Introduction to genetic algorithm

Introduction

Genetic algorithm is an artificial intelligence (AI) algorithm used to solve problems that can't be solved with a mathematic formula.

It is inspired by the theory of evolution which uses genes and chromosomes to evolve. The generations that evolve are your solutions to the problem. To understand how this algorithm works, it is good if you have some knowledge about evolution and DNA and how the 'mom' and 'dad' pass their DNA to their children.

Values

Genes: A string of binary numbers (0 and 1).
Chromosomes: A group of genes.
Population/Generation: A group of chromosomes.

To setup your values, you must first look at your problem and find what are the values required. Let's take a simple example, we want to find what equation gives us the number 30. What we need here are numbers and operators and let's say we want to have an equation with 3 numbers in it. We would setup our genes like this (note that you can give them the value you want):

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

These are our genes. To generate a chromosome, we need to choose random genes and put them together. When you generate a chromosome, you can make it completely random which could give something like "1+-+3" or make it generate valid equation from the beginning like "1+2+3". Now, we need 5 genes (3 numbers + 2 operators), so we would have a chromosome like:

00111100011110101001 <- which is 3 * 7 + 9

You can see the genes as values in the DNA, every pair of bits have their own value. We generate a population using the same function. The population can have as many chromosomes as you want. In this example, you can generate 100 chromosomes, it would be enough. To save your chromosomes you can create a string and consider every 0 and 1 as a character, or save it in an array of short (8 bits) /int (32 bits depending on your OS) which will take less memory.

Analysis

Once you have your generation, loop through every generated chromosomes and see if one of them gives you the solution you seek (in this case, 30). For every chromosome, set a fitness score which is a value that specifies how far the chromosome is to the solution. An example of a fitness score for our problem would be:

01011010011110110111 <- we will calculate the fitness score of this chromosome, which is 5 + 7 - 7
5 + 7 - 7 = 5
1/(30-5) = 0.04 <- this is the fitness score of our chromosome, in our case, the closer it is to 1, the closer it is to the solution.

If you don't find the solution, you need to generate a new generation by creating babies.

Generate New Population

There are many ways to generate a new population. We will use the one called 'Crossover Rate'. First you need to choose 2 chromosomes at random (this is where the fitness score comes handy, if you match 2 chromosomes that are near your solution, it increases the chance to find the solution) and generate a random number between 1 and the length of your chromosomes (in this case, 20). Take the 2 chromosomes and switch each parts:

1. 00111100 011110101001
2. 01011010 011110110111
---------------------------------
1. 00111100 011110110111
2. 01011010 011110101001

You can see here the random number was 8 and we switched every parts (before and after 8) and created 2 babies using their parents' DNA. Do the same until you made babies with all the chromosomes. You will now have a population of 200 chromosomes.

You can set a mutation rate for every gene, which should be at a very low level, such as 1%. If a gene mutates, you need to change it to a random gene, which can go better or worse.

Invalid Chromosomes

You can either keep them or kill them. I prefer to kill them as when they are invalid they become very useless, but you can make a function that fixes an invalid chromosome.

Conclusion

This is the very basic of the algorithm. Using this algo, you can solve many problems such as where can we fit the biggest circle in a given GUI or how many boxes can fit in a containers. It is an algo that you must program yourself to understand it well, so open your IDE and start coding!