Trying to figure out ways to check that the neural network is working is driving me batty. The neural network visualizer is working but it only helps to a degree.
Inputs to the neural network keep changing.
There are too many neurons and synapses, and synapses that loop back. I'm thinking about creating smaller neural networks or at least neural networks with fixed values to test the result of a few activation and learning steps.
Creating a test neural network is hard because I'll have to create another genome.
There seems to be too many things to test, too many possible combinations. It all feels too complex and I feel lost.
Okay, after my panic attack, I've settled on a plan. I'm going to first test the full neural network instead of a simpler one because I want to make sure that the full one works.
I'll be testing neural network structure, activation, and learning. With structure, I'll be making sure that all the neurons and synapses are in the right places. With activation, I'll be testing neuron activations after a few neural network activations. With learning, I'll be testing synapse weights after a few neural network learning cycles.
I've been spending most of my day setting this up and will continue to do so tonight.
There are too many possible synapse connections so I'm going to have to be selective.