rasa: Rasa X 0.19.1 -- interactive learning broken?

Rasa version: 1.1.3

Rasa X version: 0.19.1

Python version: 3.6.7

Operating system (windows, osx, …): Windows 10

Issue:

Hello,

I updated Rasa X to 0.19.1. It has a lot of new cool features (automatically updating domain file is very nice!) but unfortunately I ran into a new problem:

When I talk to the bot in interactive learning mode, Core is not suggesting me any action.

For example:

I say hi: NLU detects the greet intent: step1

I click next, Rasa X is not proposing me any actions! Instead, it just runs actions_listen again: step2

The shell based interactive mode however, works perfectly: shell

Talking to the bot in Rasa X’s talk mode, also works fine: talkmode

I didn’t have this problem with Rasa X version 0.19.0.

A bug?

Cheers, Philippe

About this issue

  • Original URL
  • State: closed
  • Created 5 years ago
  • Comments: 24 (14 by maintainers)

Most upvoted comments

@PhilippeBaert the fix has already been implemented, and will be part of the next release. We’re working on it and it should be out in the next few days

Thanks @PhilippeBaert - this was very helpful. The endpoint tries to serialise a set which it shouldn’t. This serialisation seems to work anyway in some json.dumps() implementations, but fails on Windows. We’ll fix this asap - I’ll get back to you once it’s released

Ok, I can reproduce this on my Windows 10 laptop, following these points:

  • download rasa-x-0-19-2.tar.gz from https://pypi.rasa.com/simple/rasa-x/

  • open Anaconda Prompt

  • conda create -n rasa-iss3802 python=3.6

  • conda activate rasa-iss3802

  • d:\Apps\Anaconda3\envs\rasa-iss3802\python.exe -m pip install rasa-x-0.19.2.tar.gz

  • create a new directory

  • do a “rasa init” and train the models

  • launch Rasa x with “rasa x” (starts in the Brave browser)

  • go to “Talk to your bot” => “Mode: interactive learning”

  • say “hello” and press enter

  • intent “greet” gets detected, I press enter again, no action gets proposed, the bot listens again

Looking into it…