I want to know the use case of unfeaturized slot

Rasa Open Source
1

Went through the documentation, still not clear where to use the unfeaturized slot. Almost every slot i declared as text with few list and categorical types. Can someone please explain what is the significance of using a unfeaturized slot ?

Thanks in advance…

2

Hi @suraj.padhy.gm! You use an unfeaturized slot when you have data you want to store which shouldn’t influence the dialogue flow. Its value is ignored when predicting the next action the bot should run. For example, this could be useful if you get information from an API call (e.g. user_id) in a custom action and want to store it

3

am facing a similar situation …

I have a bot to collect feedback/Compliments/Complaints from employees. have 2 fields

  1. Empcode - Entity, Slot- unfeaturized - Collected at the beginning of the flow, validated against a database - Works well … 2… have a form with 1 Field , Grievancetext - No Entity, Slot - Unfeaturized. Once the employee is validated and chooses to write a feedback, the flow triggers a form to collect the free form text.

However, if the text contains dates, numbers, or amount like 12,000/- , the slot empcode get updated with the date, or number or amount. What could be the problem, am I missing something? e.g.: Empcode = 10007777 - it had this Empcode =02-06-2020 - gets updated. GrieText =response was awaited on 02-06-2020, can you revert ?

can you advice on how to stop this behavior? Thanks !

4

@AnandS This sounds like an issue with entity extraction, not slot featurization. Have you tried figuring out what is going on using rasa shell --debug to see what happens?

5

thanks @tyd for the revert. checked it with --debug

I see that as one enters the text , DIETClassifer is firing and classifying it as as one of the intents in nlu.md “ask_question” and matching the appropriate entities trained in Inform intent and EmpCode is one of the entities.

However, my understanding was that the text/content in a unfeaturized slot should not trigger prediction? I would expect it to just pick up the text and store in the slot for it to be saved. is it not ?

unfeaturized slots to simply store the information without influencing the predictions.(taken from RASA Blog) .

any thoughts ?

6

Hi @tyd ,I have managed to fix the problem.

  1. Inform Intent - New training data added to with the free form text with dates, amount and numbers
  2. the Intent triggering the form - Added the conditions to ignore Entities in domain file.

Thanks for your guidance . Regards!

1 Like
7

Has the unfeaturized slot been deprecated? I don’t see it in the documentation anymore. What is its new equivalent?

8

@baurjanj “The slot type unfeaturized is deprecated. Please use the property influence_conversation as described here or the slot type any instead.” from rasa docs

1 Like
9

@tyd - can you please also share with me the latest document on how to get a user input from the assistant and pass it to the custom actions to further use to query db. Demonstration with actual example would be highly appreciated!