valentine-1.0.0

What's Changed

• Rename ComaPy to Coma, remove Java dependency by @kPsarakis in #92
• v1.0.0 API redesign: unified match API, ColumnPair, immutable results, metrics overhaul by @kPsarakis in #93
• Add take_top_n_per_source method to MatcherResults by @diogohal in #94
• Add Zensical-based documentation site by @kPsarakis in #95
• Performance: 27× speedup across matchers + Coma accuracy improvements by @kPsarakis in #96
• harden the nltk download data function by @kPsarakis in #98
• Add embeddings support, modify Jaccard, enrich one-to-one filtering by @chrisk21 in #97
• Release v1.0.0 by @kPsarakis in #99

New Contributors

• @diogohal made their first contribution in #94

Full Changelog: https://delftdata.github.io/valentine/changelog/#v100-2026-05-14

valentine-0.5.0

Release notes for v0.5.0:

New: Pure Python COMA algorithm (ComaPy) — #84

• Adds ComaPy, a pure Python implementation of the COMA 3.0 schema matching algorithm, eliminating the Java 6 runtime dependency (40MB JAR + subprocess IPC)
• Faithfully reproduces COMA's matching pipeline: trigram name/path similarity, TF-IDF instance matching with global IDF, and structural matchers — all combined via max-matching Dice set aggregation
• Configurable via use_schema, use_instances, delta, threshold, and max_n parameters
• The Java Coma class is now deprecated and will be removed in v1.0.0, when ComaPy will be renamed to Coma

Improved: Cupid algorithm aligned with the original paper — #81

• Fixed structural matching, leaf similarity range, name similarity formula, and wsim recomputation
• Added integration tests based on examples from the Cupid paper (Figures 7 & 8)

Improved: Similarity Flooding aligned with the original paper — #82

• Implemented the paper's prefix/suffix string matcher replacing Levenshtein
• Fixed Formula B and C fixpoint iteration bootstrapping
• Fixed hash/equality contracts for Node and NodePair

Improved: Distribution-Based algorithm aligned with the original paper — #83

• Added missing triangle inequality constraints to correlation clustering
• Fixed intersection EMD computation to use per-source filtering
• Type-aware sorting for global ranks (pure Python, faster, cross-platform)
• Fixed quantile histogram bucket ordering

Internal: Ruff linter and formatter — #80

• Migrated the entire codebase to Ruff for linting and formatting

valentine-0.4.1

What's Changed

• Support Python 3.14

valentine-0.3.0

• Remove pyemd in favor of POT
• Add support for numpy 2.0+

valentine-0.2.1

What's Changed

• Test case improvement by @ThanosTsiamis in #71
• Upgrade to nltk 3.9.1 to address CVE-2024-39705 by @aecio in #75

valentine-0.2.0

new metrics API

valentine-0.1.9

Address the ambiguity with Coma errors as discussed in #58

valentine-0.1.8

• New string similarity functions for the JaccardLeven method.
• Batching API support
• Python 3.12 support

valentine-0.1.7

Code cleanup and removal of unused dependencies

valentine-0.1.6

Fix issue #53 for the DistributionBased method