Introduction

This is a very simple library for Genetic Algorithm which uses Tree-Based Genetic programming to evolve some solutions to find answers or improve them. its writen in Python with the help of GraphViz library for rendering the result as a graph(tree).

NOTE : this is just one of my personal projects and first expirence with python, so be cool if you found a bug ;-).

and also feel free to fork it and tweak it ...

---

Features

- Independent Fitness Function (FF) :

one of the key features is that you can implement your own FitnessFunction (or FF) completely independent to GP system. each Individual has a variable called fitness which you can overwrite it in your FF and let the GP system handles the rest of the job.

NOTE: in this library, the Higher fitness is Better. so be careful about fitness calculation in your FF

- HyperParameter(HP) and Online Tuning :

with the use of HyperParameters you can control the decision flow of GP system. for an example you can tune the range of selection both in cross-over and mutation while your GP is running

NOTE: although there's a few of these HPs, but in the future there will be more

---

Dependencies

the only dependency is GraphViz library which you can go to their website and see the installation manual. BUT , for the sake of simplicity, here is how you can install it on Linux:

first install GraphViz compiler through apt : sudo apt install graphviz

and then through python pip : pip install pydot graphviz

now you're ready to go. -_*

---

Examples

Here you can find an example of how this works. if you're familiar with genetic programming and boolean operations, the code is simple as a NOT gate for you :-)

in this example i tried to make a XOR gate , using Only 4 NAND gate... so here is the 18'th Generation of GP :

There are some notations in graphs which are :

• G : is the current Generation number
• F : is the Fitness Value
• D : Depth of the Tree
• W : Width of the Tree
• n : Number of Nodes(both functions and terminals) in tree

---

What's Next???

well, the next step is to Optimize this and make it cleaner, also im thinking adding concurrency to make it work parallel. I might also solve some other famous problems with this and compare the performance.

But the main goal is to implement my own ideas and thoughts and test them.