Question on lookup tables


I had a general question about lookup tables in Rasa that how many examples can we provide for a particular entity. If we provide a huge amount of examples (Let’s say more than 50k examples), will the model become slow after training?

Welcome to the community, Ramesh! :slight_smile:

More entries in the table will indeed slow down the training.

If your lookup table is an exhaustive list of all possible values for an entity, then I would go for it, even if it has 50k entries. I used one with 100k.

Anyway, you can still compare training time with and without the lookup, and it would be interesting if you could share your findings :slight_smile:

Hello @ChrisRahme

Thank you so much for the warm welcome and quick response :blush:

Apart from the time it costs for training the model, will it cause the prediction to run slow while testing the trained model?

PS: I will be more than happy to share my findings :slightly_smiling_face:

1 Like

No, speaking from experience, once the model is created, prediction speed won’t be affected :slight_smile:

Or maybe I’m wrong, but if that’s the case, it means the change in speed is really unnoticeable.

1 Like

Thanks a lot for the answer @ChrisRahme. It was really helpful :slightly_smiling_face:

1 Like

Glad to be of help!

1 Like