I think this right here warrants a blog post in itself! Maybe we can get you to use Comet ML to talk about how to examine and tune hyper parameters so that we prevent overfitting