Saving the state of an EpsilonGreedy RlAgent to allow resuming q-learning model training

[theshoeshiner]

I'm new to q-learning, so correct me if Im wrong - but it seems to me that in terms of resuming training a model - the state of some agents is important as well, specifically an EpsilonGreedy agent, otherwise when resuming training we will start over and begin re-selecting more random actions rather than matching the regression towards already seen actions.

1. Won't this begin to "un-train" our model, or does the reward function essentially prevent that?
2. Is there any effect at all in starting that regression over (other than inefficiency) ?
3. If there is a negative effect - then is there a built-in way to save the state of an RlAgent? Specifically the internal state of an EpsilonGreedy agent and it's rate tracker? Or does this have to be done manually by my code?