Rasa not executing action that is registered

Rasa Open Source
1

As the title states, rasa is seemingly refusing to run the action instead falling to fallback. I’ve given it interactive training stories with the correct flow and it still doesn’t work. All I want to do is be able to call the action so I can have custom response bodies without having to wait for the form action to call submit. Essentially I want to call this action after every user input and have it return a custom response body.

Side note I’ve tried to find other ways of sending a custom response body back to the user but I need access to the tracker for slot info, which I can’t access in my custom rest api.

The action I want to call

class ResponseBody(Action):

def name(self) -> Text:
    return "response_body"

def run(self, dispatcher: CollectingDispatcher, tracker: Tracker, domain: Dict[Text, Any]):
    dispatcher.utter_message(tracker)
    return []

Right now I am just trying to get this action to work so all I have it do is return the tracker

Story data

Generated Story 6856553841717322574

  • find_a_doctor
    • search_providers_form
    • form{“name”: “search_providers_form”}
    • slot{“requested_slot”: “specialty”}
  • inform{“specialty”: “cardiology”}
    • slot{“specialty”: “cardiology”}
    • response_body
    • action_listen
    • search_providers_form
    • slot{“requested_slot”: “zipCode”}
  • form: inform{“zipCode”: “23233”}
    • slot{“zipCode”: “23233”}
    • form: search_providers_form
    • slot{“zipCode”: “23233”}
    • form{“name”: null}
    • slot{“requested_slot”: null}
    • action_restart The form action works fine but it ignores my action when I tell it to execute it

Domain actions

actions:

  • utter_set_isLive
  • utter_out_of_scope
  • utter_noworries
  • utter_greet
  • customized_fallback
  • utter_goodbye
  • utter_handle_insult
  • utter_nice_to_meet_you
  • utter_default
  • utter_telljoke
  • utter_ask_continue
  • response_body
  • utter_ask_howdoing
  • action_chitchat

Rest API

class RestInput(InputChannel): “”"A custom http input channel.

This implementation is the basis for a custom implementation of a chat
frontend. You can customize this to send messages to Rasa Core and
retrieve responses from the agent."""

@classmethod
def name(cls):
    return "bot"

@staticmethod
def on_message_wrapper(on_new_message, text, queue, sender_id):
    collector = QueueOutputChannel(queue)

    message = UserMessage(text, collector, sender_id,
                          input_channel=RestInput.name())
    on_new_message(message)

    queue.put("DONE")

def _extract_sender(self, req):
    return req.json.get("sender", None)

# noinspection PyMethodMayBeStatic
def _extract_message(self, req):
    return req.json.get("message", None)

def stream_response(self, on_new_message, text, sender_id):
    from multiprocessing import Queue

    q = Queue()

    t = Thread(target=self.on_message_wrapper,
               args=(on_new_message, text, q, sender_id))
    t.start()
    while True:
        response = q.get()
        if response == "DONE":
            break
        else:
            yield json.dumps(response) + "\n"

def blueprint(self, on_new_message):
    custom_webhook = Blueprint(
        'custom_webhook_{}'.format(type(self).__name__),
        inspect.getmodule(self).__name__, template_folder='templates', static_folder='static')

    @custom_webhook.route("/", methods=['GET'])
    def health():
        return jsonify({"status": "ok"})

    @custom_webhook.route("/ui", methods=['GET'])
    def botui():
        return render_template('index.html')

    @custom_webhook.route("/ui/", methods=['GET'])
    def botui4():
        return render_template('index.html')

    @custom_webhook.route("/parse", methods=['POST'])
    def receive():
        sender_id = self._extract_sender(request)
        text = self._extract_message(request)
        should_use_stream = utils.bool_arg("stream", default=False)

        if should_use_stream:
            return Response(
                self.stream_response(on_new_message, text, sender_id),
                content_type='text/event-stream')
        else:
            collector = CollectingOutputChannel()
            on_new_message(UserMessage(text, collector, sender_id,
                                       input_channel=self.name()))
            return jsonify(collector.messages)

    return custom_webhook

What am I doing wrong here? What do I need to do to have rasa send back a custom response body every time to the user?

2

I guess you did try to utter a string instead of the tracker object? … dispatcher.utter_message(“hello world”)

What does the action server logs say?

3

Yes I did try to utter just a string and the logs say nothing about the action I want to execute. It just skips it over. Rasa core’s memorization policy is not predicting it for what ever reason. Despite the fact I give it a number of stories with the action.

Here is the log from core to show that it doesn’t predict the action at all rasa_core.log (9.7 KB)

4

Bump, still not solved