Rasa X will not train saying nothing has changed

[Deprecated] Rasa X Community Edition
1

After making changes to my NLU and story in Rasa X when attempting to train, I get the response 'nothing has changed … so we skipped training a new one".

I have updated to the latest version of Rasa X. Deployment is standard and all pods are running fine. Pared back data and Domain to the simplest possible. X is in sync with my Git Master.

This ongoing issue is a critical impediment to our progress and I cannot seem to get any answers - please help!

2

Hi Shaun! I experienced the same just few hours ago. What I found out was that, the error message is too generic to be useful.

There is a high probability that there is a server error going on under the hood. An incompatible custom-component? A misplaced yml key?

It could be anything. If the engine cant parse or train, it just throws up its hands and say, “hey nothing changed, so lets skipped this one.”

I would suggest you open up the logs and watch out for this endpoint: api/projects/default/models/jobs.

You can also check the network tab of your browser to confirm that the result is actually an error 500

3

@wale Thanks for this advice. Your comment is spot-on that the error message is misleading and too generic. I am going to give this feedback to the product manager at Rasa.

I found an error in my Domain yml and fixing this seems to have solved the issue as I managed to train the model.

I am really enthusiastic to make progress with this now, so fingers crossed!

4

Ok, thats wonderful to know! Rasa is really awesome!!!

And I am very pumped for what we can build with it!

5

The problem recurred and I have figured this is an issue with the Rasa X application. The domain file in X syncs incorrectly with my Git master. It pulls responses into actions. So I end up with -utter_ statements sitting in the incorrect place. This prevents the model from training. Without making any changes, when I sync the domain file that X generates back into a new git branch, this also has the responses incorrectly listed under actions. If I then delete the incorrectly located -utter_ responses within the X domain and run train, this works. But only temporarily, as the error recurs with the next X sync back to Master. Responses again get placed under actions - even though the Git Master file is correct. I am logging this as a Github issue. Have lost many dev hours to this problem already :frowning: