Transitivity

.github/workflows/CI.yml

@@ -18,7 +18,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ["3.9", "3.13"]
+        python-version: ["3.9", "3.14"]
     steps:
       - uses: actions/checkout@v4
         with:
@@ -27,13 +27,14 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
       - name: Run Cargo Tests
-        run: cargo test --features python -- --nocapture --test-threads=1
+        run: cargo test -- --nocapture --test-threads=1
       - name: Run pytest
         run: |
           # just venv pytest
           rm -rf .venv
           python3 -m venv .venv
           . .venv/bin/activate
+          .venv/bin/pip install --upgrade pip
           .venv/bin/pip install wheel pytest maturin
           maturin develop
           .venv/bin/pytest python/tests
@@ -46,11 +47,11 @@ jobs:
       matrix:
         platform:
           - target: x64
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
           - target: aarch64
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
           - target: armv7
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
     steps:
       - uses: actions/checkout@v4
         with:
@@ -77,13 +78,13 @@ jobs:
         platform:
           - target: x86_64-unknown-linux-musl
             arch: x86_64
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
           - target: i686-unknown-linux-musl
             arch: x86
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
           - target: aarch64-unknown-linux-musl
             arch: aarch64
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
         # all values: [x86_64, x86, aarch64, armhf, armv7, ppc64le, riscv64, s390x]
         # { target: "armv7-unknown-linux-musleabihf", image_tag: "armv7" },
         # { target: "powerpc64le-unknown-linux-musl", image_tag: "ppc64le" },
@@ -113,7 +114,7 @@ jobs:
     strategy:
       matrix:
         target: [x64, x86]
-        interpreter: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+        interpreter: ["3.9", "3.10", "3.11", "3.12", "3.13", "3.14"]
     steps:
       - uses: actions/checkout@v4
         with:
@@ -151,9 +152,9 @@ jobs:
       matrix:
         platform:
           - target: x64
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
           - target: aarch64
-            interpreter: 3.9 3.10 3.11 3.12 3.13
+            interpreter: 3.9 3.10 3.11 3.12 3.13 3.14
     steps:
       - uses: actions/checkout@v4
         with:

CHANGELOG.md

@@ -5,11 +5,23 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## Unreleased
+## [1.1.0] - Unreleased
+
+### Added
+
+- Add opt-in transitive cost closure via `WeightedLevenshtein.transitive_closure()`.
+- Support for Python 3.14.
+- `__str__` and `__repr__` methods for `WeightedLevenshtein`.
 
 ### Changed
 
-- Small runtime improvements in Rust backend.
+- Cap `default_substitution_cost` at `default_insertion_cost + default_deletion_cost`, since a substitution can always be expressed as a deletion followed by an insertion.
+- Treat empty-sided substitution costs as insertion/deletion aliases.
+- Reject non-finite (NaN, infinite) costs during validation.
+
+### Fixed
+
+- Use the minimum cost for conflicting symmetric substitutions.
 
 ## [1.0.1] - 2025-09-21
 

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "ocr_stringdist"
-version = "1.0.1"
+version = "1.1.0"
 edition = "2021"
 description = "String distances considering OCR errors."
 authors = ["Niklas von Moers <niklasvmoers@protonmail.com>"]
@@ -14,8 +14,5 @@ name = "ocr_stringdist"
 crate-type = ["cdylib"]
 
 [dependencies]
-pyo3 = { version = "0.24.0", features = ["auto-initialize"] }
+pyo3 = { version = "0.28.3", features = ["auto-initialize"] }
 rayon = "1.10.0"
-
-[features]
-python = []

Justfile

@@ -8,15 +8,20 @@ pytest:
     uv run pytest --cov=python/ocr_stringdist python/tests
 
 test:
-    cargo llvm-cov --features python
-    #cargo test --features python
+    cargo llvm-cov
+    #cargo test
 
 mypy:
     uv run mypy .
 
 lint:
+    cargo clippy
     uv run ruff check . --fix
 
+format:
+    cargo fmt
+    uv run ruff format
+
 doc:
     uv run make -C docs html
 

docs/source/index.rst

@@ -56,6 +56,7 @@ Contents
    getting-started
    examples
    end_to_end_example
+   transitive_closure
    cost_learning_model
    api/index
    changelog

docs/source/transitive_closure.rst

@@ -0,0 +1,37 @@
+===========================
+ Transitive Cost Closure
+===========================
+
+By default, :class:`~ocr_stringdist.WeightedLevenshtein` only considers the
+costs you explicitly provide. If `("6", "G")` costs `0.5` and deleting `"G"`
+costs `0.01`, the engine does **not** automatically know that deleting `"6"`
+in context costs `0.51`.
+
+:meth:`~ocr_stringdist.WeightedLevenshtein.transitive_closure` returns a new
+instance whose cost dictionaries are filled with these effective (transitive)
+costs. The closure also materializes mixed chains, e.g. a `del("y") + ins("x")`
+sequence becoming an effective `("y", "x")` substitution.
+
+Example
+=======
+
+.. code-block:: python
+
+    from ocr_stringdist import WeightedLevenshtein
+
+    wl = WeightedLevenshtein(
+        substitution_costs={("6", "G"): 0.5},
+        deletion_costs={"G": 0.01},
+    ).transitive_closure()
+
+    # The chain "6" -> "G" -> ε is now a single effective deletion at 0.51.
+    print(wl.distance("06", "0"))  # 0.51
+
+After closure, :meth:`~ocr_stringdist.WeightedLevenshtein.explain` returns a
+single flat operation at the effective cost; the underlying chain is not
+preserved.
+
+You may pass ``prune=True`` to the ``transitive_closure`` method to remove
+generated substitutions whose costs are already represented by matches,
+insertions, deletions, or shorter substitutions. This shrinks the resulting
+cost map but is significantly more expensive to compute.

pyproject.toml

@@ -20,7 +20,7 @@ documentation = "https://niklasvonm.github.io/ocr-stringdist/"
 
 
 [tool.maturin]
-features = ["pyo3/extension-module", "python"]
+features = ["pyo3/extension-module"]
 python-source = "python"
 module-name = "ocr_stringdist._rust_stringdist"