Neural Network with Genetic Algorithms Experiment (food and water) with online demo!
An experiment using Neural Networks and genetic algorithms I have always found you understand something better if you build it yourself.
Since creating this I've created a new demo with new C++ code to explore some different situations. I recommend watching the video at https://youtu.be/bq3FdlUeOTU I've also updated this demo to use more modern javascript. (ES6) I also fixed a long standing issue with the way they wrap-around the edges of the screen.
My aim was to make a "lifeform" that learnt basic needs. The lifeform needs water and food to survive. Water is always available, food is scattered and once eaten re-spawns elsewhere. If the "lifeform" runs out of food or water then it is classed as "dead"
See it running at https://youtu.be/yZUhprqGVTE Live demo at http://experiments.robsmithdev.co.uk/NeuralNetworkGA1/
There are 30 "lifeforms" in each generation, and each generation runs for a maximum of 6000 iterations. The project is setup with a Neural Network with 8 inputs, two hidden layers and 2 outputs.
With the current configuration and topology, by the 950th generation 2/3 of the lifeforms survive.
I have included data for the network after I left it running over night (it reached 17120 generations)
I found the following resources useful and inspirational while creating this:
MIT AI Courseware - Worth a watch: https://www.youtube.com/playlist?list=PLUl4u3cNGP63gFHB6xb-kVBiQHYe_4hSi
Copyright © RobSmithDev 2021-2022 https://robsmithdev.co.uk