Using Custom SpaCy model

Rasa Open Source
1

Hi, I was developing a chatbot in Urdu language and I wanted to know if there is an existing pipeline that I could use for Urdu language. Moreover, I found a custom SpaCy model, could that be used for training my chatbot? Thank you in advance.

2

Check out Language Support for info on adding a custom spaCy model and linking it to the language code. Once you have done that you could use the pretrained_embeddings_spacy pipeline which would look like:

language: ur
pipeline: pretrained_embeddings_spacy

in your config file.

3

Hi @mloubser thanks for helping me out. I’ve set these in the config file now but I have another issue regarding linking the model. The first command I ran was:

pip install dist/ur_model-0.0.0.tar.gz

Then, when I try to link the SpaCy model (ur-model-0.0.0.tar.gz) from terminal using the command

python -m spacy link ur-model ur

I get the error: Could not load model data Can you help me out on this? Any help would be greatly appreciated.

4

Sounds like an issue with your custom model - you’ll probably find better info on spaCy’s docs & github, since this is unrelated to rasa.

5

@taimoor-ahmed Are you done with Urdu bot? Can you help me with that. I am creating a bot too and its urgent.

6

Yes, it’s still in development but I have developed a prototype. What do you need help with?

7

Dear can you plz share the sample of prototype. It will be very gratefull of you. I have the model you mentioned earlier in the post but i am not able to connect it to rasa. Secondly how you use slot filling for Urdu.

8

@Talhag958 I didn’t end up using the SpaCy model for Urdu because I was unable to integrate it in RASA. Here is my config.yml file which I am using.

Configuration for Rasa NLU.

Components

language: ur

pipeline:

  • name: WhitespaceTokenizer
  • name: RegexFeaturizer
  • name: LexicalSyntacticFeaturizer
  • name: CountVectorsFeaturizer
  • name: CountVectorsFeaturizer analyzer: “char_wb” min_ngram: 1 max_ngram: 4
  • name: DIETClassifier epochs: 100
  • name: EntitySynonymMapper
  • name: ResponseSelector epochs: 100

policies:

  • name: MemoizationPolicy
  • name: KerasPolicy
  • name: MappingPolicy
  • name: FormPolicy
9

Dear are your slots working??

10

They do not work for most times. Did you get them to work ?

11

@mirfan899 has done marvelous job. Here is link of repository for urdu bot.

1 Like