Skip to content

Fix TEDS distance over explicit ordered table trees - #76

Closed
lanyue-llk wants to merge 1 commit into
opendatalab:mainfrom
lanyue-llk:codex/fix-teds-ordered-tree
Closed

lanyue-llk wants to merge 1 commit into
opendatalab:mainfrom
lanyue-llk:codex/fix-teds-ordered-tree

Conversation

@lanyue-llk

Copy link
Copy Markdown

Summary

APTED was receiving a serialized string from TEDSMetric. Its default child accessor therefore saw a leaf rather than the parsed table tree. This PR passes explicit nodes and exposes their children to APTED. It also removes the node-count-only fallback, which could silently produce an optimistic score after an algorithm error.

The two existing score assertions that encoded the leaf behavior now expect the actual ordered-tree distance. Added regression cases cover a missing cell and changed cell text.

Validation

tests/test_teds.py: 24 passed using the repository's metric modules and the benchmark runner's locked dependency environment. The full package initializer was bypassed because that environment does not include unrelated WebMainBench data-loader dependencies.

This changes historical TEDS values when table trees differ; consumers should rescore from preserved raw outputs and retain the former official score for provenance.

@lanyue-llk lanyue-llk closed this Sep 23, 2026
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