Hello,
I made a custom connector like the one in here Tutorial and it works.
When I send a post to the custom channel I can see the parsed data like this:
rasa.core.processor - Received user message ‘Hi there!’ with intent ‘{‘id’: 172637494295832716, ‘name’: ‘greet’, ‘confidence’: 0.5666244626045227}’ and entities ‘’
Now my question is: how do I get this parsed data and send it to the OutputChannel?
can you elaborate more about this use case or with some example.
PS: please update the tutorial link for the ref its directing to thread
I made a custom connector and in my CollectingOutputChannel I post some data to an endpoint. In that endpoint I want to do some processing on the intent end entities extracted by my nlu model.
Data extracted by nlu model example:
"text": "some text", "intent": { "name": "some_intent", "confidence": 1 }, "intent_ranking": [ { "id": 3545025623098539000, "name": "some_intent", "confidence": 1 }, { "id": -6116029395920568000, "name": "some_intent_2", "confidence": 0 } ], "entities": [ { "entity": "entity1", "start": 0, "end": 3, "confidence_entity": 0.9999949932, "value": "2", "extractor": "DIETClassifier", "processors": [ "EntitySynonymMapper" ] }, { "entity": "entity2", "start": 11, "end": 16, "confidence_entity": 0.9997606874, "value": "today", "extractor": "DIETClassifier" } ] }
In my RestInput I get the request from my service in this format:
'{ "sender": "test_user", message: "Hi there!", metadata: {}}'
Between these two operations, rasa does some processing for extracting the entities and intents, which I can see in the logs:
rasa.core.processor - Received user message ‘Hi there!’ with intent ‘{‘id’: 172637494295832716, ‘name’: ‘greet’, ‘confidence’: 0.5666244626045227}’ and entities ‘’
My intention is to get the data extracted by nlu model (see example above) and send it to my output channel which will send this payload forward to my endpoint which does the processing.
The way I’m doing this currently: I load the last trained model into my custom channel like this
class CustomCallbackInput(RestInput):
"""A custom REST http input channel that responds using a callback server.
Incoming messages are received through a REST interface. Responses
are sent asynchronously by calling a configured external REST endpoint."""
@classmethod
def name(cls) -> Text:
return "customcallback"
@classmethod
def from_credentials(cls, credentials: Optional[Dict[Text, Any]]) -> InputChannel:
return cls(EndpointConfig.from_dict(credentials))
def __init__(self, endpoint: EndpointConfig) -> None:
self.callback_endpoint = endpoint
self.model = RasaNLUInterpreter('./models/current_nlu')
The problem with this approach is that when I run the channel, I have two models loaded, the one from rasa.core and this one, from my custom channel. I only want to use the model from rasa.core and send the predictions to the output channel, but I don’t know how to extract those predictions
@nik202 Any update on this one?
You’ll need to subclass the OutputChannel class provided by Rasa and implement the necessary methods to send messages.
Here’s a basic example:
from rasa.core.channels.channel import OutputChannel
class CustomOutputChannel(OutputChannel): def init(self): pass
async def send_text_message(self, recipient_id: Text, text: Text, **kwargs: Any) -> None:
# Implement logic to send text message to your custom channel
pass
# Implement other methods as needed, such as sending images, buttons, etc.
In your custom connector where you handle incoming messages, you’ll need to create an instance of your custom output channel and use it to send messages back to the user for crm enrichment. Here’s a basic example of how you might modify your connector: from rasa.core.channels.channel import InputChannel
class CustomInputChannel(InputChannel): def init(self): pass
async def send_message(self, message):
# Parse the incoming message if needed
parsed_data = self.parse_message(message)
# Get the intent and entities from the parsed data
intent = parsed_data.get('intent')
entities = parsed_data.get('entities')
# Here, you'll likely have some logic to process the intent and entities
# Send the processed response back to the user using your custom output channel
output_channel = CustomOutputChannel()
await output_channel.send_text_message(message['sender_id'], "Your response here")
# Implement other methods as needed, such as message parsing, etc.