opusfilter 3.2.0__tar.gz → 3.3.1__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (94) hide show
  1. {opusfilter-3.2.0 → opusfilter-3.3.1}/.github/workflows/ci.yml +2 -2
  2. {opusfilter-3.2.0/opusfilter.egg-info → opusfilter-3.3.1}/PKG-INFO +8 -8
  3. {opusfilter-3.2.0 → opusfilter-3.3.1}/README.md +1 -1
  4. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter-duplicates +13 -3
  5. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter-scores +1 -2
  6. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/CHANGELOG.md +32 -1
  7. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/CONTRIBUTING.md +2 -2
  8. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/command_line_tools.md +2 -1
  9. opusfilter-3.3.1/docs/filters/script_and_language_identification_filters.md +204 -0
  10. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/filters/special_character_and_similarity_filters.md +6 -0
  11. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/functions/filtering_and_scoring.md +9 -0
  12. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/installation.md +23 -11
  13. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/references.bib +13 -0
  14. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/autogen.py +11 -2
  15. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/filters.py +31 -11
  16. opusfilter-3.3.1/opusfilter/lid.py +364 -0
  17. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/opusfilter.py +4 -1
  18. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/segment_hash.py +29 -6
  19. {opusfilter-3.2.0 → opusfilter-3.3.1/opusfilter.egg-info}/PKG-INFO +8 -8
  20. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter.egg-info/SOURCES.txt +1 -0
  21. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter.egg-info/requires.txt +11 -6
  22. {opusfilter-3.2.0 → opusfilter-3.3.1}/requirements.txt +4 -2
  23. {opusfilter-3.2.0 → opusfilter-3.3.1}/setup.py +12 -6
  24. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_autogen.py +3 -1
  25. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_filters.py +5 -3
  26. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_lid.py +89 -28
  27. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_segment_hash.py +18 -0
  28. opusfilter-3.2.0/docs/filters/script_and_language_identification_filters.md +0 -58
  29. {opusfilter-3.2.0 → opusfilter-3.3.1}/.gitignore +0 -0
  30. {opusfilter-3.2.0 → opusfilter-3.3.1}/.pylintrc +0 -0
  31. {opusfilter-3.2.0 → opusfilter-3.3.1}/LICENSE +0 -0
  32. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter +0 -0
  33. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter-autogen +0 -0
  34. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter-cmd +0 -0
  35. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter-diagram +0 -0
  36. {opusfilter-3.2.0 → opusfilter-3.3.1}/bin/opusfilter-test +0 -0
  37. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/automatic_configuration.md +0 -0
  38. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/conf.py +0 -0
  39. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/filters/alignment_model_filters.md +0 -0
  40. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/filters/custom_filters.md +0 -0
  41. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/filters/language_model_filters.md +0 -0
  42. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/filters/length_filters.md +0 -0
  43. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/filters/sentence_embedding_filters.md +0 -0
  44. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/functions/downloading_and_selecting_data.md +0 -0
  45. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/functions/preprocessing_text.md +0 -0
  46. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/functions/training_and_using_classifiers.md +0 -0
  47. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/functions/training_language_and_alignment_models.md +0 -0
  48. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/functions/using_score_files.md +0 -0
  49. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/index.rst +0 -0
  50. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/bpe_segmentation.md +0 -0
  51. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/custom_preprocessors.md +0 -0
  52. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/detokenizer.md +0 -0
  53. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/monolingual_sentence_splitter.md +0 -0
  54. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/morfessor_segmentation.md +0 -0
  55. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/reg_exp_sub.md +0 -0
  56. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/tokenizer.md +0 -0
  57. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/preprocessors/whitespaceNormalizer.md +0 -0
  58. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/references.rst +0 -0
  59. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/requirements.txt +0 -0
  60. {opusfilter-3.2.0 → opusfilter-3.3.1}/docs/usage.md +0 -0
  61. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/README.md +0 -0
  62. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/paracrawl_fi-en/create_ce_sets.yaml +0 -0
  63. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/paracrawl_fi-en/create_domain_sets.yaml +0 -0
  64. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/paracrawl_fi-en/create_roc_auc_sets.yaml +0 -0
  65. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/paracrawl_fi-en/devset_100_1_labels.jsonl +0 -0
  66. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/paracrawl_fi-en/devset_100_2_labels.jsonl +0 -0
  67. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/paracrawl_fi-en/prepare_data.yaml +0 -0
  68. {opusfilter-3.2.0 → opusfilter-3.3.1}/example_configs/qed_lm_langid.yaml +0 -0
  69. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/__init__.py +0 -0
  70. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/autogen_cluster.py +0 -0
  71. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/classifier.py +0 -0
  72. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/embeddings.py +0 -0
  73. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/lm.py +0 -0
  74. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/pipeline.py +0 -0
  75. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/preprocessors.py +0 -0
  76. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/subwords.py +0 -0
  77. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/tokenization.py +0 -0
  78. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/util.py +0 -0
  79. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter/word_alignment.py +0 -0
  80. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter.egg-info/dependency_links.txt +0 -0
  81. {opusfilter-3.2.0 → opusfilter-3.3.1}/opusfilter.egg-info/top_level.txt +0 -0
  82. {opusfilter-3.2.0 → opusfilter-3.3.1}/pyproject.toml +0 -0
  83. {opusfilter-3.2.0 → opusfilter-3.3.1}/setup.cfg +0 -0
  84. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_classifier.py +0 -0
  85. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_embeddings.py +0 -0
  86. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_filter_pipeline.py +0 -0
  87. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_lm_filter.py +0 -0
  88. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_opusfilter.py +0 -0
  89. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_preprocessors.py +0 -0
  90. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_subwords.py +0 -0
  91. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_tokenization.py +0 -0
  92. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_util.py +0 -0
  93. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_wordalign_filter.py +0 -0
  94. {opusfilter-3.2.0 → opusfilter-3.3.1}/tests/test_yaml.py +0 -0
@@ -21,7 +21,7 @@ jobs:
21
21
  fail-fast: false
22
22
  matrix:
23
23
  os: [ubuntu-latest]
24
- python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
24
+ python-version: ["3.8", "3.9", "3.10", "3.11", "3.12", "3.13"]
25
25
  requirements-file: ["requirements.txt"]
26
26
  runs-on: ${{ matrix.os }}
27
27
  steps:
@@ -49,7 +49,7 @@ jobs:
49
49
  build-n-publish:
50
50
  name: Build and publish Python 🐍 distributions 📦 to PyPI and TestPyPI
51
51
  needs: build-n-test
52
- runs-on: ubuntu-20.04
52
+ runs-on: ubuntu-latest
53
53
  steps:
54
54
  - uses: actions/checkout@v4
55
55
  - name: Check for main release tag
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: opusfilter
3
- Version: 3.2.0
3
+ Version: 3.3.1
4
4
  Summary: Toolbox for filtering parallel corpora
5
5
  Home-page: https://github.com/Helsinki-NLP/OpusFilter
6
6
  Author: Mikko Aulamo, Sami Virpioja
@@ -18,7 +18,7 @@ Requires-Dist: graphviz
18
18
  Requires-Dist: py3langid>=0.2.2
19
19
  Requires-Dist: matplotlib
20
20
  Requires-Dist: morfessor
21
- Requires-Dist: opus-fast-mosestokenizer>=0.0.8.5
21
+ Requires-Dist: opus-fast-mosestokenizer>=0.0.8.11
22
22
  Requires-Dist: pandas>=1.0.0
23
23
  Requires-Dist: xxhash>=3.2.0
24
24
  Requires-Dist: sentence-splitter
@@ -29,7 +29,9 @@ Requires-Dist: requests
29
29
  Requires-Dist: scikit-learn
30
30
  Requires-Dist: subword_nmt
31
31
  Requires-Dist: tqdm
32
- Requires-Dist: lingua-language-detector>=1.3.0
32
+ Requires-Dist: lingua-language-detector<2.1,>=1.3.0; python_version < "3.10"
33
+ Requires-Dist: lingua-language-detector>=2.1.1; python_version >= "3.10"
34
+ Requires-Dist: iso639-lang
33
35
  Provides-Extra: test
34
36
  Requires-Dist: pytest; extra == "test"
35
37
  Provides-Extra: pycld2
@@ -49,16 +51,14 @@ Provides-Extra: laser
49
51
  Requires-Dist: laserembeddings; extra == "laser"
50
52
  Provides-Extra: varikn
51
53
  Requires-Dist: varikn; extra == "varikn"
54
+ Provides-Extra: heliport
55
+ Requires-Dist: heliport>=0.10.0; extra == "heliport"
52
56
  Provides-Extra: docs
53
57
  Requires-Dist: myst-parser; extra == "docs"
54
58
  Requires-Dist: sphinx; extra == "docs"
55
59
  Requires-Dist: sphinx-rtd-theme; extra == "docs"
56
60
  Requires-Dist: sphinxcontrib-bibtex; extra == "docs"
57
61
  Provides-Extra: all
58
- Requires-Dist: pycld2; extra == "all"
59
- Requires-Dist: py3langid<0.3.0; extra == "all"
60
- Requires-Dist: numpy<2.0.0; extra == "all"
61
- Requires-Dist: fasttext; extra == "all"
62
62
  Requires-Dist: eflomal>=2.0.0; extra == "all"
63
63
  Requires-Dist: jieba>=0.42; extra == "all"
64
64
  Requires-Dist: mecab-python3>=1.0.8; extra == "all"
@@ -98,7 +98,7 @@ Install from source:
98
98
 
99
99
  ### Troubleshooting
100
100
 
101
- OpusFilter should generally work fine on Python 3.8 to 3.12. In the case of troubles, try installing the exact versions in `requirements.txt`:
101
+ OpusFilter should generally work fine on Python 3.8 to 3.13. In the case of troubles, try installing the exact versions in `requirements.txt`:
102
102
 
103
103
  * `pip install -r requirements.txt`
104
104
 
@@ -25,7 +25,7 @@ Install from source:
25
25
 
26
26
  ### Troubleshooting
27
27
 
28
- OpusFilter should generally work fine on Python 3.8 to 3.12. In the case of troubles, try installing the exact versions in `requirements.txt`:
28
+ OpusFilter should generally work fine on Python 3.8 to 3.13. In the case of troubles, try installing the exact versions in `requirements.txt`:
29
29
 
30
30
  * `pip install -r requirements.txt`
31
31
 
@@ -3,6 +3,7 @@
3
3
  import argparse
4
4
  import collections
5
5
  import logging
6
+ import json
6
7
 
7
8
  from tqdm import tqdm
8
9
 
@@ -24,20 +25,29 @@ parser.add_argument('--hash', type=str, default='xxh64',
24
25
  help=('hash function from xxhash library, empty string '
25
26
  'for no hashing (default "xxh64")'))
26
27
  parser.add_argument('--letters-only', action='store_true', default=False,
27
- help='lowercase input strings before hashing')
28
- parser.add_argument('--lowercase', action='store_true', default=False,
29
28
  help='remove all non-letters from intput strings before hashing')
29
+ parser.add_argument('--letter-words-only', action='store_true', default=False,
30
+ help='remove words with non-letter characters before hashing')
31
+ parser.add_argument('--lowercase', action='store_true', default=False,
32
+ help='lowercase input strings before hashing')
33
+ parser.add_argument('--tokenizers', type=str, metavar='JSON', default=None,
34
+ help=('load tokenizer specifications from a JSON list (e.g. \'[["moses", "en"], ["jieba", "zh"]]\'); '
35
+ 'use with --letter-words-only'))
30
36
 
31
37
  args = parser.parse_args()
32
38
 
33
39
  if args.overlap and len(args.overlap) != len(args.files):
34
40
  raise ValueError("The number of the main and overlap input files should match")
35
41
 
42
+ tokenizers = json.loads(args.tokenizers) if args.tokenizers else None
43
+
36
44
  hasher = SegmentHasher(
37
45
  compare='all',
38
46
  method=args.hash,
39
47
  letters_only=args.letters_only,
40
- lowercase=args.lowercase
48
+ letter_words_only=args.letter_words_only,
49
+ lowercase=args.lowercase,
50
+ tokenizers=tokenizers
41
51
  )
42
52
 
43
53
  total = 0
@@ -14,8 +14,7 @@ from tqdm import tqdm
14
14
 
15
15
  from pandas.plotting import scatter_matrix
16
16
 
17
- from opusfilter.util import file_open
18
- from opusfilter.classifier import lists_to_dicts
17
+ from opusfilter.util import file_open, lists_to_dicts
19
18
 
20
19
 
21
20
  logger = logging.getLogger(__name__)
@@ -7,6 +7,35 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
7
7
 
8
8
  ## [Unreleased]
9
9
 
10
+ ## [3.3.1] - 2025-12-10
11
+
12
+ ### Changed
13
+
14
+ - changed `HtmlTagFilter` to accept broken tags in order to avoid inconsistent results
15
+ - update required `opus-fast-mosestokenizer` version for better platform support
16
+
17
+ ### Fixed
18
+
19
+ - broken citations and links in the documentation
20
+
21
+ ## [3.3.0] - 2025-07-02
22
+
23
+ ### Added
24
+
25
+ - `letter_words_only` option for duplicate detection
26
+ - support for selection of possible languages for `lingua` language detection
27
+ - support for language detection based on `heliport` (a port of HeLI-OTS)
28
+
29
+ ### Changed
30
+
31
+ - update library requirements and include Python 3.13 tests
32
+ - removed pycld2 and fasttext from the `[all]` extras
33
+ - refactored all language identification filters to have their own classes and marked `LanguageIDFilter` as deprecated
34
+
35
+ ### Fixed
36
+
37
+ - fix broken import in `opusfilter-scores`
38
+
10
39
  ## [3.2.0] - 2024-08-14
11
40
 
12
41
  ### Changed
@@ -216,7 +245,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
216
245
  First tagged version.
217
246
 
218
247
 
219
- [Unreleased]: https://github.com/Helsinki-NLP/OpusFilter/compare/3.2.0...develop
248
+ [Unreleased]: https://github.com/Helsinki-NLP/OpusFilter/compare/3.3.1...develop
249
+ [3.3.1]: https://github.com/Helsinki-NLP/OpusFilter/compare/3.3.0...3.3.1
250
+ [3.3.0]: https://github.com/Helsinki-NLP/OpusFilter/compare/3.2.0...3.3.0
220
251
  [3.2.0]: https://github.com/Helsinki-NLP/OpusFilter/compare/3.1.0...3.2.0
221
252
  [3.1.0]: https://github.com/Helsinki-NLP/OpusFilter/compare/3.0.0...3.1.0
222
253
  [3.0.0]: https://github.com/Helsinki-NLP/OpusFilter/compare/2.6.0...3.0.0
@@ -5,7 +5,7 @@ issues page. We are also happy to consider pull requests. There are a
5
5
  few rules for pull requests:
6
6
 
7
7
  * Make a pull request to the `develop` branch instead of `master`.
8
- * The code should support at least Python versions from 3.8 to 3.12.
8
+ * The code should support at least Python versions from 3.8 to 3.13.
9
9
  * Please follow [PEP 8](https://www.python.org/dev/peps/pep-0008/). Exception: The maximum line length is 127 characters instead of 79.
10
10
  * Especially for new features, please include test cases for unit testing.
11
11
 
@@ -20,7 +20,7 @@ skips the respective tests if not.)
20
20
 
21
21
  GitHub workflows defined in the project run automatically `flake8`
22
22
  checks and unit testing with `pytest` using Python 3.8, 3.9, 3.10,
23
- 3.11, and 3.12.
23
+ 3.11, 3.12, and 3.13.
24
24
 
25
25
  Especially for larger contributions, consider using a code analysis
26
26
  tool like [Pylint](https://github.com/PyCQA/pylint). Install it
@@ -30,7 +30,8 @@ number of duplicates in it, or two corpora for calculating the overlap
30
30
  between them. The syntax for the `opusfilter-duplicates` is:
31
31
 
32
32
  ```
33
- opusfilter-duplicates [--overlap FILE [FILE ...]] [--hash HASH] [--letters-only] [--lowercase] FILE [FILE ...]
33
+ opusfilter-duplicates [--overlap FILE [FILE ...]] [--hash HASH] [--letters-only]
34
+ [--letter-words-only] [--lowercase] [--tokenizers JSON] FILE [FILE ...]
34
35
  ```
35
36
 
36
37
  The options are essentially the same as for [`remove_duplicates`](remove_duplicates).
@@ -0,0 +1,204 @@
1
+ # Script and language identification filters
2
+
3
+ ## AlphabetRatioFilter
4
+
5
+ Filter segments based on what proportion of the characters are alphabetic characters.
6
+
7
+ Parameters:
8
+
9
+ * `threshold`: minimum proportion of alphabets in a segment (default 0.75)
10
+ * `exclude_whitespace`: whether to exclude whitespace characters from the ratio (default `false`)
11
+
12
+ Returned scores are proportions of alphabetic characters in the
13
+ segments (after removing whitespace if `exclude_whitespace` is
14
+ true). In filtering, all values have to be equal to or greater than
15
+ the minimum threshold.
16
+
17
+ In order to allow have different thresholds per language, the
18
+ threshold parameter can also be given as a list.
19
+
20
+ ## CharacterScoreFilter
21
+
22
+ Filter segments based on what proportion of their alphabetic characters are in a given script. For a list of valid scripts, see e.g. [www.regular-expressions.info/unicode.html](https://www.regular-expressions.info/unicode.html).
23
+
24
+ Parameters:
25
+
26
+ * `scripts`: scripts for input segments
27
+ * `thresholds`: minimum proportion of characters in a script (default 1)
28
+
29
+ Returned scores are proportions of valid characters in the segments. In filtering, all values have to be equal to or greater than the minimum thresholds.
30
+
31
+ ## LanguageIDFilter (deprecated)
32
+
33
+ *Note: The common LanguageIDFilter is now deprecated. Please see the method-specific language identification filters below.*
34
+
35
+ Filter segments based on their language identification confidence scores.
36
+
37
+ Parameters:
38
+
39
+ * `languages`: expected languages (ISO639 language codes) for the segments
40
+ * `id_method`: language indentification method (`langid`, `lingua`, `cld2`, `fasttext`, `heliport`; default `langid`)
41
+ * `thresholds`: minimum identification confidence score for the segments (a single float or a list of floats per language)
42
+ * `fasttext_model_path`: path for a `fasttext` model (required only for the `fasttext` method; default `null`)
43
+ * `langid_languages`: limit detection to a list of ISO 639-1 codes for possible languages (valid only for the `langid` and `lingua` methods; default `null`)
44
+ * `cld2_options`: a dictionary of options for the `cld2` method (valid only for the `cld2` method; default `null`)
45
+ * `lingua_mode`: a string specifying whether to use lingua's `high` or `low` accuracy mode
46
+
47
+ Returned scores are the language identification confidence scores from a given identification method for the segments. The scores range from 0 to 1. In filtering, all values have to be greater than the minimum thresholds. A negative threshold can be used to skip filtering for a language.
48
+
49
+ Currently the following identification methods are supported:
50
+
51
+ * `langid` (default) {cite:p}`lui-baldwin-2012-langid`
52
+ * See <https://github.com/adbar/py3langid>
53
+ * `lingua`
54
+ * See <https://github.com/pemistahl/lingua-py>
55
+ * `cld2`
56
+ * See <https://github.com/CLD2Owners/cld2>
57
+ * Requires [installing optional libraries](../installation.md).
58
+ * `fasttext` {cite:p}`joulin-etal-2016-fasttext` and {cite:p}`joulin-etal-2017-bag`
59
+ * A pretrained model can be downloaded from [fasttext.cc/docs/en/language-identification.html](https://fasttext.cc/docs/en/language-identification.html).
60
+ * Requires [installing optional libraries](../installation.md).
61
+
62
+ ## LangidFilter
63
+
64
+ Filter segments based on the langid method {cite:p}`lui-baldwin-2012-langid`; see
65
+ <https://github.com/adbar/py3langid>.
66
+
67
+ Parameters:
68
+
69
+ * `languages`: expected languages (ISO639 language codes) for the segments
70
+ * `thresholds`: minimum identification confidence score for the segments (a single float or a list of floats per language)
71
+ * `langid_languages`: limit detection to a list of ISO 639-1 codes for possible languages
72
+
73
+ Returned scores are the language identification confidence scores for
74
+ the segments. The scores range from 0 to 1. In filtering, all values
75
+ have to be greater than the minimum thresholds. A negative threshold
76
+ can be used to skip filtering for a language.
77
+
78
+ ## Cld2Filter
79
+
80
+ Filter segments based on the CLD2 method; see <https://github.com/CLD2Owners/cld2>.
81
+
82
+ Requires [installing optional libraries](../installation.md).
83
+
84
+ Parameters:
85
+
86
+ * `languages`: expected languages (ISO639 language codes) for the segments
87
+ * `thresholds`: minimum identification confidence score for the segments (a single float or a list of floats per language)
88
+ * `options`: a dictionary of options for the method (see <https://github.com/CLD2Owners/cld2>; default `null`)
89
+
90
+ Returned scores are the language identification confidence scores from
91
+ a given identification method for the segments. The scores range from
92
+ 0 to 1. In filtering, all values have to be greater than the minimum
93
+ thresholds. A negative threshold can be used to skip filtering for a
94
+ language.
95
+
96
+ ## FastTextFilter
97
+
98
+ Filter segments based on the Fasttext language identification models
99
+ ({cite:t}`joulin-etal-2016-fasttext`, {cite:t}`joulin-etal-2017-bag`), see
100
+ <https://github.com/CLD2Owners/cld2>.
101
+
102
+ Requires [installing optional libraries](../installation.md).
103
+
104
+ Parameters:
105
+
106
+ * `languages`: expected languages (ISO639 language codes) for the segments
107
+ * `thresholds`: minimum identification confidence score for the segments (a single float or a list of floats per language)
108
+ * `model_path`: path for a `fasttext` model
109
+
110
+ Returned scores are the language identification confidence scores from
111
+ a given identification method for the segments. The scores range from
112
+ 0 to 1. In filtering, all values have to be greater than the minimum
113
+ thresholds. A negative threshold can be used to skip filtering for a
114
+ language.
115
+
116
+ ## LinguaFilter
117
+
118
+ Filter segments based on the Lingua method; see <https://github.com/pemistahl/lingua-py>.
119
+
120
+ Parameters:
121
+
122
+ * `languages`: expected languages (ISO639 language codes) for the segments
123
+ * `thresholds`: minimum identification confidence score for the segments (a single float or a list of floats per language)
124
+ * `langid_languages`: limit detection to a list of ISO 639-1 codes for possible languages (default `null`)
125
+ * `lingua_mode`: a string specifying whether to use lingua's `high` or `low` accuracy mode (default `low`)
126
+
127
+ Returned scores are the language identification confidence scores from
128
+ a given identification method for the segments. The scores range from
129
+ 0 to 1. In filtering, all values have to be greater than the minimum
130
+ thresholds. A negative threshold can be used to skip filtering for a
131
+ language.
132
+
133
+ ## HeliportSimpleFilter
134
+
135
+ Filter segments based on the HeLI-OTS method {cite:p}`jauhiainen-etal-2022-heli`; see
136
+ <https://github.com/ZJaume/heliport>.
137
+
138
+ Requires [installing optional libraries](../installation.md).
139
+
140
+ Parameters:
141
+
142
+ * `languages`: expected languages (ISO639 language codes) for the segments
143
+ * `thresholds`: minimum identification score for the segments (a single float or a list of floats per language)
144
+
145
+ This version uses three distinct score values. Score 1.0 is used if
146
+ the method predicts the correct language. If the low-confidence label
147
+ `und` is returned, the language is considered correct with score
148
+ 0.5. Finally, if a wrong language is predicted, the score will
149
+ 0.0. A negative threshold can be used to skip filtering for a language.
150
+
151
+ ## HeliportConfidenceFilter
152
+
153
+ Filter segments based on the HeLI-OTS method {cite:p}`jauhiainen-etal-2022-heli`; see
154
+ <https://github.com/ZJaume/heliport>.
155
+
156
+ Requires [installing optional libraries](../installation.md).
157
+
158
+ Parameters:
159
+
160
+ * `languages`: expected languages (ISO639 language codes) for the segments
161
+ * `thresholds`: minimum identification score for the segments (a single float or a list of floats per language)
162
+
163
+ This version uses the confidence value returned by the method. It is
164
+ the difference between raw scores of the the best (lowest) and second
165
+ best language. The maximum value is 7.0, which is a penalty value for
166
+ the languages that do not match. The minimum value 0 is returned if
167
+ the best language does not match. The low-confidence label `und` is
168
+ considered as a match. A negative threshold can be used to skip
169
+ filtering for a language.
170
+
171
+ ## HeliportRawScoreFilter
172
+
173
+ Filter segments based on the HeLI-OTS method {cite:p}`jauhiainen-etal-2022-heli`; see
174
+ <https://github.com/ZJaume/heliport>.
175
+
176
+ Requires [installing optional libraries](../installation.md).
177
+
178
+ Parameters:
179
+
180
+ * `languages`: expected languages (ISO639 language codes) for the segments
181
+ * `thresholds`: minimum identification score for the segments (a single float or a list of floats per language)
182
+ * `topk`: the number of top languages to consider (default 50)
183
+
184
+ This version uses the raw scores returned by the method. A lower score
185
+ is better. If the correct language is among the top-k languages, the
186
+ score if that language is returned. Otherwise, the penalty value 7 is
187
+ returned. A threshold over the penalty value can be used to skip
188
+ filtering for a language.
189
+
190
+ ## HeliportProbabilityFilter
191
+
192
+ Filter segments based on the HeLI-OTS method {cite:p}`jauhiainen-etal-2022-heli`; see
193
+ <https://github.com/ZJaume/heliport>.
194
+
195
+ Requires [installing optional libraries](../installation.md).
196
+
197
+ * `languages`: expected languages (ISO639 language codes) for the segments
198
+ * `thresholds`: minimum identification score for the segments (a single float or a list of floats per language)
199
+ * `topk`: the number of top languages to consider (default 50)
200
+
201
+ This version transforms the raw scores of of the top-k languages into
202
+ probability-like values. If the correct language is not in the top-k
203
+ list, zero is returned. A negative threshold can be used to skip
204
+ filtering for a language.
@@ -8,6 +8,12 @@ The returned scores are two boolean values indicating whether the
8
8
  segments contain HTML tags. In filtering, a segment pair is accepted
9
9
  if none of the segments contains HTML tags.
10
10
 
11
+ The beautifulsoup4 library is used for detecting HTML. Up to
12
+ OpusFilter version 3.3.0, the behavior on broken HTML tags varies
13
+ depending on the input and installed library versions. For newer
14
+ versions, broken HTML is not considered as HTML, i.e., it is accepted
15
+ by the filter.
16
+
11
17
  ## TerminalPunctuationFilter
12
18
 
13
19
  Filter segments based on a penalty score with respect to the
@@ -11,6 +11,10 @@ Parameters:
11
11
  * `compare`: select files for duplicate comparison (optional; default `all` or a list of indices)
12
12
  * `hash`: select hash algorithm from xxhash (optional; default `xxh64`)
13
13
  * `overlap`: remove overlap with a second set of files (optional; default `null`)
14
+ * `letters_only`: remove all non-letters from intput strings before hashing (optional; default `false`)
15
+ * `letter_words_only`: remove words with non-letter characters before hashing (optional; default `false`)
16
+ * `lowercase`: lowercase input strings before hashing (optional; default `false`)
17
+ * `tokenizers`: load tokenizer specifications from a list (optional; use with `letter_words_only`)
14
18
 
15
19
  Duplicate filtering is recommended as a first step especially if you
16
20
  combine different corpus collections (e.g. data crawled from web) and
@@ -38,6 +42,11 @@ sizes, but if do not care about memory use and want to be extra sure
38
42
  there are no collisions, you can disable hashing by setting the `hash`
39
43
  parameter as empty string or null.
40
44
 
45
+ For the `letter_words_only`, unless the input is pre-tokenized, you
46
+ should select tokenizers for each of the input files. See
47
+ [Tokenizer](../preprocessors/tokenizer.md) for description of the
48
+ tokenizer parameters.
49
+
41
50
  ## filter
42
51
 
43
52
  Filter parallel data with a combination of filters.
@@ -12,7 +12,7 @@ Install from source:
12
12
 
13
13
  Note that all required libraries are not available to install via PyPI
14
14
  on Windows OS. On Linux and MacOS, it should work directly for Python
15
- versions from 3.8 to 3.12.
15
+ versions from 3.8 to 3.13.
16
16
 
17
17
  ## Required libraries
18
18
 
@@ -39,23 +39,29 @@ See `setup.py` for possible version requirements.
39
39
 
40
40
  ## Optional libraries and tools
41
41
 
42
- ### FastText and PyCLD2 language identification
42
+ ### FastText, PyCLD2, and heliport language identification
43
43
 
44
44
  The language identification libraries currently supported out-of-the-box
45
45
  are [py3langid](https://github.com/adbar/py3langid) and
46
- [lingua](https://github.com/pemistahl/lingua-py). The support for for
46
+ [lingua](https://github.com/pemistahl/lingua-py). The support for
47
47
  [PyCLD2](https://github.com/aboSamoor/pycld2) and
48
48
  [FastText models](https://fasttext.cc/docs/en/language-identification.html)
49
49
  have been changed to optional due to the lack of support especially
50
- for newer Python versions.
50
+ for newer Python versions. Also the support for
51
+ [heliport](https://github.com/ZJaume/heliport) is optional. Please
52
+ note that `heliport` has GPL-3.0 license.
51
53
 
52
54
  The PyCLD2 support can be installed automatically with pip by
53
- including the extras `[pycld2]` or `[all]` (e.g.
54
- `pip install opusfilter[pycld2]`).
55
+ including the extras `[pycld2]` (e.g. `pip install
56
+ opusfilter[pycld2]`).
55
57
 
56
58
  The support for FastText models can be installed automatically with
57
- pip by including the extras `[fasttext]` or `[all]` (e.g.
58
- `pip install opusfilter[fasttext]`).
59
+ pip by including the extras `[fasttext]` (e.g. `pip install
60
+ opusfilter[fasttext]`).
61
+
62
+ The support for `heliport` can be installed automatically with pip by
63
+ including the extras `[heliport]` (e.g. `pip install
64
+ opusfilter[heliport]`).
59
65
 
60
66
  ### Jieba and MeCab word segmentation
61
67
 
@@ -72,24 +78,30 @@ automatically with pip by including the extras `[mecab]` or `[all]`
72
78
  ### LASER sentence embeddings
73
79
 
74
80
  For using sentence embeddings filters, you need to install
75
- `laserembeddings` (https://github.com/yannvgn/laserembeddings). It can
81
+ `laserembeddings` (<https://github.com/yannvgn/laserembeddings>). It can
76
82
  be installed automatically with pip by including the extras `[laser]`
77
83
  or `[all]` (e.g. `pip install opusfilter[laser]`). The package will also
78
84
  require a number of additional libraries, including PyTorch, jieba,
79
85
  and MeCab. Note that you need also to download the prebuild models
80
86
  with `python -m laserembeddings download-models`.
81
87
 
88
+ **Note:** The current latest release `laserembeddings==1.1.2` does not
89
+ support `torch>=2.0.0` and thus fails on Python 3.12 and above. The
90
+ code itself works on Python 3.12 and 3.13, but you need to fix
91
+ `pyproject.toml` to allow later `torch` versions, and install it
92
+ manually.
93
+
82
94
  ### VariKN n-gram models
83
95
 
84
96
  For using n-gram language model filters, you need to install the
85
- Python wrapper for VariKN (https://github.com/vsiivola/variKN). It can
97
+ Python wrapper for VariKN (<https://github.com/vsiivola/variKN>). It can
86
98
  be installed automatically with pip by including the extras `[varikn]`
87
99
  or `[all]` (e.g. `pip install opusfilter[varikn]`).
88
100
 
89
101
  ### Eflomal word alignment
90
102
 
91
103
  For using word alignment filters, you need to install elfomal
92
- (https://github.com/robertostling/eflomal). It can be installed
104
+ (<https://github.com/robertostling/eflomal>). It can be installed
93
105
  automatically with pip by including the extras `[eflomal]` or `[all]`
94
106
  (e.g. `pip install opusfilter[eflomal]`). Note that you will need `Cython`
95
107
  for the installation.
@@ -68,6 +68,19 @@
68
68
  pages = "261--266"
69
69
  }
70
70
 
71
+ % Jauhiainen et al. (2022)
72
+ @inproceedings{jauhiainen-etal-2022-heli,
73
+ title = "{H}e{LI}-{OTS}, Off-the-shelf Language Identifier for Text",
74
+ author = "Jauhiainen, Tommi and Jauhiainen, Heidi and Lind{\'e}n, Krister",
75
+ booktitle = "Proceedings of the Thirteenth Language Resources and Evaluation Conference",
76
+ month = jun,
77
+ year = "2022",
78
+ address = "Marseille, France",
79
+ publisher = "European Language Resources Association",
80
+ url = "https://aclanthology.org/2022.lrec-1.416/",
81
+ pages = "3912--3922"
82
+ }
83
+
71
84
  % Joulin et al. (2016)
72
85
  @article{joulin-etal-2016-fasttext,
73
86
  author = {Armand Joulin and Edouard Grave and Piotr Bojanowski and Matthijs Douze and Herv{\'{e}} J{\'{e}}gou and Tom{\'{a}}s Mikolov},
@@ -14,6 +14,7 @@ import ruamel.yaml
14
14
  from . import CLEAN_LOW, CLEAN_HIGH, CLEAN_BETWEEN, CLEAN_TRUE, CLEAN_FALSE
15
15
  from . import OpusFilterError, ConfigurationError
16
16
  from . import filters as filtermodule
17
+ from . import lid as lidmodule
17
18
  from .autogen_cluster import ScoreClusters
18
19
  from .classifier import load_dataframe
19
20
  from .opusfilter import OpusFilter
@@ -171,6 +172,13 @@ def parse_filter_specs(specs):
171
172
  return name, params
172
173
 
173
174
 
175
+ def get_lang_id_filters():
176
+ """Return all language identification filters"""
177
+ names = ['LanguageIDFilter'] # old deprecated filter
178
+ names += [var for var in dir(lidmodule) if var.endswith('Filter')]
179
+ return names
180
+
181
+
174
182
  class AutoFiltersABC(metaclass=abc.ABCMeta):
175
183
  """Abstract base class for automatic filter configuration"""
176
184
 
@@ -181,13 +189,14 @@ class AutoFiltersABC(metaclass=abc.ABCMeta):
181
189
  filters = self.DEFAULT_FILTERS
182
190
  filters = [parse_filter_specs(spec) for spec in filters]
183
191
  self.filters_to_add = []
192
+ lang_id_filters = get_lang_id_filters()
184
193
  for filter_name, filter_params in filters:
185
194
  if filter_name == 'CharacterScoreFilter' and 'scripts' not in filter_params:
186
195
  if not scripts:
187
196
  logger.warning('Cannot add CharacterScoreFilter (no scripts provided)')
188
197
  continue
189
198
  filter_params['scripts'] = scripts
190
- if filter_name == 'LanguageIDFilter' and 'languages' not in filter_params:
199
+ if filter_name in lang_id_filters and 'languages' not in filter_params:
191
200
  if not langs:
192
201
  logger.warning('Cannot add LanguageIDFilter (no languages provided)')
193
202
  continue
@@ -217,7 +226,7 @@ class DefaultParameterFilters(AutoFiltersABC):
217
226
  'AverageWordLengthFilter', 'AlphabetRatioFilter',
218
227
  'TerminalPunctuationFilter', 'NonZeroNumeralsFilter',
219
228
  'LongestCommonSubstringFilter', 'SimilarityFilter', 'RepetitionFilter',
220
- 'CharacterScoreFilter', ('LanguageIDFilter', {'id_method': 'lingua'})]
229
+ 'CharacterScoreFilter', 'LinguaFilter']
221
230
 
222
231
  def set_filter_thresholds(self):
223
232
  """Set filter thresholds"""