orneryd / NornicDB / 38496198439

89%

Build Type

push

github

Committed by orneryd

Commit Message

fix(linkpredict): predict.stream compares named embeddings by name, bounds its chunk comparison, and combines the full ensemble before topK (#1186)

Review:
- Named embeddings may come from different models: each is compared only
  with the other node's embedding of the same name, and chunk embeddings
  with chunk embeddings (the default group, with a named "default").
- The all-pairs chunk comparison is bounded: past 4,096 pairs in a group,
  each vector is compared with the other side's mean vector, so the cost is
  linear (benchmark: 300 × 300 chunks of 1,024 dimensions in 0.6 ms). The
  source node's embeddings are read once per call.
- The ensemble combines every candidate of every algorithm, then applies
  topK, so a pair's ensemble score doesn't depend on topK.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Coverage Stats

61 of 61 new or added lines in 2 files covered. (100.0%)

11 existing lines in 4 files now uncovered.

192255 of 216492 relevant lines covered (88.8%)

1.05 hits per line