A concept I came up which ditches the idea of "layers" in a neural network.

Copy Dynet.py to your project.

Install matplotlib with pip install matplotlib to run the example in main.py.

Classic neural networks use layers as a way of organizing neurons. "Dynet" uses a single layers to process inputs and outputs where neurons can directly connect to outputs or pass through multiple neurons and even connect to themselves

To create a Dynet object simple use myNet = Dynet(numInputs, numOutputs, numHiddens). Dynet objects have to have at least 1 hidden neuron. To use the Dynet object, use myNet.feedForward(inputArr) which will return an array of outputs generated by the neural network. To mutate the network use myNet.mutate(mutationRate, amountOfMutations, modifyHiddens). modifyHiddens is an optional parameter which will mutate hidden nodes if set to True (True by default). myNet.printNetwork() will print the current structure of the network and myNet.copy() will return a copy of the network

Here is an example that uses "The Coding Train's" flappy bird project altered to use Dynet instead https://editor.p5js.org/akgaming1322/full/g4ipfcCZb