Agreed, the code is not meant to be a good solution.

This was actually my starting point for an implementation of an "evolution" based algorithm that I wanted to experiment with. I was surprised that completely random models were able to optimally solve the Cartpole environment so I thought I'd share :). Of course I don't expect this to be the case for more complex environments.