Genetic Algorithms

Homework 2

Ciobotaru Mihai , 2B3

December 7, 2021

Abstract
The purpose of this study is to prove the efficiency of genetic algorithms in solving multi-modal
functions. In order to conduct this research, I compared the results I gained with the results of
other heuristic search algorithms. Doing that, I have found genetic algorithms are a very useful
tool in solving multi-modal functions in cases where the search space is too large for deterministic
algorithms to solve.

1 Introduction
This paper‘s purpose is to present the behaviour and efficiency of a basic genetic algorithm(GA) and
compare it with other heuristic search algorithms: Steepest-Ascent Hill Climbing(SAHC), Nearest-
Ascent Hill Climbing(NAHC) and Simulated Annealing(SA). In the following pages I will explain what
a genetic algorithm is in detail and prove it‘s power through an experiment . This experiment involves
search for the global minima of four different multi-modal functions with the algorithms mentioned
earlier. With the help of the results, we will make a comparison between them and draw conclusions.

2 Genetic Algorithm
2.1 Definition and explanation
Genetic algorithms are a family of algorithms inspired by evolution. These algorithms encode
a potential solution to a specific problem on a simple chromosome-like data structure and apply
recombination operators to these structures in order to reach the best possible solution it can find.

An implementation of a genetic algorithm begins with a population of chromosomes and after that, for
multiple iterations called generations, there are used reproductive methods (genetic operators) in such
a way that the chromosomes which represent a better solution for the target problem are given more
chances to ’reproduce’ than those chromosomes which represent poorer solutions. The ’goodness’ of a
solution is typically defined with respect to the current population.

It is a good idea to use a genetic algorithm when the search space is so large that using an exhaustive
search would be impossible. More than that, genetic algorithms do not provide an exact global
maxima/minima but they offer a good enough solution in a good amount of time. Other heuristic
search algorithms such as Hill Climbing may be better to use if the function to be tested is uni-modal.

In nature, the individual that has better survival traits will survive for a longer.This in turn provides
it a better chance to produce offspring with its genetic material. In time, the population will consist
with lot of genes from the superior individuals and less from the inferior individuals. The existence of
competition among individuals of a population determine them to get fitter.

1
The concept described above is also present in genetic algorithms.As a consequence, mostly only the
fittest solutions get passesd to the next generation. By random chance, a small number of unfit
solutions are also picked in order to prevent getting stuck in a local accumulation point.A very small
number of unfit solutions are closer to the global maxima/minima that a fit solution. There are many
selection functions to choose from, but the fitness function(used to evaluate a solution) is dependent
on the problem.

2.2 Genetic operators

Crossover is a genetic operator in which the parents exchange information(genetic material) at a
random position in the chromosome. In some genetic algorithms the offspring are added to the new
population and in other algorithms they replace their parents.

Mutation is an asexual recombination operator(genetic operator) in which the information of a chro-

mosome is slightly modified in order to maintain the diversity of the population. The mutation operator
ensures that the algorithms doesn‘t converge in a local maxima/minima and gives the random chance
to find a better result.

3 Methods used
3.1 Algorithm

Algorithm 1 Genetic Algorithm

Require: generationsN um > 0 and i = 0
1: population ← randomP opulation()
2: while i < generationsN um do
3: mutateP opulation(population)
4: crossoverP opulation(population)
5: mutateP opulation(population)
6: evaluateP opulation(population)
7: selectP opulation(population)
8: end while
9: return population.bestSol()

3.2 Mutation
The mutation function iterates through every bit of every solution in the current population and has
a random probability of 1% to invert that bit. In case a chromosome‘s information is changed, the
parent remains in the population and offspring too. Applying this algorithm once before the crossover
operator and once after it, allows us to have chromosomes untainted, chromosomes with only 1% of
information changed due to mutation and chromosomes with 2% changed due to mutation.

3.3 Crossover
The crossover operator picks randomly three pairs of positions in chromosomes. For each pair we will
swap the bits from the starting position to the end position. The offspring resulted in the process
replace the parents in the population.Also we will randomly pick only 60% of the population to
crossover.

3.4 Fitness function

f itness(candidate) = 1 − | candidate.evalV alue()−min
(max−min)+ϵ |

Max variable is the maximum evaluation value from the population and min is the minimum.

2
The result value will be in the interval [0,1].

3.5 Selection
The selection function sorts all the solutions from the best to the worst and then iterates n times
through a loop (n being the population size). For each iteration we pick a candidate from the population
and add it it to the new one. In order to pick a solution we generate a random number from 0 to 98
and compare it to a certain value and ,if it is lower than that value, the solution is picked. That value
is equal with the fitness of the candidate ( a number from 0 to 1) multiplied by the acceptr ate variable
that is equal with 0,70. Therefore, every solution has a maximum probability of 70% to be picked.

4 Functions to be tested
4.1 De Jong‘s function .
The simplest test function is De Jong’s function 1. It is also known as sphere model. It is continuos,
convex and unimodal.
n
f : [−5, 12, 5, 12]n ⇒ R, f (x) = x2i
P
i=1

4.2 Schwefel’s function .

Schwefel’s function [Sch81] is deceptive in that the global minimum is geometrically distant, over the
parameter space, from the next best local minima. Therefore, the search algorithms are potentially
prone to convergence in the wrong direction.
n p
f : [−500, 500]n ⇒ R, f (x) = −xi · sin( |xi | )
P
i=1

3
4.3 Rastrigin’s function .
Rastrigin’s function is based on function 1 with the addition of cosine modulation to produce many
local minima. Thus, the test function is highly multimodal. However, the location of the minima are
regularly distributed.
n
f : [−5, 12, 5, 12]n ⇒ R, f (x) = 10·n+ (x2i − 10 · (2 · π · xi ))
P
i=1

4.4 Michalewicz’s function .

The Michalewicz function [Mic92] is a multimodal test function (n! local optima). The parameter m
defines the ”steepness” of the valleys or edges. Larger m leads to more difficult search. For very large
m the function behaves like a needle in the haystack (the function values for points in the space

4
outside the narrow peaks give very little information on the location of the global optimum).
n
i·x2
f : [0, π]n ⇒ R, f (x) = − sin(xi ) · (sin( π i ))2·m
P
m = 10
i=1

5 Experiment
For this experiment we will be comparing the time and value result‘s of a Genetic Algorithm, Hill
Climbing algorithms and Simulated Annealing in relation with the functions presented above.Each
algorithm will target to find the global minimum of the function and test it for 5th, 10th and 30th
dimension each with a precision of 5 decimals.We are interested in finding the best,worst and average
time and values and inserting them in tables to have a good visual representation of the results . The
algorithms depicted will be implemented in C++ ,compiled with g++ and ran on Ubuntu 18.04.6
LTS OS with Intel® Core™ i5-8265U CPU @ 1.60GHz × 8 processor .

Hill Climbing and Simulated Annealing algorithms go through 500 iterations and we will have a
sample of 400 results. For the genetic algorithm we will have a population of 450 solution before the
genetic operators with 450 generations and a sample of 50 results.

5.1 De Jong‘s 1 function .

Dimensions Algorithm Average Worst Best Avg. Time(s)
GA 0.00000 0.00000 0.00000 22.58351
NAHC 0.00000 0.00001 0.00000 0.50772
5
SAHC 0.00000 0.00001 0.00000 0.52688
SA 0.00001 0.00003 0.00000 0.53893
GA 0.00000 0.00000 0.00000 28.79274
NAHC 0.00001 0.00001 0.00000 1.76225
10
SAHC 0.00001 0.00001 0.00000 1.85105
SA 0.00001 0.00001 0.00000 1.88168
GA 0.00010 0.00006 0.00003 63.60260
NAHC 0.00002 0.00002 0.00001 12.41868
30
SAHC 0.00002 0.00002 0.00001 15.14539
SA 0.00002 0.00002 0.00001 15.23265

5
5.2 Schwefel’s function .
Dimensions Algorithm Average Worst Best Avg. Time(s)
GA -2094.61625 -2094.39409 -2094.91367 26.23100
NAHC -2090.77954 -1701.52869 -2094.91444 1.04017
5
SAHC -2091.86359 1701.61822 -2094.91444 1.08996
SA -2084.94706 -1701.49871 -2094.91443 1.10650
GA -4189.27586 -4188.89152 -4189.61848 38.08962
NAHC -4178.62142 -3796.01101 -4189.82880 3.61197
10
SAHC -4182.73306 -3796.08427 -4189.82887 4.00180
SA -4167.90092 -3796.08427 -4189.82633 4.02178
GA -12394.98571 -12118.25510 -12567.87570 61.84964
NAHC -12242.16245 -10432.01064 -12568.95746 8.72211
30
SAHC -12551.84651 -12175.45655 -12569.48148 17.91231
SA -11957.89946 -12491.76075 -12569.37348 16.52397

5.3 Rastrigin’s function .

Dimensions Algorithm Average Worst Best Avg. Time(s)
GA 0.04967 1.23934 0.00010 20.64697
NAHC 6.49286 21.01186 0.00025 0.52538
5
SAHC 1.43719 11.13935 0.00010 0.53120
SA 3.04881 16.08704 0.00010 0.54977
GA 1.78470 4.95716 0.00019 28.75531
NAHC 12.45414 27.20517 1.24021 1.82583
10
SAHC 2.79471 22.23488 0.00035 1.89475
SA 6.47576 29.62160 0.00081 1.95420
GA 20.67289 51.11093 4.95925 59.64764
NAHC 70.49311 21.01064 1.214796 9.10564
30
SAHC 31.18246 9.90016 0.00244 9.19241
SA 28.43003 10.28954 0.00306 9.18831

5.4 Michalewicz’s function .

Dimensions Algorithm Average Worst Best Avg. Time(s)
GA -4.64707 -4.51579 -4.68765 21.11427
NAHC -4.14842 -3.00788 -4.68752 0.53119
5
SAHC -4.12068 -2.53429 -4.68764 0.54422
SA -4.11096 -2.03949 -4.68757 0.54804
GA -9.38963 -9.05056 -9.62041 29.61517
NAHC -8.17759 -6.38718 -9.51387 1.88050
10
SAHC -8.29387 -6.01194 -9.59792 1.98624
SA -8.29976 -5.66804 -9.55632 1.99319
GA -27.35190 -25.75154 -28.37507 61.84964
NAHC -24.20723 -20.87499 -26.85518 7.67801
30
SAHC -25.01129 -22.67337 -27.47162 9.19241
SA -24.93673 -22.32254 -27.20926 9.18831

6 Comparison
The results shown above depict the efficiency of each algorithm in finding the global minima. Based
on the results we can conclude how efficient genetic algorithms are. As shown above the results of the
genetic algorithms are more concentrated, meaning that the interval in which the results are is
smaller compared to the ones from the other heuristic search algorithms. We can observe that, in
most cases, the genetic algorithm proved to return better results but in more time. More than that,
we can see that for bigger dimensions, the efficiency of the algorithm may give poorer results that
being a consequence of not enough optimisation of the algorithm.

6
7 Conclusion
This study proved the efficiency of genetic algorithms in relation with finding the global minima.
The report succeeded to show that these algorithms are slightly better than other heuristic search
algorithms for this problem in particular. As a final remark, with the help of this research, we
observed genetic algorithms‘s importance for solving difficult computing problems.

References
[1] Mathew, Tom V. ”Genetic algorithm.” Report submitted at IIT Bombay (2012). |
http://datajobstest.com/data-science-repo/Genetic-Algorithm-Guide-[Tom-Mathew].pdf
[2] Wang SC. (2003) Genetic Algorithm. In: Interdisciplinary Computing in Java Programming. The
Springer International Series in Engineering and Computer Science, vol 743. Springer, Boston,
MA |
https://link.springer.com/chapter/10.1007/978-1-4615-0377-4 6f
[3] Wikipedia - Hill Climbing |
https://en.wikipedia.org/wiki/Hill climbing
[4] Wikipedia - Simulated annealing |
https://en.wikipedia.org/wiki/Simulated annealing
[5] Wikipedia - Genetic algorithm |
https://en.wikipedia.org/wiki/Genetic algorithm
[6] Eugen Croitoru: Teaching: Genetic Algorithms |
https://profs.info.uaic.ro/∼eugennc/teaching/ga/
[7] GEATbx - Example Functions |
http://www.geatbx.com/docu/fcnindex-01.html