Does anyone know if you need to retrain Rasa when you add a lookup tables and do you only have to make sure that there are enough examples in your NLU training data to make them work?
Or do i need to retrain everything, because it doesn’t retrain if a lookup table has more values?
You definitely don’t need to retrain core. Probably also not NLU, because as far as I know a lookup table is just reduced to a regex.
You need to retrain NLU. Rasa models are meant to work just by themselves, so all the lookup table data is saved in the model files.
Once your lookup tables got a certain size, youll even notice that the training process takes some time for the RegexFeaturizer
Is it possible that you can learn the bot to extract entities while it is running (and to not bring your bot offline for retrain?
Yes, you can do this in the Rasa X UI or via the API for example