pgml sdk examples #669
pgml sdk examples #669
Conversation
| @@ -777,5 +777,8 @@ def vector_search( | |||
|
|
|||
| search_results = run_select_statement(conn, cte_select_statement) | |||
| self.pool.putconn(conn) | |||
|
|
|||
| # Sort the list of dictionaries based on the 'score' key in descending order | |||
| search_results = sorted(search_results, key=lambda x: x['score'], reverse=True) | |||
There was a problem hiding this comment.
That's already done in SQL above. Postgres is much better at sorting than Python.
There was a problem hiding this comment.
I convert this to JSON using SQL that returns unsorted array. Is there a way to return a sorted dict using SQL?
SELECT array_to_json(array_agg(row_to_json(t)))
FROM ({select_statement}) t;There was a problem hiding this comment.
select array(select unnest(array[1.0, 2.0, 3.0]) order by 1 desc);
https://stackoverflow.com/questions/2913368/sorting-array-elements
| context = " ".join(results[0]["chunk"].strip().split()) | ||
| context = context.replace('"', '\\"').replace("'", "''") | ||
|
|
||
| select_statement = """SELECT pgml.transform( |
There was a problem hiding this comment.
I think a great future iteration for the pgml.transform call, would be to have collection.vector_search() return a query object rather than a result, and then pass that query object as "context" to a new python function question_answering api. That API should then build the context as a sub-select inside the question-answering, so the documents never leave the database.
There was a problem hiding this comment.
Running the scripts by setting LOGLEVEL=INFO writes the query to the terminal. We could have a dry_run option that outputs a query instead of result.
LOGLEVEL=INFO python examples/question_answering.py
Details on examples here
Also added a roadmap section to main readme