mlfoundations / mlfoundations/datacomp

Potential duplicates in rewritten subsets

Open
#92 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
792
Forks
65
PR merge metrics
No merged PRs in 30d

Description

Hello DataComp team!

I'm seeking some clarification on the problem setup. To my understanding, when specifying a subset, if I assign a weight > 1 to a particular datapoint, it can appear multiple times in the rewritten dataset. This duplication may result in the same datapoint appearing twice in the same batch during contrastive training, potentially degrading performance (as the same datapoint would be contrasted against another copy of itself).

Do you have any mechanisms or suggestions within DataComp to help detect or handle these duplicate datapoints? If not, how would you recommend mitigating potential issues caused by having duplicates in the final dataset?

Thank you in advance for your guidance!

Contributor guide

No contributing guide indexed for this repository

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 with resharder.py around line 948, where the issue says subset weights can duplicate datapoints in the rewritten dataset. Trace how weighted records flow into dataset batches and review the contrastive-training assumptions described in the issue. The scope is unresolved: completion would require an agreed mechanism or recommendation for detecting or handling duplicates.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
data, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.