How to train multiple agents at once with Neural Network

Help and Feedback Scripting Support
,
#1

How to train multiple agents at once with Neural Networks?

I am using Kironte’s Neural Network and i was wondering how to train multiple agents at once with Genetic Evolution.

Training a single agent at a time is very slow and time-consuming, so I need to train around 120 agents simultaneously per generation.

Here is the base code:

math.randomseed(os.clock()+os.time())

local Package = game:GetService("ReplicatedStorage").NNLibrary
local Base = require(Package.BaseRedirect)
local FeedforwardNetwork = require(Package.NeuralNetwork.FeedforwardNetwork)
local ParamEvo = require(Package.GeneticAlgorithm.ParamEvo)
local Momentum = require(Package.Optimizer.Momentum)
local clock = os.clock()

local generations = 64
local population = 10
local setting = {
	HiddenActivationName = "LeakyReLU";
	OutputActivationName = "Tanh";
}

local geneticSetting = {
	ScoreFunction = function(net)
		local score = 0
		
		return score
	end;
	PostFunction = function(geneticAlgo)
		local info = geneticAlgo:GetInfo()
		print("Generation "..info.Generation..", Best Score: "..info.BestScore)
	end;
}
local tempNet = FeedforwardNetwork.new({"x"},2,3,{"out"},setting)
geneticAlgo = ParamEvo.new(tempNet,population,geneticSetting)

geneticAlgo:ProcessGenerations(generations)

local net = geneticAlgo:GetBestNetwork()
#2

thats a question you should ask the creator of this libary if thats even a option

#3

To train multiple agents at once with Neural Networks and Genetic Evolution, you will need to make some modifications to the code.

  1. Create a population of networks: Instead of creating a single network, you will need to create a population of networks, one for each agent you want to train.

For example, you could create an array of networks like this:

javaCopy code

local population = {}
for i = 1, 120 do
    population[i] = FeedforwardNetwork.new({"x"},2,3,{"out"},setting)
end
  1. Evaluate each network: To evaluate each network, you will need to run them through the simulation and calculate a fitness score for each one. You can use the ScoreFunction in the geneticSetting table to do this.

For example:

luaCopy code

local geneticSetting = {
    ScoreFunction = function(net)
        local score = 0
        -- Run the simulation for the network and calculate the score
        return score
    };
    PostFunction = function(geneticAlgo)
        local info = geneticAlgo:GetInfo()
        print("Generation "..info.Generation..", Best Score: "..info.BestScore)
    end;
}

for i = 1, #population do
    local net = population[i]
    local score = geneticSetting.ScoreFunction(net)
    geneticAlgo:SetFitness(net, score)
end
  1. Evolve the population: Once you have evaluated each network and assigned a fitness score, you can use the ProcessGeneration() method to evolve the population.

For example:

geneticAlgo:ProcessGeneration()

This will create a new generation of networks based on the fitness scores of the previous generation. You can repeat this process for as many generations as you want.

  1. Select the best network: Once you have completed all of the generations, you can select the best network from the final population.

For example:

local bestNet
local bestScore = -math.huge
for i = 1, #population do
    local net = population[i]
    local score = geneticAlgo:GetFitness(net)
    if score > bestScore then
        bestNet = net
        bestScore = score
    end
end

This will loop through each network in the final population and select the one with the highest fitness score as the best network.

Note that this is just a basic outline of the changes you will need to make. You will likely need to make additional modifications to the code to adapt it to your specific needs.

#4

By any chance, did you use ChatGPT or any language model to say that? Yes you did. By looking at your other posts, many people accuse you of using a language model called ChatGPT, which is simply obvious.