openml / openml/OpenML

No way to submit prediction uncertainties that are not probabilities

Open
#1,006 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
PHP
Stars
755
Forks
128
PR merge metrics
No merged PRs in 30d

Description

From talking with @mfeurer it seems like there's no way to submit continuous predictions for classification that are not probabilities.
I think that's an issue because you can compute binary AUC or OVR AUC without probabilities.
Not using decision values leads to inconsistencies (the AUC of an SVM is much lower in openml than if computed with sklearn because the openml computation is wrong).
Also see https://github.com/openml/openml-python/issues/786

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 the prediction-submission behavior described here and the related openml-python issue #786. Determine how continuous classification predictions are currently represented and how AUC is computed, then define what support for non-probability decision values would require and how correctness would be verified against sklearn.

Written by the indexing model from the issue text.

Assessment

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

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.