Questions about Neural Network Library?

Help and Feedback Scripting Support
,
#1

Okay, so I have a lot of questions about this library again.
So, there are a few questions that I want to ask,

  1. what are the parameters for ParamEvo.new
    I didn’t really get any examples for it… They just said this:
    NeuralNetwork neuralNetworkTemp , number popSize , dictionary geneticSettings
    which im not gonna lie, i dont really get it… Whats neuralNetworkTemp? sorry dont know much, is it a template?? i guess i know what a popSize is and what a geneticSettings are…
  1. FeedforwardNetwork.new
    yeah okay, i just dont generally get this… is the first one sending the information? next is node number? and third is output? and last is settings?

yeah okay i dont get this… Any help? I would appreciate it… Thanks!

#2

Well, neuralNetworkTemp(As the name implies Temperature) is a hyperparameter of LSTMs (and neural networks generally) used to control the randomness of predictions by scaling the logits before applying softmax

You can see more about feedforwoard NN here: https://towardsdatascience.com/deep-learning-feedforward-neural-network-26a6705dbdc7

If I were you, I would start with more simple examples to understand what is going on. self learning zombie - Roblox

1 Like
#3

Alright… thanks so much i really appreciate it!! :smiley:

#4

You should visit the library’s website as well as check out some sites on how machine learning works, like the one inpad_2 linked.

The API is here. neuralNetworkTemp is the template network that describes what the entire population’s structure should look like, popSize is the number of networks in the population, and geneticSettings is a table containing the configuration; it’s structure and default values are noted in the API.

The API is also here. inputNamesArray is an array of names that the input nodes should have, numberOfLayers is the number of hidden layers, numberOfNodesPerLayer is the number of nodes per hidden layer, and customSettings contains the configuration for the network.

2 Likes
#5

I have one question though, is it possible to apply wights and bias to the created nodes in FeedForwardNetwork.new?

Note: have been looking at the api documentation and i found a SetBias function and synapse one…
Thanks for the help, i really appreciate u and @inpad_2 for helping me and taking the time to!!
(Ill prob read more about the library on the documentation)

One question tho, if i wanted to… i could set the weight and bias’s output to moving the ai? This one is from jabril’s video @ 3:20… pretty much any movement is a certain number, for example; left will be 0 and right will be 1 and slightly left will be 0.5 and you get the point, and @ 6:30 he explains how it will work… from my understanding, he calculated the output of where to go from going on every node and synapse and calculating the math of biases and weights and from that number it will move the ai.
But my main question is, how would i let the ai configure the weights and biases? It might be kinda complicated so thats why i am asking for help… Thanks!!

#6

To edit the weights and biases automatically, you should use the BackPropagator class. It allows you to backward propagate a network so its parameters adapt to the specific problem it is assigned.

The API isn’t great and my OOP implementation isn’t the best, but it was a year ago so what can you do.

2 Likes