Save and load agents

[jbirkesteen]

We should be able to save and load agents.
Consider which parts of the object should be saved, as mentioned in the Parameter loading section of https://github.com/RasmusBrostroem/ConnectFourRL/issues/42#issuecomment-1310874145.

[jbirkesteen]

I unfortunately lost my progress due to an unexpected shutdown of my computer.
I will add a comment on the chosen approach one of the coming days, and have a look at implementing it before next time we meet.

[jbirkesteen]

We want to save

• an agent's learned parameters in a minimal manner, ideally compatible with any type of agent having the same network architecture

• metadata related to the parameters, for instance

  • name of the agent class

  • the procedure used for training the agent, e.g. as the hash of the current git hash and the name of the script training the agent

  • some (or all) of the data being logged in #65, most importantly an identifier of the connected neptune run

  • number of epochs, generation, etc.? Is this stored directly in Neptune?

@RasmusBrostroem, do you have anything to add?

[jbirkesteen]

Merged into main with #70.