Problems building an agent


(Helga Sigurdardottir) #1

Hey,

I’m using jupyter notebook and python 3.6.8 to build a conversational agent. My interpreter is parsing intents correctly but I’ve tried five different ways to build an agent and it either goes straight to fallback or gives an answer that is far off from the intent found with the interpreter.

I want to build an agent that:

  1. is in python, not bash code
  2. supports custom fallback
  3. allows an interactive mode so I can see the predicted intents and confidence
  4. answers correctly :grin:

This doesn’t go together and neither building of agent works correctly in my case:

I found some way that seemed to work but doesn’t now because passing policy configuration parameters to agent.train(...) is not supported anymore.

I hope there’s a simple way to solve this, I just feel like I’m getting sent in endless circles so any help would be greatly appreciated. Thanks!


(Juste) #2

Hey @helga. Can you also share how much training data do you have for your NLU and dialogue models?