blevesearch / bleve

54%

LAST BUILD ON BRANCH master

branch: SELECT

CHANGE BRANCH

x

• No branch selected
• HEAD
• ajroetker/add-dynamic-index-templates
• alice
• backport_MB-37392
• backport_json_tags
• backport_merger_fix
• backport_polygon1
• backport_polygon_changes
• close_err
• compaction_bytes_stats
• conditional_snappy
• config_fix
• cosineFix
• coveralls
• cut_snappy_calls
• docnum_missing_fix
• error_fuzzy
• firestorm
• govariant_removal
• knnDup
• loadchunk_minor
• mad-hatter-release
• master
• max_segment_size
• minor_docvalue_space_savings
• minor_fixes
• new_lev
• newlev_adoption
• numsnapshots_config
• open_fix
• persister_fix
• pre-bbolt-bugfix
• scorch
• single-disjunction-min
• temp
• tfr_advance
• travis_failure_fix
• v0.1.0
• v0.2.0
• v0.3.0
• v0.4.0
• v0.5.0
• v0.6.0
• v0.7.0
• v0.8.0
• v0.8.1
• v0.8.2
• v0.9.0
• v1.0.5
• vellum_for_fuzzy_regex
• vellum_sha
• vet_fixes
• vulcan
• zap_version

Build # 20225609024

Build Type

push

github

Committed by web-flow

Commit Message

MB-69655: Fix vector normalization to handle multi-vectors correctly (#2264)

Author: @CascadingRadium from
https://github.com/blevesearch/bleve/pull/2260

- When indexing multi-vector fields (e.g., `[[3,0,0], [0,4,0]]`) with
`cosine` similarity, normalization was incorrectly applied to the entire
flattened array instead of each sub-vector independently, resulting in
degraded similarity scores.
- Added `NormalizeMultiVector(vec, dims)` that normalizes each
sub-vector separately, fixing scores for multi-vector documents (e.g.,
score now correctly returns 1.0 instead of 0.6 for exact matches).

---------

Co-authored-by: Rahul Rampure <rahul.rampure@couchbase.com>

Run Details

19095 of 35544 relevant lines covered (53.72%)

0.6 hits per line