Possible to use new regex & lookup with tensorflow?

Hi, I’m taking a look at Rasa and seeing if it might meet my requirements. My timing looks great, as the new regex and lookup features in the latest release are part of my requirements!

Looking at the examples, the regex & lookup features all seem to be implemeneted using a spacy/sklearn pipeline. Is it possible to use these with the tensorflow pipeline? I get the impression from some discussions here that tensorflow is the preferred option now, so it would be great to see an example nlu_config.yml that allows me to use all these great new features together.

you can use them with tensorflow as well! just add them on to the other components of the tensorflow pipeline

Thanks… I’m a big fan of showing verified examples of things that work, so here I share my pipeline that successfully uses lookups, regex and duckling with rasa/tensorflow:

language: "en"
- name: "tokenizer_whitespace"
- name: "intent_entity_featurizer_regex"
- name: "ner_crf"
- name: "intent_featurizer_count_vectors"
- name: ner_duckling_http
  url: http://localhost:8000
  - email
  - phone-number
  - amount-of-money
- name: "intent_classifier_tensorflow_embedding"

I hope this is useful to some others - it’s important that “intent_entity_featurizer_regex” comes before “ner_crf” (which seems obvious when I think about it, but it’s always nice to have such things documented so that I don’t need to think!).