Story paths misfiring (Rasa 1.x vs 2.x)

Rasa Open Source
1

I’m having one of those moments and finally giving in and asking for help (I may need professional help soon)

I have a new Rasa 2.6.3 project ( rasa init ) I’ve added a single starter conversation, and I have a yes/no question - Do you need help with that?

and when I choose Yes (/affirm) It displays the out-of-the-box responses to “I’m good” - you know the one

Great, carry on

WHAT?! It’s 100% not following the story below so I’m curious what part of my brain has died to revert my skills back to a first-time-bot-builder.

Here’s my one and only custom story (remember I have the “init bot” code)

  - story: Adobe Photoshop Upgrade Issue - YYx - Yes Help with Deflection
    steps:
      - intent: adobe_photoshop_upgrade_issue
      - action: utter_adobe_photoshop_reset_preferences
      - action: utter_general_support_ask_do_you_need_help_with_that
      - intent: affirm
      - action: utter_adobe_photoshop_reset_preferences_url
      - action: utter_adobe_photoshop_ask_video_helped
      - intent: affirm
      - action: utter_general_support_tell_glad_it_helped

Here’s the shell interaction

2021-06-08 17:00:21 INFO     root  - Rasa server is up and running.
Bot loaded. Type a message and press enter (use '/stop' to exit): 
Your input ->  dragging an image onto a canvas no longer center snaps it
This could be caused by corrupted preferences.  Could you please try resetting the preferences of Photoshop.
? Do you need help with that?  1: Yes I do (/affirm)
Great, carry on!
Your input ->

Here’s the generated config.yml

language: en
pipeline:
# # No configuration for the NLU pipeline was provided. The following default pipeline was used to train your model.
# # If you'd like to customize it, uncomment and adjust the pipeline.
# # See https://rasa.com/docs/rasa/tuning-your-model for more information.
#   - name: WhitespaceTokenizer
#   - name: RegexFeaturizer
#   - name: LexicalSyntacticFeaturizer
#   - name: CountVectorsFeaturizer
#   - name: CountVectorsFeaturizer
#     analyzer: char_wb
#     min_ngram: 1
#     max_ngram: 4
#   - name: DIETClassifier
#     epochs: 100
#     constrain_similarities: true
#   - name: EntitySynonymMapper
#   - name: ResponseSelector
#     epochs: 100
#     constrain_similarities: true
#   - name: FallbackClassifier
#     threshold: 0.3
#     ambiguity_threshold: 0.1
policies:
# # No configuration for policies was provided. The following default policies were used to train your model.
# # If you'd like to customize them, uncomment and adjust the policies.
# # See https://rasa.com/docs/rasa/policies for more information.
#   - name: MemoizationPolicy
#   - name: TEDPolicy
#     max_history: 5
#     epochs: 100
#     constrain_similarities: true
#   - name: RulePolicy

Any ideas what the gap in my brain can’t figure out?

UPDATE I ran rasa interactive and get the exact same story as I manually put together.

1 Like
2

Update #2 : Reverted Rasa

I reverted back to Rasa 1.10.14 to make sure I wasn’t crazy. I reformatted the stories back to markdown along with the nlu data, added my responses to my domain file and it worked. I got the expected (boring) story working in 1.x

1 Like
3

@jonathanpwheat ,

Can you talk to it in debug mode, using rasa shell --debug, and post the full log?

Maybe that explains why it predicted the wrong next action.

4

You got me wondering, so I tried it out myself, and it does work.

I am using rasa 2.7

Bot loaded. Type a message and press enter (use '/stop' to exit): 
Your input ->  dragging an image onto a canvas no longer center snaps it                                                                                                  
This could be caused by corrupted preferences.  Could you please try resetting the preferences of Photoshop.
Do you need help with that?
Your input ->  Yes I do                                                                                                                                                   
Reset your preferences here...
Did that photoshop video help?
Your input ->

Attached my version.

Maybe this helps you find what is going on.

testbot.zip (7.0 KB)

5

Hey @Arjaan, Thank you for building the testbot, I appreciate you throwing that together. Its weird, your bot works fine for me (I’m using 2.6.3), can’t really imagine that would cause a difference.

Here’s a link to mine if you’re interested, I wasn’t able to upload it for some reason, it’s only 77k. Dropbox - rasa-photoshop-demo.zip - Simplify your life I know you can’t get into loading everyone’s bots or you’d be doing that all day long… but if you’re bored it’s there :slight_smile:

As an aside you mentioned using 2.7 - Is that officially out yet? I don’t see it in the change log Rasa Open Source Change Log

6

Ok. So this is interesting.

I’m in Windows, and as you know it likes to reboot seemingly randomly in the night. I came in this morning and it had rebooted. I tested your bot and it worked perfectly. I tested my bot and it also worked perfectly.

I have a hunch. When doing development, I train and run the shell constantly. Sometimes I /stop out of the shell and sometimes - depending on what happens on the action server side, I get a console error and have to break out (CTRL+C) of the shell. This sometimes leaves a rogue python 3.7 process running and I think the next time I run a shell it attaches to this, or they conflict and I get the problems I was seeing. I work 100% in a Ubuntu WSL shell which also may lead to the process getting left in memory.

I’ve had this happen in the past where a console was actually hitting a different rogue rasa server in memory in a different directory and not the one I just trained up.

I’ll be a little more vigilant with checking Task Manager before I post from now on. This have caught me a couple of times.

Sorry to waste your time, but hopefully this will help other people that may develop on Windows in a WSL shell.

1 Like