WIP: add query baseline query classifier and boilerplate in pipeline

[shahrukhx01]

Proposed changes:

• Query classifier baseline and its boilerplate in the pipeline

Status (please check what you already did):

• [*] First draft (up for discussions & feedback)
• Final code
• Added tests
• Updated documentation

Discussion Points:
I have added the baseline model, however,

• Currently, I'm reading models from temporarily hosted raw files on Github, I'd need some direction in replacing GitHub placeholder files with S3 hosted models for classification.
• Could you please comment on how we are going to extend the QueryClassifier class since I intend to add a transformer finetuned model for the same purpose. Should we then completely replace sklearn based classifier or keep both?
• Any other major pointer that I missed.

Issue: Linked Issue

PS:
This is my first PR on here, please let me know any contribution guideline that I might have missed. Also, any instructions manuals for contributors, which would help me get started with the codebase quickly since I'd like to actively contribute to the haystack in general. Thanks!

[shahrukhx01]

@tholor please let me know how to proceed with this.

[lalitpagaria]

I think Deepset team host these model on their s3 (@tholor WDYT?)
Also It is better to pass model via constructor

[tholor]

Yep, I agree. I will upload it to our s3.

[tholor]

Done. You can find it at https://ext-models-haystack.s3.eu-central-1.amazonaws.com/gradboost_query_classifier/model.pickle

Also added a tiny readme: https://ext-models-haystack.s3.eu-central-1.amazonaws.com/gradboost_query_classifier/readme.txt

[shahrukhx01]

@lalitpagaria thanks for your feedback.
@tholor could you also please upload the TF-IDF vectorizer pickle for feature extraction on S3. https://raw.githubusercontent.com/shahrukhx01/ocr-test/main/query_vectorizer.pickle

[tholor]

Done: https://ext-models-haystack.s3.eu-central-1.amazonaws.com/gradboost_query_classifier/vectorizer.pickle

[lalitpagaria]

I think it should be subclass of BaseComponent

[lalitpagaria]

@shahrukhx01 Thank you for PR.
I did quick review here is my comments -

Currently, I'm reading models from temporarily hosted raw files on Github, I'd need some direction in replacing GitHub placeholder files with S3 hosted models for classification.

Yes I think deepset team can host these model on their s3

Could you please comment on how we are going to extend the QueryClassifier class since I intend to add a transformer finetuned model for the same purpose. Should we then completely replace sklearn based classifier or keep both?

In this case think

• there should be DecisionNode (kind of switch case) derived from BaseComponent like JoinDocuments, which will use BaseClassifier to route incoming request to other direction.
• There should be BaseClassifier base class and then we can have SklearnQueryClassifier and TransformerQueryClassifier
  @tholor What do you think of this design.

Any other major pointer that I missed.

• Better to add code/script to train and benchmark Sklearn classifier