Switch to TF 2.0 and new NLU components

[tabergma]

Proposed changes:

• Switched to Tensorflow 2.0
• Added new NLU components, such as DIETClassifier
• Deprecated some old NLU components, such as CRFEntityExtractor and KerasPolicy

(We still need to finish up some parts of the documentation and some minor changes inside the code itself. Test time has increased quite a bit and we will try to reduce it, but this is not our highest priority for now. This can be also tackled after the merge if needed.)

Status (please check what you already did):

• added some tests for the functionality
• updated the documentation
• updated the changelog (please check changelog for instructions)
• reformat files using black (please check Readme for instructions)

[tmbo]

Test time has increased quite a bit and we will try to reduce it, but this is not our highest priority for now.

I don't think doubling the test time is something we can accept - I do not think we should merge this before looking into this.

[wochinge]

Just had a glimpse, but we should add docstrings to at least all classes and public functions / methods

[tabergma]

Created some issues to tackle remaining cleanup tasks after the merge:

• Docstrings and naming #5319
• Update default docker files #5316
• Update documentation on components and policies #5308
• Get rid of pickle #5303

[Ghostvv]

@wochinge regarding attributes: we kept the way it was done before, just substituted magic strings with constants. I agree it is not very clear from the constants how they are used, but as soon as you see then in context, in my opinion it is quite clear, and additional wording just make it more complicated.

I agree that the whole message attribute logic is not very intuitive, but changing it is out of the scope of this paper PR

[suzhaomin666]

how to use bert chinese

[wochinge]

@suzhaomin666 Please ask questions in the Rasa Forum. This link should hopefully also help you: https://rasa.com/docs/rasa/nlu/components/#hftransformersnlp

[yuxuan2015]

@suzhaomin666 This link should hopefully also help you: https://github.com/lhr0909/rasa-v2-nlu-bert-chinese