Silent failure when the number of intents is less than 2

[furqan-shakoor]

Rasa version: '0.15.1'

Rasa X version (if used & relevant):

Python version: 3.7

Operating system (windows, osx, ...): MacOS

Issue: If the number of intents in an application is less than 2, the application does not correctly classify any inputs to an intent.

The endpoints /train and /parse both return 200 in such a case, indicating no errors. But the server logs show the following errors when /parse is called

2019-09-04 17:16:23+0200 [-] 2019-09-04 17:16:23 WARNING  rasa_nlu.classifiers.embedding_intent_classifier  - Failed to load nlu model. Maybe path /Users/furqan/sandbox/furqs/a3ea383570354a609776bbfeb26493dd/2 doesn't exist
2019-09-04 17:16:23+0200 [-] 2019-09-04 17:16:23 ERROR    rasa_nlu.classifiers.embedding_intent_classifier  - There is no trained tf.session: component is either not trained or didn't receive enough training data

I traced the error to the following snippet of code in embedding_intent_classifier.py, in the function train

        intent_dict = self._create_intent_dict(training_data)
        if len(intent_dict) < 2:
            logger.error("Can not train an intent classifier. "
                         "Need at least 2 different classes. "
                         "Skipping training of intent classifier.")
            return

Since this block bails out of the train function, as a consequence the training of the EmbeddingIntentClassifier never goes through and the relevant training files are never saved.

Eventually when /train is called, the load function of EmbeddingIntentClassifier fails because of the snippet if model_dir and meta.get("file"). This translates into the session attribute not being set and then the process method failing because self.session is None

I believe the /train endpoint should return an error in such a case and let the user know that train was in fact not successful

[sara-tagger]

Thanks for the issue, @Ghostvv will get back to you about it soon!

You may find help in the docs and the forum, too 🤗

[erohmensing]

This makes sense. Would you be interested in contributing a fix @furqan-shakoor ?

[jeanfrancois8512]

I have reproduced the behavior in 0.15.1. If failing early is the good behavior when training a component, one way to do it is:

        intent_dict = self._create_intent_dict(training_data)
        if len(intent_dict) < 2:
            raise Exception("Can not train an intent classifier. "
                         "Need at least 2 different classes. "
                         "Skipping training of intent classifier.")

This will make the server return a 500 status code with the message as content. I haven't tested this in master. The server has changed since 0.15.1 and more test will be needed.

[lucianaribeiro]

I would like to work on that

[erohmensing]

awesome @lucianaribeiro 🎉would be great if you could look into this issue on master. Let us know if you need any help! Maybe @jeanfrancois8512 can also help!

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[erohmensing]

not stale, in discussion atm

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale[bot]]

This issue has been automatically closed due to inactivity. Please create a new issue if you need more help.