Repository navigation
Score paths embeddings v2 - #189
Conversation
Codecov Report❌ Patch coverage is
Continue to review full report in Codecov by Harness.
🚀 New features to boost your workflow:
|
|
The subclass fix uses a broad signal. Any edge with a biolink:support_graphs attribute is treated as “inferred” and pushed behind the pair’s own edges. The docstring talks about Gandalf subclass inference, but this also catches other inferred or creative-mode edges and anything merged with support graphs (merge_message/worker.py:130 adds that attribute). If the reranker’s training paths only excluded subclass-inferred edges, this may hide legitimate inferred predicates when a direct edge exists. Please confirm that matches training. The new test covers the intended cases well, including “negated own edge falls back to inferred”. |
ARS payload store and retention, Redis/monitor hardening, score_paths v2 embeddings, Omnicorp decimal shared counts: TRAPI-neutral, taken as is. #194 removes the seven shared-regulator treats rules from ARAGORN's rules file, which this branch keeps in its TRAPI 1.x query shapes: the same seven rules (matched by their Rule text) are removed from the 1.x file, leaving the same 11 treats rules in the same order; the rest of the file is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TtsBikU9y9AFQ5uu9tcioz
Uses a smaller embeddings database (which has no effect on scores, just makes the database slight smaller and faster to read). Also fixes a bug when scoring paths from a scublass node, which caused misses, and preloads search graph into the pathfinder worker so the first query after worker restart doesn't time out the search.