Skip to content

build_subsets takes shared terms from free text, so a note that rejects a term creates an overlap #254

Description

@realmarcin

Found in the round-1 review of #244 (#84).

Where: scripts/fleet/build_subsets.py scan(). rx.findall(txt) runs over the whole record text, including notes, change logs and data_source lines.

Example: MediaIngredientMech's mapped/Cobalamine.yaml is grounded at CHEBI:30411 (cobalamin). Its notes say the old CAS number belonged to cob(I)alamin, "ChEBI:15982 … a THIRD oxidation state", and was cleared. Once #244 folds the ChEBI: spelling, the CommunityMech–MediaIngredientMech overlap lists CHEBI:15982 cob(I)alamin as a term the two share. The record rejects that term. Uppercase CHEBI: mentions in prose have always been matched the same way, so this predates #244; #244 only adds one such instance.

Fix: decide whether overlap terms should come only from structured fields (the STRICT positions, lists of ids) rather than any mention. That is a methodology change to the overlaps, for its own PR.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions