Write intent

Rasa Open Source
1

Hello again,

I’m trying to get the user to type a character string, consisting of + or -, followed by three letters and a number, example: +COJ1, -TPL23, +ERD12

But when I train I get the following:

UserWarning: Misaligned entity annotation in message 'Code +COJ1' with intent 'code_specific'. Make sure the start and end values of entities ([(8, 13, '+COJ1')]) in the training data match the token boundaries ([(0, 7, 'Code'), (9, 13, 'COJ1')]). Common causes:
  1) entities include trailing whitespaces or punctuation
  2) the tokenizer gives an unexpected result, due to languages such as Chinese that don't use whitespace for word separation
  More info at https://rasa.com/docs/rasa/training-data-format#nlu-training-data

This is the intent:

- intent: code_specific
  examples: |
    - Code [new code](code_number)
    - Code [+COJ1](code_number)

The problem is that if I put Code [1](code_number) in the intent, when I run rasa shell the next action works correctly, but if I leave it like this it does not.

Does anyone know why?

Thank you very much!

2

Try using Regex to parse the entities.

It’s the recommended way for parsing ordered data like yours, always +/- followed by 3 letters and a number.

1 Like
3

I’ve read it but I don’t understand how to apply it :frowning:

It isn’t necessary to separate the code, i.e. it’s the same group of characters, it isn’t necessary to separate the sign from the letters and the number.

4

@Alonso

Demo code with a regular expression please do check the regular expression again. ref: https://regex101.com

nlu:
- regex: code_specific
  examples: |
    - [+-a-zA-Z0-9a-zA-Z]+
- intent: inform
  examples: |
     -  Code [-TPL23](code_specific)
     -  Code [+COJ1](code_specific)

You just need to update the config file with RegexFeaturizer and RegexEntityExtractor | NLU Training Data

I hope this will help you.

1 Like
5

Yes, thank you both very much

6

I just wanna propose another Regex since Nik’s will work for anything (m_m, 4>3, 12+A will all match):

^(\+|\-)[a-zA-Z]{3}\d+$

Nik’s works for everything containing +, letters, numbers, and every ASCII character between + and a, all optional and in any order (basically the only thing that will break the pattern are the ASCII characters #32 to #42).

Mine works for tokens starting with + or -, followed by 3 uppercase or lowercase letters, then end with at least one digit.

In my Regex above, any amount of numbers at the end will be taken. If you want for example a minimum of 1 number and a maximum of 4, you can do the following:

^(\+|\-)[a-zA-Z]{3}\d{1,4}$

And, as Nik said, you need to keep at least 2 examples and add RegexFeaturizer and RegexEntityExtractor in your pipeline.

1 Like
7

Thanks!!!

1 Like