NVIDIA / NVIDIA/cuvs

[FEA] BSDG: Explore query generation techniques

Open
#2,242 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

feature request
Dominant language
Cuda
Stars
854
Forks
236
Avg merge
3d 3h
Merged PRs (30d)
62

Description

BSDG (Billion Scale Data Generator)

Explore other query generation techniques that allow us to model the relationship between real data and real queries.

Right now we random sample from data and add noise.
In the kNN Decoder model we generate the whole data pool and use heldout queries (from a data distribution aspect, it's still similar to the random sampling above).

Ideally, we would want to

  1. Figure out the relevant distribution translation (if there is any) between the real data and the real queries.
  2. And generate our synthetic data and queries to have that same translation.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing BSDG's current random-sampling-with-noise approach and the kNN Decoder model described in the issue. Compare how real data and real queries are distributed, then identify a candidate distribution-translation and generation technique. Done should include a documented approach for generating synthetic data and queries with the desired relationship.

Written by the indexing model from the issue text.

Assessment

Domain
data, machine-learning, search
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.