah prispele boljše rešitve, medtem ko bodo najmanj primerni umrli.
Vsaka nova generacija ima v povprečju več boljših genov kot posamezna (rešitev) prejšnjih generacij. Tako ima vsaka nova generacija boljšega delne rešitve kot prejšnje generacije. Ko se proizvedeni potomci ne razlikujejo bistveno od potomcev prejšnjih populacij, se populacija konvergira. Algoritem naj bi bil konvergiran k naboru rešitev za problem.
Operatorji genetskih algoritmov
Ko je ustvarjena začetna generacija, jo algoritem razvije z uporabo naslednjih operatorjev –
1) Izbirni operater: Ideja je dati prednost posameznikom z dobrimi rezultati telesne pripravljenosti in jim omogočiti prenos svojih genov na naslednje generacije.
2) Navzkrižni operater: To predstavlja parjenje med posamezniki. Dva posameznika sta izbrana z uporabo izbirnega operaterja, mesta križanja pa so izbrana naključno. Nato se geni na teh mestih križanja izmenjajo in tako nastane popolnoma nov posameznik (potomec). Na primer –
3) Operator mutacije: Ključna ideja je vstavljanje naključnih genov v potomce, da se ohrani raznolikost v populaciji in se izogne prezgodnji konvergenci. Na primer –
zemljevid v tipkopisu
Celoten algoritem lahko povzamemo kot –
1) Randomly initialize populations p 2) Determine fitness of population 3) Until convergence repeat: a) Select parents from population b) Crossover and generate new population c) Perform mutation on new population d) Calculate fitness for new population>
Primer problema in rešitve z uporabo genetskih algoritmov
Glede na ciljni niz je cilj ustvariti ciljni niz, ki se začne z naključnim nizom enake dolžine. V naslednji izvedbi so narejene naslednje analogije –
- Znaki A-Z, a-z, 0-9 in drugi posebni simboli se štejejo za gene
- Niz, ki ga ustvarijo ti znaki, se obravnava kot kromosom/raztopina/posameznik
Rezultat telesne pripravljenosti je število znakov, ki se razlikujejo od znakov v ciljnem nizu pri določenem indeksu. Tako imajo posamezniki z nižjo kondicijsko vrednostjo večjo prednost.
C++
// C++ program to create target string, starting from> // random string using Genetic Algorithm> > #include> using> namespace> std;> > // Number of individuals in each generation> #define POPULATION_SIZE 100> > // Valid Genes> const> string GENES =>'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOP'>> 'QRSTUVWXYZ 1234567890, .-;:_!'#%&/()=?@${[]}'>;> > // Target string to be generated> const> string TARGET =>'I love techcodeview.com'>;> > // Function to generate random numbers in given range> int> random_num(>int> start,>int> end)> {> >int> range = (end-start)+1;> >int> random_int = start+(>rand>()%range);> >return> random_int;> }> > // Create random genes for mutation> char> mutated_genes()> {> >int> len = GENES.size();> >int> r = random_num(0, len-1);> >return> GENES[r];> }> > // create chromosome or string of genes> string create_gnome()> {> >int> len = TARGET.size();> >string gnome =>''>;> >for>(>int> i = 0;i gnome += mutated_genes(); return gnome; } // Class representing individual in population class Individual { public: string chromosome; int fitness; Individual(string chromosome); Individual mate(Individual parent2); int cal_fitness(); }; Individual::Individual(string chromosome) { this->kromosom = kromosom; fitnes = cal_fitness(); }; // Izvedite parjenje in ustvarite nove potomce Individual Individual::mate(Individual par2) { // kromosom za potomce string child_chromosome = ''; int len = chromosome.size(); for(int i = 0;i { // naključna verjetnost float p = random_num(0, 100)/100; // če je verjetnost manjša od 0,45, vstavi gen // iz nadrejenega 1 if(p<0.45) child_chromosome += chromosome[i]; // if prob is between 0.45 and 0.90, insert // gene from parent 2 else if(p <0.90) child_chromosome += par2.chromosome[i]; // otherwise insert random gene(mutate), // for maintaining diversity else child_chromosome += mutated_genes(); } // create new Individual(offspring) using // generated chromosome for offspring return Individual(child_chromosome); }; // Calculate fitness score, it is the number of // characters in string which differ from target // string. int Individual::cal_fitness() { int len = TARGET.size(); int fitness = 0; for(int i = 0;i { if(chromosome[i] != TARGET[i]) fitness++; } return fitness; }; // Overloading bool operator<(const Individual &ind1, const Individual &ind2) { return ind1.fitness } // Driver code int main() { srand((unsigned)(time(0))); // current generation int generation = 0; vector population; bool found = false; // create initial population for(int i = 0;i { string gnome = create_gnome(); population.push_back(Individual(gnome)); } while(! found) { // sort the population in increasing order of fitness score sort(population.begin(), population.end()); // if the individual having lowest fitness score ie. // 0 then we know that we have reached to the target // and break the loop if(population[0].fitness <= 0) { found = true; break; } // Otherwise generate new offsprings for new generation vector new_generation; // Perform Elitism, that mean 10% of fittest population // goes to the next generation int s = (10*POPULATION_SIZE)/100; for(int i = 0;i new_generation.push_back(population[i]); // From 50% of fittest population, Individuals // will mate to produce offspring s = (90*POPULATION_SIZE)/100; for(int i = 0;i { int len = population.size(); int r = random_num(0, 50); Individual parent1 = population[r]; r = random_num(0, 50); Individual parent2 = population[r]; Individual offspring = parent1.mate(parent2); new_generation.push_back(offspring); } population = new_generation; cout<< 'Generation: ' << generation << ' '; cout<< 'String: '<< population[0].chromosome <<' '; cout<< 'Fitness: '<< population[0].fitness << '
'; generation++; } cout<< 'Generation: ' << generation << ' '; cout<< 'String: '<< population[0].chromosome <<' '; cout<< 'Fitness: '<< population[0].fitness << '
'; }> |
>
>
Java
bash razdeli niz z ločilom
kaj dela ravel v pythonu
import> java.util.ArrayList;> import> java.util.Collections;> import> java.util.List;> import> java.util.Random;> > public> class> GeneticAlgorithm {> >// Number of individuals in each generation> >private> static> final> int> POPULATION_SIZE =>100>;> > >// Valid Genes> >private> static> final> String GENES =>'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ 1234567890, .-;:_!'#%&/()=?@${[]}'>;> > >// Target string to be generated> >private> static> final> String TARGET =>'I love techcodeview.com'>;> > >// Function to generate random numbers in given range> >private> static> int> randomNum(>int> start,>int> end) {> >Random rand =>new> Random();> >return> rand.nextInt(end - start +>1>) + start;> >}> > >// Create random genes for mutation> >private> static> char> mutatedGenes() {> >int> len = GENES.length();> >int> r = randomNum(>0>, len ->1>);> >return> GENES.charAt(r);> >}> > >// Create chromosome or string of genes> >private> static> String createGnome() {> >int> len = TARGET.length();> >StringBuilder gnome =>new> StringBuilder();> >for> (>int> i =>0>; i gnome.append(mutatedGenes()); return gnome.toString(); } // Class representing individual in population private static class Individual implements Comparable { String chromosome; int fitness; Individual(String chromosome) { this.chromosome = chromosome; fitness = calFitness(); } // Perform mating and produce new offspring Individual mate(Individual par2) { StringBuilder childChromosome = new StringBuilder(); int len = chromosome.length(); for (int i = 0; i // random probability float p = randomNum(0, 100) / 100f; // if prob is less than 0.45, insert gene from parent 1 if (p <0.45) childChromosome.append(chromosome.charAt(i)); // if prob is between 0.45 and 0.90, insert gene from parent 2 else if (p <0.90) childChromosome.append(par2.chromosome.charAt(i)); // otherwise insert random gene(mutate), for maintaining diversity else childChromosome.append(mutatedGenes()); } // create new Individual(offspring) using generated chromosome for offspring return new Individual(childChromosome.toString()); } // Calculate fitness score, it is the number of characters in string which differ from target string private int calFitness() { int len = TARGET.length(); int fitness = 0; for (int i = 0; i if (chromosome.charAt(i) != TARGET.charAt(i)) fitness++; } return fitness; } @Override public int compareTo(Individual o) { return Integer.compare(this.fitness, o.fitness); } } // Driver code public static void main(String[] args) { // current generation int generation = 0; List population = new ArrayList(); boolean found = false; // create initial population for (int i = 0; i String gnome = createGnome(); population.add(new Individual(gnome)); } while (!found) { // sort the population in increasing order of fitness score Collections.sort(population); // if the individual having lowest fitness score i.e. 0 then we know that we have reached to the target // and break the loop if (population.get(0).fitness <= 0) { found = true; break; } // Otherwise generate new offsprings for new generation List newGeneration = new ArrayList(); // Perform Elitism, that mean 10% of fittest population goes to the next generation int s = (10 * POPULATION_SIZE) / 100; for (int i = 0; i newGeneration.add(population.get(i)); // From 50% of fittest population, Individuals will mate to produce offspring s = (90 * POPULATION_SIZE) / 100; for (int i = 0; i int len = population.size(); int r = randomNum(0, 50); Individual parent1 = population.get(r); r = randomNum(0, 50); Individual parent2 = population.get(r); Individual offspring = parent1.mate(parent2); newGeneration.add(offspring); } population = newGeneration; System.out.print('Generation: ' + generation + ' '); System.out.print('String: ' + population.get(0).chromosome + ' '); System.out.println('Fitness: ' + population.get(0).fitness); generation++; } System.out.print('Generation: ' + generation + ' '); System.out.print('String: ' + population.get(0).chromosome + ' '); System.out.println('Fitness: ' + population.get(0).fitness); } }> |
>
>
Python3
# Python3 program to create target string, starting from> # random string using Genetic Algorithm> > import> random> > # Number of individuals in each generation> POPULATION_SIZE>=> 100> > # Valid genes> GENES>=> '''abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOP> QRSTUVWXYZ 1234567890, .-;:_!'#%&/()=?@${[]}'''> > # Target string to be generated> TARGET>=> 'I love techcodeview.com'> > class> Individual(>object>):> >'''> >Class representing individual in population> >'''> >def> __init__(>self>, chromosome):> >self>.chromosome>=> chromosome> >self>.fitness>=> self>.cal_fitness()> > >@classmethod> >def> mutated_genes(>self>):> >'''> >create random genes for mutation> >'''> >global> GENES> >gene>=> random.choice(GENES)> >return> gene> > >@classmethod> >def> create_gnome(>self>):> >'''> >create chromosome or string of genes> >'''> >global> TARGET> >gnome_len>=> len>(TARGET)> >return> [>self>.mutated_genes()>for> _>in> range>(gnome_len)]> > >def> mate(>self>, par2):> >'''> >Perform mating and produce new offspring> >'''> > ># chromosome for offspring> >child_chromosome>=> []> >for> gp1, gp2>in> zip>(>self>.chromosome, par2.chromosome):> > ># random probability> >prob>=> random.random()> > ># if prob is less than 0.45, insert gene> ># from parent 1> >if> prob <>0.45>:> >child_chromosome.append(gp1)> > ># if prob is between 0.45 and 0.90, insert> ># gene from parent 2> >elif> prob <>0.90>:> >child_chromosome.append(gp2)> > ># otherwise insert random gene(mutate),> ># for maintaining diversity> >else>:> >child_chromosome.append(>self>.mutated_genes())> > ># create new Individual(offspring) using> ># generated chromosome for offspring> >return> Individual(child_chromosome)> > >def> cal_fitness(>self>):> >'''> >Calculate fitness score, it is the number of> >characters in string which differ from target> >string.> >'''> >global> TARGET> >fitness>=> 0> >for> gs, gt>in> zip>(>self>.chromosome, TARGET):> >if> gs !>=> gt: fitness>+>=> 1> >return> fitness> > # Driver code> def> main():> >global> POPULATION_SIZE> > >#current generation> >generation>=> 1> > >found>=> False> >population>=> []> > ># create initial population> >for> _>in> range>(POPULATION_SIZE):> >gnome>=> Individual.create_gnome()> >population.append(Individual(gnome))> > >while> not> found:> > ># sort the population in increasing order of fitness score> >population>=> sorted>(population, key>=> lambda> x:x.fitness)> > ># if the individual having lowest fitness score ie.> ># 0 then we know that we have reached to the target> ># and break the loop> >if> population[>0>].fitness <>=> 0>:> >found>=> True> >break> > ># Otherwise generate new offsprings for new generation> >new_generation>=> []> > ># Perform Elitism, that mean 10% of fittest population> ># goes to the next generation> >s>=> int>((>10>*>POPULATION_SIZE)>/>100>)> >new_generation.extend(population[:s])> > ># From 50% of fittest population, Individuals> ># will mate to produce offspring> >s>=> int>((>90>*>POPULATION_SIZE)>/>100>)> >for> _>in> range>(s):> >parent1>=> random.choice(population[:>50>])> >parent2>=> random.choice(population[:>50>])> >child>=> parent1.mate(parent2)> >new_generation.append(child)> > >population>=> new_generation> > >print>(>'Generation: {} String: {} Fitness: {}'>.> >format>(generation,> >''.join(population[>0>].chromosome),> >population[>0>].fitness))> > >generation>+>=> 1> > > >print>(>'Generation: {} String: {} Fitness: {}'>.> >format>(generation,> >''.join(population[>0>].chromosome),> >population[>0>].fitness))> > if> __name__>=>=> '__main__'>:> >main()> |
>
>
C#
binarno drevo po vrstnem redu
using> System;> using> System.Collections.Generic;> using> System.Linq;> > public> class> GeneticAlgorithm> {> >// Number of individuals in each generation> >private> const> int> POPULATION_SIZE = 100;> > >// Valid Genes> >private> const> string> GENES =>'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOP'> +> >'QRSTUVWXYZ 1234567890, .-;:_!'#%&/()=?@${[]}'>;> > >// Target string to be generated> >private> const> string> TARGET =>'I love techcodeview.com'>;> > >private> static> readonly> Random random =>new> Random();> > >// Function to generate random numbers in given range> >private> static> int> RandomNum(>int> start,>int> end)> >{> >return> random.Next(start, end + 1);> >}> > >// Create random genes for mutation> >private> static> char> MutatedGenes()> >{> >int> len = GENES.Length;> >int> r = RandomNum(0, len - 1);> >return> GENES[r];> >}> > >// Create chromosome or string of genes> >private> static> string> CreateGnome()> >{> >int> len = TARGET.Length;> >char>[] gnome =>new> char>[len];> >for> (>int> i = 0; i { gnome[i] = MutatedGenes(); } return new string(gnome); } // Class representing individual in population private class Individual { public string Chromosome { get; } public int Fitness { get; } public Individual(string chromosome) { Chromosome = chromosome; Fitness = CalculateFitness(); } // Calculate fitness score, it is the number of // characters in string which differ from target string. private int CalculateFitness() { return Chromosome.Zip(TARGET, (a, b) =>a == b? 0 : 1).Vsota(); } // Izvedite parjenje in ustvarite nove potomce public Individual Mate(Individual parent2) { char[] childChromosome = new char[Chromosome.Length]; for (int i = 0; i { double p = random.NextDouble(); if (p<0.45) childChromosome[i] = Chromosome[i]; else if (p <0.90) childChromosome[i] = parent2.Chromosome[i]; else childChromosome[i] = MutatedGenes(); } return new Individual(new string(childChromosome)); } } // Overloading private class FitnessComparer : IComparer { public int Compare(Individual ind1, Individual ind2) { return ind1.Fitness.CompareTo(ind2.Fitness); } } // Driver code public static void Main() { // current generation int generation = 0; List population = new List(); bool found = false; // create initial population for (int i = 0; i { string gnome = CreateGnome(); population.Add(new Individual(gnome)); } while (!found) { // sort the population in increasing order of fitness score population.Sort(new FitnessComparer()); // if the individual having lowest fitness score ie. // 0 then we know that we have reached the target // and break the loop if (population[0].Fitness == 0) { found = true; break; } // Otherwise generate new offsprings for new generation List newGeneration = new List(); // Perform Elitism, that means 10% of fittest population // goes to the next generation int s = (10 * POPULATION_SIZE) / 100; for (int i = 0; i newGeneration.Add(population[i]); // From 50% of fittest population, Individuals // will mate to produce offspring s = (90 * POPULATION_SIZE) / 100; for (int i = 0; i { int len = population.Count; int r = RandomNum(0, 50); Individual parent1 = population[r]; r = RandomNum(0, 50); Individual parent2 = population[r]; Individual offspring = parent1.Mate(parent2); newGeneration.Add(offspring); } population = newGeneration; Console.WriteLine('Generation: ' + generation + ' ' + 'String: ' + population[0].Chromosome + ' ' + 'Fitness: ' + population[0].Fitness); generation++; } Console.WriteLine('Generation: ' + generation + ' ' + 'String: ' + population[0].Chromosome + ' ' + 'Fitness: ' + population[0].Fitness); } }> |
>
>
Javascript
// Number of individuals in each generation> const POPULATION_SIZE = 100;> > // Valid Genes> const GENES =>'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOP'> +> >'QRSTUVWXYZ 1234567890, .-;:_!'#%&/()=?@${[]}'>;> > // Target string to be generated> const TARGET =>'I love techcodeview.com'>;> > // Function to generate random numbers in given range> function> RandomNum(start, end) {> >return> Math.floor(Math.random() * (end - start + 1)) + start;> }> > // Create random genes for mutation> function> MutatedGenes() {> >let len = GENES.length;> >let r = RandomNum(0, len - 1);> >return> GENES.charAt(r);> }> > // Create chromosome or string of genes> function> CreateGnome() {> >let len = TARGET.length;> >let gnome =>''>;> >for> (let i = 0; i gnome += MutatedGenes(); } return gnome; } // Class representing individual in population class Individual { constructor(chromosome) { this.Chromosome = chromosome; this.Fitness = this.CalculateFitness(); } // Calculate fitness score, it is the number of // characters in string which differ from target string. CalculateFitness() { let fitness = 0; for (let i = 0; i |
>
>
Izhod:
a b c številke
Generation: 1 String: tO{'-?=jH[k8=B4]Oe@} Fitness: 18 Generation: 2 String: tO{'-?=jH[k8=B4]Oe@} Fitness: 18 Generation: 3 String: .#lRWf9k_Ifslw #O$k_ Fitness: 17 Generation: 4 String: .-1Rq?9mHqk3Wo]3rek_ Fitness: 16 Generation: 5 String: .-1Rq?9mHqk3Wo]3rek_ Fitness: 16 Generation: 6 String: A#ldW) #lIkslw cVek) Fitness: 14 Generation: 7 String: A#ldW) #lIkslw cVek) Fitness: 14 Generation: 8 String: (, o x _x%Rs=, 6Peek3 Fitness: 13 . . . Generation: 29 String: I lope Geeks#o, Geeks Fitness: 3 Generation: 30 String: I loMe GeeksfoBGeeks Fitness: 2 Generation: 31 String: I love Geeksfo0Geeks Fitness: 1 Generation: 32 String: I love Geeksfo0Geeks Fitness: 1 Generation: 33 String: I love Geeksfo0Geeks Fitness: 1 Generation: 34 String: I love techcodeview.com Fitness: 0> Opomba: Vsakič se algoritem začne z naključnimi nizi, zato se izhod lahko razlikuje
Kot lahko vidimo iz izhoda, se je naš algoritem včasih zataknil pri lokalni optimalni rešitvi, kar je mogoče še izboljšati s posodobitvijo algoritma za izračun ocene telesne pripravljenosti ali s prilagajanjem operatorjev mutacije in križanja.
Zakaj uporabljati genetske algoritme
- So robustni
- Zagotovite optimizacijo v stanju velikega prostora.
- Za razliko od tradicionalne umetne inteligence se ne pokvarijo ob majhni spremembi vnosa ali prisotnosti šuma
Uporaba genetskih algoritmov
Genetski algoritmi imajo veliko aplikacij, nekatere od njih so –
- Ponavljajoča se nevronska mreža
- Testiranje mutacij
- Razbijanje kode
- Filtriranje in obdelava signalov
- Učenje baze mehkih pravil itd