Story structure analysis

[JEM-Mosig]

Description of Problem:
Stories might be inconsistent with one another. E.g. when story 1 is A > B > C, and story 2 is A > B > X, and C and X are bot actions, then rasa core would now know what to learn. Presently, it is hard to identify such history branches in large story files.

Overview of the Solution:
I wrote the story_tree.py script that solves this problem: https://github.com/RasaHQ/research/blob/master/report_multiWOZ/multiwoz/scripts/story_tree.py (private repo). This issue is about integrating it into rasa, so it can be called to check for inconsistencies with rasa test core, and perhaps also be called for statistics / deleting duplicate stories / exploring the story tree.

Examples (if relevant):

These stories are inconsistent:

## Story 1
* greet
  - utter_greet
* mood_unhappy
  - utter_cheer_up
  - utter_did_that_help
* affirm
  - utter_happy

## Story 2
* greet
  - utter_greet
* mood_unhappy
  - utter_cheer_up
* affirm
  - utter_happy

## Story 3
* greet
  - utter_greet
* mood_happy
  - utter_happy

Blockers (if relevant):

Definition of Done:

• UX is decided (how would users call this function?)
• Original story-tree functionality can be leveraged with Rasa
• Tests are added
• Feature is documented
• Blog post is published (?)

[JEM-Mosig]

Created a Colab notebook, testing out the solution: https://colab.research.google.com/drive/1We1EcSp_uA31OooC78sReryhMCkOjEKM