Neural Network - Genetic Algorithm - C++ Implementation

(work in progress)

This project is still a work in progress and pretends to create an implementation of an Engine capable of genetically evolve AI agents that contain a feedforward neural network.

All of project main structures are independent and self contained, meaning that the neural network and genetic algorithm systems are not tightly coupled.

Directory Structure

  ├── bin             - executable bin folder
  ├── build           - default ai lib build folder
  ├── src             - ai lib source folder
  ├── test            - ai lib tests folder
      └── memory      - memory source code
  └── sample          - samples source folder
      ├── xor         - xor agent to solve XOR function
      └── tictactoe   - tic tac toe agent example source code

How to build the library

  nn-ga$ make
  make -C src/ objects
  make[1]: Entering directory 'nn-ga/src'
  g++ -Wall -c Genome.cpp Neuron.cpp GeneticAlgorithm.cpp NeuralNetwork.cpp Engine.cpp -O3
  ar rcs ../build/libai.a *.o
  make[1]: Leaving directory 'nn-ga/src'

The lib file (libai.a) was built to the build/ directory and can be linked from now on with any of your projects. In the test/ folder several examples can be found on how to link the lib, and how to use it on your own agents.

Core Source Structures

Neural Network

Genetic Algorithm

Engine

Proof of Concept

The XOR example implements a a XOR agent using a feedforward neural network to derive the XOR function. After a couple million iterations it converges to the desired output:

Current output: 2152738
1, 0 => 0.93
0, 1 => 0.93
1, 1 => 0.07
0, 0 => 0.07

The implementation is suboptimal since it is not using backpropagation yet.