Change Networks and Network Copying

[mbecker12]

Replace the manual initialization of layers by a more automated way:

The number of layers and their features (e.g. number of neurons, number of channels) should be read from a config file. Then all the parameters can be stored in an nn.Sequential object.

For this to work, we would also have to change the copying of parameters to both the offline network in double deep q learning and the actor networks: Instead of using torch native functions (such as vector_to_parameters) as it is now, we could use the deepcopy function and maybe send those copied objects through the communication pipelines.

[mbecker12]

Currently, the agent class files are rather messy as we still try to keep it somewhat configurable without using nn.Sequential. The current solution allows us to keep using the original way of sending network weights (i.e. via vector_to_parameters).

It would be beneficial for adaptability and code readability to implement this feature but it has no priority right now