llnl / llnl/professor

Add options for handling zero-inflated datasets

Open
#16 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
2
Forks
1
Avg merge
4d 21h
Merged PRs (30d)
1

Description

When testing the new synthetic dataset that is available in PR #2 , I found that the existing generators struggled to handle zero-inflated datasets. In this case, the typical failure mode was for the model to land on a degenerate solution of all zero values.

I've run into a similar problem for a dataset on another project. The fix that we applied in this case was to split the model output head into two parts:

1) A classifier that predicts whether the absolute value is greater than a user-supplied threshold
2) A regressor that will only predict the amplitude for non-zero cells

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 by reviewing PR #2 and the existing dataset generators to understand how the synthetic dataset exposes zero-inflated behavior. Define how user-supplied thresholds and the classifier/regressor split should be configured, then validate that zero-inflated datasets no longer produce an all-zero degenerate solution.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.