Skip to content

Report argmax accuracy on TALENT and lift the KumoTabular class limit - #992

Merged
JingangQu merged 1 commit into
mainfrom
talent-argmax-kumo-classes
Sep 26, 2026
Merged

JingangQu merged 1 commit into
mainfrom
talent-argmax-kumo-classes

Conversation

@JingangQu

Copy link
Copy Markdown
Collaborator

Summary

  • Report argmax accuracy on TALENT by passing tune_threshold=False. TALENT's published result tables use argmax predictions (they predate TALENT's threshold tuning), while tune_threshold=True tuned the binary decision threshold for F1 on the validation set, which lowers accuracy and made our results incomparable with the baselines.
  • Drop the 10-class limit of KumoTabular, which handles more than 10 classes through ECOC (Integrate ECOC into KumoTabular #936). TabFM keeps its limit.

TALENT's published accuracies use argmax predictions, while threshold
tuning picked an F1-optimal threshold on the validation set.
`KumoTabular` handles more than 10 classes through ECOC (#936).

Signed-off-by: Jingang Qu <jqu@nvidia.com>
@copy-pr-bot

copy-pr-bot Bot commented Sep 26, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@JingangQu
JingangQu merged commit 6382d58 into main Sep 26, 2026
3 of 4 checks passed
@JingangQu
JingangQu deleted the talent-argmax-kumo-classes branch September 26, 2026 10:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant