UserWarning: Intent 'nlu_fallback' has only 1 training examples! Minimum is 2, training may fail

Rasa Open Source
1

how can we solve this user warninig…in the rasa document there is nothing defined as we have to add nlu_fallback…if we happen to add what a user might say as a fallback then how would we know all those fallbacks?

1 Like
2

When exactly do you see this kind of warning? I assume you see this when training, for example, the DIETClassifier.

You don’t need to add any examples to nlu_fallback in your NLU training data. The FallbackClassifier you add to your config file does not actually train on any example. It rather checks what intent was set and what confidence was associated with it. If the confidence is below a certain threshold, which you can define in your config file, the FallbackClassifier will overwrite the intent with the default intent name nlu_fallback. If you have a rule in place that reacts on this intent, you can inform the user that, for example, the provided method could not be handled probably and that the user should rephrase. So instead of “teaching” the FallbackClassifier when to fallback by providing examples, the classifier checks how confident the other classifiers are and overwrites their intent if the confidence is too low, e.g. the other classifier is not really sure about the intent.

Does that help?

3

Thanks…but the one that says nlu_fallback has only one training examples…I add some and it starts working

4

I am facing a similar issue on rasa 2.2.1, action_default_fallback the following rule was automated added while migrating form 1.x to 2.x

- rule: Rule to handle messages with low NLU confidence (automated conversion from 'FallbackPolicy')
  steps:
  - intent: nlu_fallback
  - action: action_default_fallback

And the same warning is coming as mentioned by @faiza_conte

1 Like
5

@madanmeena…what is your issue ?

6

I am getting the same warning as you.

7

@madanmeena…they say we are not suppose to add a training example for it ,it should work just like that…but I just add some training example for nlu_fallback and the warning just gone…just add nlu_fallback in nlu data with some training example of what’s not related to u just 2 or 3 is enough so that the warning will disappear…it might not be a best solution but just give it a try…the nlu_fallback works fine but just to get ride of the warining try that

1 Like
8

Yes will do that, I recently migrated from 1.x to 2.2.x, facing a lot of warning, stabilizing the model, and getting rid of them one by one.

This is a big blocker as I am not able to get the insight from rasa, had you faced this issue or any clue.

9

@madanmeena…did you face the issue while migrating?? or after migrating ?

10

after migration.

11

Maybe while migrating u didnt give a permission for some of the installation I guss…try to install again while giving permission by adding –user at the end for all of the command line u use for installation…when i see ur error looks like some of them need permission

12

Make sense, I did no installation, was running directly through docker command I Will try what you suggested, thanks

13

Ur welcome

14

That did not work for me, seems like a version issue

the following does not work

docker run -v $(pwd):/app rasa/rasa:2.2.1-full data validate --debug
docker run -v $(pwd):/app rasa/rasa:2.2.2-full data validate --debug

The following does

docker run -v $(pwd):/app rasa/rasa:2.1.3-full data validate --debug

15

This is a known bug. Also, there are duplicates of this thread in the forum, it looks like it shows up once a month or so.

1 Like
16

I also have this issue - should we ignore it?

17

yes: