How to set slots in custom components?

Rasa Open Source
1

Hi Rasa Community,

I am currently developing a custom component for sentimental analysis. What I want is after the user input is classified by the pretrained model, the slot of the predicted label could be set to true. Like this:

User input: “I am feeling excited today!” → Slot: {“excitement”: True}

According to this post, it seems like I have to store the label in an entity first, so I did the following.

image
image1479×1097 110 KB

Then in the domain file, I set auto_fill of these slots to be True. So far the code is working properly and when I am in debug mode, the slot is actually set. But the problem is that the rasa test couldn’t run:

image
image1422×231 28.8 KB

I got the test running by assigning zeros the starting position and ending position when I convert to rasa compatible format:

image
image948×312 16 KB

But the story test failed, and the error log is:

image
image1473×57 8.5 KB
Here is the test_stories.yml:
image
image1240×128 4.25 KB

I think it is failing because of the random number I assigned to “start” and “end”. But as you can see in the above example, “excitement” is not a part of the sentence, so I couldn’t specify the values for “start” and “end”.

In summary, my main questions are:

  1. Is there any way to get around with it (e.g. setting specific numbers to “start” and “end”), so that I can pass rasa test?
  2. Is there any way to set slots in rasa NLU pipeline (before the core)?

Thanks for your help! Yinzhou

2

You can set entity in message attribute, like below:

        extracted = [
            {
                "start": 0,
                "end": len(text),
                "text": text,
                "value": "San",
                "confidence": 1.0,
                "additional_info": None,
                "entity": 'city',
                EXTRACTOR: self.name
            }
        ]

        message.set(ENTITIES, message.get(ENTITIES, []) + extracted, add_to_output=True)

It’s work fine for me. You can clone my project, and add above code in tokenizer file.

1 Like