NEAT (NeuroEvolution of Augmenting Topologies) implemented in Go

https://github.com/jinyeom/neat/blob/7aa4ca49072822816508b2aa18ddf04b33b7e1ae/genome.go#L297

If the function always returns false, it is infinitely called, leading to a runtime: goroutine stack exceeds 1000000000-byte limit error and a stack overflow

Hello ! First of all, thanks for this amazing library.

I'm trying to get into machine learning, and I saw on youtube a guy that used the NEAT algorythm ( I don't know if it's called like that ^^' ), so I'm wanted to create a sort of AI-only pong. Your library seemt the most complete, but I wasn't able to go get-it.

I'm a complete noob when it comes to machine learning, and I'm still learning golang, but today, I'm glad I think I've fixed the many bugs that made this library not compilable and made the test to not pass.

I don't know if I removed some usefull changes ( did I just reverted the library to an old state ? I don't really know ). I've put some // +build ignore to completely ignore files, so I don't know.

There are comments explaining changes. If they're not clear, just tell me, and I'll explain what was the problem.

Also, tell me if you want me to remove these ugly github/@jesuiscamille: comments :) !

Have a good day !

PS: I don't know what the normal go test results should be. I'll do a simple XOR manual test, and write here if nothing got wrong 😄

Hi! First i wanted to thank you for the effort you put into this project. I'm having issues evaluating even the example you posted (XOR). So i trained the network until the avg Fitness was around 1.5 and the best Fitness = 0 (which should solve all possible xor cases). But when i create a Network using the best genome and run some data through it, it doesn't seem to work. Am i doing something wrong or is there some kind of a bug regarding retrieving the best individual? Here's the code i use for the test:

package main

import (
	neat "github.com/jinyeom/neat"
	"log"
	"math"

)

func main(){
	// First, create a new instance of Config from the JSON file created above.
	// If there's a file import error, the program will crash.
	config, err := neat.NewConfigJSON("config.json")
	if err != nil{
		log.Fatal(err)
	}
	
	neatInst := neat.New(config, neat.XORTest())
	neatInst.Run()
	log.Printf("Fitness: %f\n", neatInst.Best.Fitness)
	nn := neat.NewNeuralNetwork(neatInst.Best)
	
	inputs := make([]float64, 3)
	inputs[0] = 1.0 // bias	
	b := []bool{true,false}
	for i := 0; i < len(b); i++{
		for j := 0; j < len(b); j++{
			a := b[i]
			b := b[j]
			if a{inputs[1] = 1.0}else{inputs[1] = 0.0}
			if b{inputs[2] = 1.0}else{inputs[2] = 0.0}
			output, _ := nn.FeedForward(inputs)
			log.Printf("%t, %t = %f\n",a,b,output[0])
		}
	}
}

The output of this is in most cases:

2017/07/01 03:09:31 Fitness: 0.000000 2017/07/01 03:09:31 true, true = 0.000000 2017/07/01 03:09:31 true, false = 0.000000 2017/07/01 03:09:31 false, true = 0.000000 2017/07/01 03:09:31 false, false = 0.000000

But sometimes it also goes for all of it: 2017/07/01 03:22:18 Fitness: 0.000000 2017/07/01 03:22:18 true, true = 1.000000 2017/07/01 03:22:18 true, false = 1.000000 2017/07/01 03:22:18 false, true = 1.000000 2017/07/01 03:22:18 false, false = 1.000000

I'd appreciate your feedback on this issue. Chances are i'm simply too dumb/tired to use it :)

Is there a way to save and load prior trained models?

Forgive my ignorance as I am new but would it be as simple as serializing/deserializing the neat instance?