Resume training of a DQN agent. How to avoid Epsilon from being reset to max value?

When I want to resume training of an agent, I simply load it and set the "resetexperiencebuffer" option to false, but this does not avoid the exploration (depending on epsilon) to start anew. Is there any way to make the agent start from the exact point it left off without manually setting the epsilon value?

Answer 1

[Emmanouil Tzorakoleftherakis]

Hello,

This is currently not possible, but it is a great enhancement idea. I have informed the developers about your request and it will be considered for a future release.

[Cecilia S.]

excelent, thank you!!