Rasa X can't find Response Selector data

[Deprecated] Rasa X Community Edition
1

My Rasa model has many FAQ-type responses stored in data/responses.md. I do it this way because i use the ResponseSelector in my pipeline. But Rasa X shows this as inconsistency in my domain, because it says the intents are not present in my training data. It works fine on local deployment without Rasa X.

2

You should still have the corresponding intents in your NLU training data - only the responses go in responses.md. What does your NLU data & responses file look like? And the intents section of your domain file?

3

This is a sample of my nlu.md:

## intent:faq/ask_reset_pw
- where do i reset my password?
- password reset page
- how to reset my password
- i need to reset my password

## intent:faq/ask_where_asset_tag
- where can i find my asset tag?
- where is asset tag?
- where to look for asset tag?

All my FAQ intents are in the NLU training data.

In responses.md, the individual responses in this form:

## ask where is asset tag
* faq/ask_where_asset_tag
	- The asset tag can be found on the label on the back of your computer.

And I have listed the faq intent in my domain.yml. Could this issue be specific to ResponseSelector?

I’m also getting errors such as this in Rasa X logs (```docker-compose logs rasa-x``), and I’m unable to train my model:

rasa-x_1           | /usr/local/lib/python3.6/site-packages/rasa/core/training/dsl.py:422: UserWarning: Found unknown intent 'why' on line 7. Please, make sure that all intents are listed in your domain yaml.
rasa-x_1           |   docs=DOCS_URL_DOMAINS,
rasa-x_1           | /usr/local/lib/python3.6/site-packages/rasa/core/training/dsl.py:422: UserWarning: Found unknown intent 'install_software' on line 4. Please, make sure that all intents are listed in your domain yaml.
rasa-x_1           |   docs=DOCS_URL_DOMAINS,

These intents are also in my NLU, stories, and listed in the domain file.

4

Hi @visvamba you can’t train your model on Rasa X with the ResponseSelector component. You can, however, still annotate all your data without any issues. Training could then happen via CI/CD e.g. on your pull requests in your github repo

5

Thanks for the clarification @akelad. That being the case, do you think a better approach would be to implement the FAQs using regular intents, stories, and responses? Or would this issue be addressed in a future update?

6

It will be addressed in a future update for sure. So I would say continue implementing them with the ResponseSelector