Skip to main content
Version: Main branch

Execute

db.execute is superduper's wrapper around standard database queries:

  • Inserts
  • Selects
  • Updates
  • Deletes

As well as model predictions:

  • Prediction on single data points (streaming)
  • Predictions on multiple data points (batch prediction)

And also queries which consist of a combination of model computations and data operations:

  • Vector-search queries
  • Complex model predictions which include database queries (e.g. "RAG")

Standard database queries are built using a compositional syntax similar to that found in Python database clients such as pymongo and ibis. The API also includes extensions of this paradigm to cover model predictions and vector-searches.

Read more about the differences and approaches to document stores/ SQL data-backends here.

Building queries/ predictions​

All queries consist of a "chain" of methods executed over a base object. The base object can refer to a table/ collection or a model:

q = base_object.method_1(*args_1, **kwargs_1).method_2(*args_2, **kwargs_2)....

Selects​

MongoDB

A MongoDB find query can be built like this:

q = db['collection'].find().limit(5).skip(2)

SQL

A query with on an SQL data-backend can be built with ibis syntax like this:

q = db['documents'].filter(t.brand == 'Nike').limit(5)

Inserts​

MongoDB

Typically insert queries wrap Document instances and call the insert method on a table or collection:

q = db['documents'].insert_many([Document(r) for r in data])

SQL

The ibis insert is slightly different:

q = db['documents'].insert([Document(r) for r in data])

Executing the query​

results = q.execute()

Multiple results

Iterables of results are sent wrapped in a cursor

Indiviudal results

Individual results are sent wrapped in a Document

Read more about .execute here.