Model_confidence: linear_norm

I always get this error every time: /home/charles/Documents/DU/chatbot/venv/lib/python3.7/site-packages/rasa/utils/train_utils.py:535: UserWarning: model_confidence is set to softmax. It is recommended to try using model_confidence=linear_norm to make it easier to tune fallback thresholds. category=UserWarning,

But I have set model_confidence: linear_lorm in my config.yml file. How do I solve this, please?

1 Like

That is not an error, but a warning. If you prefer to use softmax, just ignore it.

To avoid the warning you need to set model_confidence: linear_lorm both for the DIETClassifier and for the TEDPolicy

2 Likes

Thank you but sadly, this still didn’t work. I have

  • name: DIETClassifier

    epochs: 100

    constrain_similarities: true

    model_confidence: linear_norm

and

  • name: TEDPolicy

    max_history: 5

    epochs: 100

    constrain_similarities: true

    model_confidence: linear_norm

1 Like

Are you using a ResponseSelector? If so, you also need to set model_confidence: linear_norm for it to get rid of the warning. If that is not the case, I am afraid I do not know why you get the warning.

I guess the warning message could be improved to specify what component it refers to.

4 Likes

Thank you. This worked.

I think this was as a result of poor familiarity with the documentation