Can anyone explain me the following components from rasa house?

i wanted to know in short about how the following pipline and policies are created in rasa house…what kind of algorithms that is used to create the following piplines and policies?

  - name: "DucklingEntityExtractor"
    url: "http://localhost:8000"
    dimensions: ["time", "email"]  
  #  timezone: "Ethiopia/addis abeba"
    timeout: 3
  - name: WhitespaceTokenizer
  - name: RegexFeaturizer
  - name: RegexEntityExtractor 
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: "char_wb"
    min_ngram: 1
    max_ngram: 4
  #- name: CRFEntityExtractor  
  - name: DIETClassifier
    epochs: 100
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100
    #retrieval_intent: faq
  - name: FallbackClassifier
    threshold: 0.7  ```

  - name: MemoizationPolicy
  - name: TEDPolicy
    max_history: 8
    epochs: 200
  #- name: "coco_rasa.CoCoContextPolicy"  
  - name: RulePolicy```

Hello @faiza_conte

Please have a look at our docs page on pipeline components.

@j.mosig Okay thanks …what is the use of tensorflow_embedding in rasa??

What do you mean? All our machine learning algorithms are implemented with Tensorflow. Or do you refer to the former KerasPolicy (it doesn’t exist any more and is superseded by TEDPolicy)?

@j.mosig Sorry I mean what is the use of tensorflow in rasa

@j.mosig…can I contact you if you are from rasa team…I have a school presentation tommorow and I have some issues that I couldn’t understand from the rasa…if you are okay please answer me it would help me with my paper thanks

Sorry @faiza_conte, but I cannot help you there. But you might find a lot of good resources on our YouTube channel. Good luck with your presentation!

@j.mosig …okay thanks…just answer me this question…the difference between NLP and machine learning in rasa…which part rasa uses NLP and which part is the machine learning algorithm in rasa

The NLP pipeline has multiple components. Some use ML and some don’t. The main ML component that is typically used is the DIETClassifier.

@j.mosig…so the pipeline it self is the NLP model …am I right??..from the pipeline I sent you…which one of them uses machine learning to build and which one doesn’t use machine learning ?

As I’ve mentioned above, the DIETClassifier.

@j.mosig…okay…the diet classifier is the inbuilt nlp model based on machine learning …did I get u? …

The entire pipeline does natural language processing. Most of that processing does not involve machine learning because machine learning is not a good tool, e.g. to split text into words. But machine learning is useful to classify the featurized token sequence (produced by the components before DIET) into intent categories and/or to find entities. That is what DIET does, as explained in our docs.

@j.mosig…thanks a lot…I have now a clear understanding of the pipelines…one more question…among the pipelines that I have posted above the tokenizers and feuterizers doesn’t use machine learning but the rest uses machine learning … So do u know the kind of machine learning algorithm used to build these pipelines that has machine learning ??

You can use a language model featurizer, such as BERT, if you like.

@j.mosig Are lexicalsynteticfeuterizer ,and count vector feuterizers are rasa inbuilt language feuterizers ??

What do you mean “inbuilt”?

For example diet classifier is rasa inbuilt nlp model I guss…and ducklinghttpextracter is custom component that we include in rasa pipeline if I am not wrong…u said I can use bert if I want for language model feature…and what is the benefit of using that or rasa provide feuterizers so that we can train our data and it could be based on our training data…is bert pretrained model??

Yes, BERT is a pre-trained model from Google. You don’t really need it, but you can add it to the pipeline with the LanguageModelFeaturizer if you like.

@j.mosig…by default what kind of language model feuterizers does rasa provides??