Closed NoB0 closed 11 months ago
Current Branch | Main Branch |
---|---|
Why do we need completely separate agent if the focus is on DP?
MovieBotAgent
uses a platform and dialogue connector to send messages between the two participant. However, in the case of RL, we have an environment (part of another PR) performing what we call a step, that is, the DP chooses a dialogue act that is sent to the user simulator, after collecting the user simulator responses, the dialogue state is updated and sent to the DP. To do so, we use a different agent.
Is this PR for training DP only or inference as well?
This PR is for training DP, for inference we will use MovieBotAgent
.
Add agent and dialogue manager updated for RL.