TrainingDataExporter

[wochinge]

I propose a component TrainingDataExporter which mirrors the TrainingDataImporter.

class TrainingDataExporter:

 @staticmethod
 def save_domain(domain: Domain, path: Text) -> None:
 pass

 @staticmethod
 def save_training_data(training_data: TrainingData, path: Text) -> None:
 pass

 @staticmethod
 def save_stories(stories: List[Story], path: Text) -> None:
 pass

 @staticmethod
 def save_config(config: Dict, path: Text) -> None:
 pass

Goals:

• have the option to change the export behavior globally
• users should be able to implement their own module for exporting data
• we should move out the persistence stuff out of our domain objects (improves encapsulation)

Implementations:

• one which exports markdown / yaml
• one which overwrites the last one and exports NLU as json

[imLew]

How would this be used? Since the data is already very accessible I imagine this would mainly be a tool that converts between different formats. I definitely like the idea of allowing users to have their own format!

Why put it in the config.yml though? My understanding is that that is config for the actual bot. If we have more tools like this that are project specific but not bot specific we should consider having a project-config somewhere.

[wochinge]

Why put it in the config.yml though?

Good point. Currently that's slowly diverging from a bot-config to a general config (DataImporter is also using that). I'm personally not a huge fan of having even more files in a rasa project than we have now, but I see your point.

How would this be used? Since the data is already very accessible I imagine this would mainly be a tool that converts between different formats. I definitely like the idea of allowing users to have their own format!

That's what I wrote in the goals.

• converting between the data
• making our entity objects / modules leaner

I imagine to configure that in a config and when I use rasa or even rasa x then it would dump data in the specified format.

[wochinge]

re-opening this after the short discussion in #6323