Handle categorical slots in command generator - [ENG 569]

[varunshankar]

Proposed changes:

• Skips slot set command if the LLM proposes a slot value that is not part of the categorical list of values defined in the domain
• For each collection step in a workflow, the default pattern, namely "pattern_collect_information," is executed. The initial step defined within this pattern is the ActionRunSlotRejections action. I've implemented extra validations within this action to reset any invalid categorical values set by the LLM. This approach allows for the reuse of code responsible for fetching slot values and generating bot responses. Additionally, it ensures that user-defined rejections within the flow take precedence.

Among the potential solutions considered but not yet implemented are:

• Adding a new default action to implement the categorical slot check as the initial step in the pattern_collect_information. This method doesn't leverage most of the existing code in ActionRunSlotRejections, and its necessity for introducing a new default action is uncertain.

• Implementing the categorical slot check before the slotset event in the slot set command. The skip information for the slot is stored in a default slot, triggering the bot utterance step in the default pattern - pattern_collect_information. This approach, however, lacks computational efficiency.

Status (please check what you already did):

• added some tests for the functionality
• updated the documentation
• updated the changelog (please check changelog for instructions)
• reformat files using black (please check Readme for instructions)

[twerkmeister]

Looks good, just a small typing thing to take care of :)

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

5 Bugs
0 Vulnerabilities
3 Security Hotspots
53 Code Smells

0.0% Coverage
0.9% Duplication

The version of Java (11.0.20) you have used to run this analysis is deprecated and we will stop accepting it soon. Please update to at least Java 17.
Read more here

[twerkmeister]

I really like the idea here, I think this is close to a really good way of handling this issue 😎! A few things that I'd like to see improved:

• I think the completely right approach is to have all type checks be a default rejection
  • don't just execute this type check rejection when no others are there. It should be always there
  • apply it here for all types, like floats and bools and categorial slots, not only the categorial slots. This means moving some code from the command generator that catches these cases for floats, bools, etc. here
  • The message should be something like "Sorry, it seems 'X' is not a valid 'Type'"
    • x being the value
    • type being something like "number" if the slot is float, "option" if bool and categorial
    • just first suggestions, feel free to improve upon this yourself. But I think it would be really nice to see the value and the slot this is about
• Maybe you can also undo the big indentation change, it makes the diff really challenging to read