Automatic Column Selection

[ritugala]

Description

A good starting point for reviewing this PR would be: https://github.com/neulab/prompt2model/pull/359/files#diff-952ef426d80e42d77fe849ad91b5ef92bcd907ea702707fb1d9a714d9e31820cR169

Changes:

-Added a file for creating the prompt for column selection, it is similar to the instr_parser_prompt file; The prompt includes incontext learning as well.
-Created a function in dataset_retriever for automatic column selection
-Also included the prompt spec to be passed in canonicalize_dataset_by_cli so that I can use the instruction in column selection prompt.
-Added tests for automatic column selection
-Modified the JSON parser file to handle responses where the JSON keys would have lists as the values (and not just strings)
-Set an explicit value for max_api in json parser incase it is not set - it was getting stuck in an infinite loop otherwise, and constantly calling the API

Issues/Discussions/Question

• Small considerations: like whether I should be throwing an error here because otherwise if the response is not appropriate (like not a json or something) then it throws a max API recall limit reached, rather than the actual error. Should not be hard to resolve, but just wanted to know if I should fix it or let it be :

  prompt2model/prompt2model/utils/parse_json_responses.py

  Line 101 in ad238c8

  logger.error("Maximum number of API calls reached.")

[neubig]

Thanks @ritugala , this looks really exciting!

I had several comments, and wherever possible I tried to use the github suggestion functionality so you can just accept the comments through github if you want.

[zhaochenyang20]

Add some minimal comments in the last minute. Sorry that preparing for my English test takes too much time.

[viswavi]

I left a few minor comments - please address them, and then feel free to merge!

This is great work. This PR clearly required a lot of understanding of our system and external components, so this is a great milestone for your first major PR @ritugala!