In videos like this, it talks about how you need to find the right fit for your data which is shown on a scatter plot. I understand how this works when you have a dataset for something but how does it work when you are trying to train a DQN to play snake (eating apples and getting longer game). I have been struggling to tune my hyperparameters as well as figure out how many hidden neurons and hidden layers I need. I have found that right now 256 hidden neurons and 2 hidden layers works best. Please tell me if this topic has flown completely over my head and I am missing something. Thank you!
I have tried using less hidden neurons (8-32) but the results are always very poor. I have 11 input nodes and 3 output.
