Merge branch 'master' into wmt23

.github/workflows/check-build.yml

@@ -1,7 +1,9 @@
 name: check-build
 
 on:
-  pull_request
+  push:
+  pull_request:
+  workflow_dispatch:
 
 env:
   PYTHONUTF8: "1"
@@ -27,8 +29,6 @@ jobs:
           - os: ubuntu-20.04
             python-version: '3.6'   # test Python 3.6 on older Ubuntu instead
     steps:
-      # - name: update
-      #  run: sudo apt-get -y update
       - uses: actions/checkout@v3
       - name: Setup Python ${{ matrix.python-version }}
         uses: actions/setup-python@v3
@@ -42,12 +42,18 @@ jobs:
         run: choco install wget unzip
       - name: Install python dependencies
         run: |
-          python -m pip install --upgrade pip
-          pip install pytest
-          pip install .[ja]
-          pip install .[ko]
+          python3 -m pip install --upgrade pip
+          pip3 install .[dev]
+          pip3 install .[ja]
+          pip3 install .[ko]
+      - name: Lint with Mypy
+        run: mypy sacrebleu scripts test
+      - name: Lint with Ruff
+        uses: chartboost/ruff-action@v1
       - name: Python pytest test suite
         run: python3 -m pytest
       - name: CLI bash test suite
         shell: bash
         run: ./test.sh
+      - name: Build
+        run: python3 setup.py sdist bdist_wheel

.gitignore

@@ -6,4 +6,5 @@ __pycache__
 sacrebleu.egg-info
 .sacrebleu
 *~
-.DS_Store
+.DS_Store
+.idea/

CHANGELOG.md

@@ -1,10 +1,14 @@
 # Release Notes
 
-
-- 2.4.0 (2023-11-07)
+- 2.4.0 (2023-12-11)
   Added:
   - WMT23 test sets (test set `wmt23`)
 
+- 2.3.3 (2023-11-28)
+  Fixed:
+  - Typing issues (#249, #250)
+  - Improved builds (#252)
+
 - 2.3.2 (2023-11-06)
   Fixed:
   - Special treatment of empty references in TER (#232)

Makefile

@@ -1,5 +1,6 @@
 .PHONY: test
 test:
+	mypy sacrebleu scripts test
 	python3 -m pytest
 	bash test.sh
 

mypy.ini

@@ -18,3 +18,12 @@ ignore_missing_imports = True
 
 [mypy-MeCab.*]
 ignore_missing_imports = True
+
+[mypy-mecab_ko.*]
+ignore_missing_imports = True
+
+[mypy-mecab_ko_dic.*]
+ignore_missing_imports = True
+
+[mypy-sentencepiece.*]
+ignore_missing_imports = True

sacrebleu/__init__.py

@@ -18,14 +18,26 @@
 __description__ = 'Hassle-free computation of shareable, comparable, and reproducible BLEU, chrF, and TER scores'
 
 
-from .utils import smart_open, SACREBLEU_DIR, download_test_set  # noqa: F401
-from .utils import get_source_file, get_reference_files  # noqa: F401
-from .utils import get_available_testsets, get_langpairs_for_testset  # noqa: F401
-from .metrics.helpers import extract_word_ngrams, extract_char_ngrams  # noqa: F401
-from .dataset import DATASETS  # noqa: F401
-from .metrics import BLEU, CHRF, TER  # noqa: F401
+from .utils import smart_open, SACREBLEU_DIR, download_test_set
+from .utils import get_source_file, get_reference_files
+from .utils import get_available_testsets, get_langpairs_for_testset
+from .metrics.helpers import extract_word_ngrams, extract_char_ngrams
+from .dataset import DATASETS
+from .metrics import BLEU, CHRF, TER
 
 # Backward compatibility functions for old style API access (<= 1.4.10)
-from .compat import corpus_bleu, raw_corpus_bleu, sentence_bleu  # noqa: F401
-from .compat import corpus_chrf, sentence_chrf  # noqa: F401
-from .compat import corpus_ter, sentence_ter  # noqa: F401
+from .compat import corpus_bleu, raw_corpus_bleu, sentence_bleu
+from .compat import corpus_chrf, sentence_chrf
+from .compat import corpus_ter, sentence_ter
+
+__all__ = [
+    'smart_open', 'SACREBLEU_DIR', 'download_test_set',
+    'get_source_file', 'get_reference_files',
+    'get_available_testsets', 'get_langpairs_for_testset',
+    'extract_word_ngrams', 'extract_char_ngrams',
+    'DATASETS',
+    'BLEU', 'CHRF', 'TER',
+    'corpus_bleu', 'raw_corpus_bleu', 'sentence_bleu',
+    'corpus_chrf', 'sentence_chrf',
+    'corpus_ter', 'sentence_ter'
+]

sacrebleu/compat.py

@@ -64,7 +64,7 @@ def raw_corpus_bleu(hypotheses: Sequence[str],
 def sentence_bleu(hypothesis: str,
                   references: Sequence[str],
                   smooth_method: str = 'exp',
-                  smooth_value: float = None,
+                  smooth_value: Optional[float] = None,
                   lowercase: bool = False,
                   tokenize=BLEU.TOKENIZER_DEFAULT,
                   use_effective_order: bool = True) -> BLEUScore: