Spike for reducing coupling between tracker featurizer, state featurizer, domain and policy

[TyDunn]

We discussed reducing coupling between tracker featurizer, state featurizer, domain and policy in Rasa Open Source 3.0. However, we don't have enough information to prioritise this with confidence and include it as part of the 3.0 action plan. We are aiming to make a decision on if we move forward with this based on the outcome of this spike by Thu, Jul 15th.

Definition of done

• [x] Problems with current coupling clearly defined
• [x] State of the codebase at the end of this decoupling articulated
• [x] How will we know that we've achieved it described
• [x] What would be needed of research and engineering explained
• [x] Risks of decoupling identified (in parallel with architecture revamp)
• [x] Enable and research team members have reviewed this document

Probably out of scope of this spike but might be needed to decide re: 3.0?

• [ ] Prototype (quick and dirty approach)
• [ ] Implementation proposal

[TyDunn]

document here

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.