Grables: Tabular Learning Beyond Independent Rows

Tamara Cucumides, Floris Geerts · 2026

Read Grables: Tabular Learning Beyond Independent Rows by Tamara Cucumides, Floris Geerts (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03945.txt.

Description

Tables whose rows depend on each other defeat simple predictors. Tamara Cucumides and 1 co-author pin down when graph structure helps tabular prediction in a 2026 paper. Their tool, grables, splits how a table becomes a graph from how predictions run on that graph.

Row-wise models provably cannot express targets driven by counts, overlaps, or relations across rows. The tool ranks architectures from boosted trees to relational graph networks on one axis. Tests on made-up, transactional, and relational data confirm the gaps they predicted. The hierarchy tells practitioners when a light look at neighbors suffices and when full graph building cannot be avoided.

Catalog record

CREATORTamara Cucumides, Floris Geerts
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE98 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-03945

Browse the full library · About the Tower of Babel