Narrowing type hints, update docstring

[bact]

• Passed code styles and structures
• Passed code linting checks and unit test

[Copilot]

Pull request overview

This PR focuses on tightening type hints and docstrings across PyThaiNLP modules, largely by replacing # type: ignore with explicit typing constructs (cast, NDArray[...], narrowed generics) and adjusting a few global model singletons to cache-based patterns.

Changes:

• Replace many # type: ignore return annotations with cast(...) and more precise types (including numpy.typing.NDArray).
• Refactor some global model singletons into keyed caches (e.g., by device/model) to avoid cross-call state issues.
• Update docstrings and tooling configuration (notably pyproject.toml mypy/tox formatting changes).

Reviewed changes

Copilot reviewed 57 out of 57 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
pythainlp/wsd/core.py	Refactor WSD model handling to a device-keyed cache; tighten typing and return conversions.
pythainlp/word_vector/core.py	Replace ignores with casts; clarify sentence_vectorizer return type and dtype.
pythainlp/wangchanberta/core.py	Replace ignore with cast for tokenizer output typing.
pythainlp/util/thai.py	Reformat constant map; narrow return types for Thai character analysis helpers.
pythainlp/util/pronounce.py	Formatting-only readability change in comprehension.
pythainlp/util/numtoword.py	Broaden bahttext parameter type and update docstring typing.
pythainlp/util/keywords.py	Narrow rank() return type and clarify docstring contract.
pythainlp/util/collate.py	Type collate() input as Iterable[str] and align docstring types.
pythainlp/util/abbreviation.py	Replace return ignore with cast for optional-score tuples.
pythainlp/ulmfit/core.py	Tighten rule/token types, NDArray returns, and float32 conversion behavior.
pythainlp/transliterate/wunsen.py	Replace ignore with cast for transliteration output.
pythainlp/transliterate/w2p.py	Narrow internal NDArray typing; add docstrings for numeric helpers.
pythainlp/transliterate/umt5_thaig2p.py	Replace ignore with cast and structured pipeline output typing.
pythainlp/transliterate/tltk.py	Replace ignores with casts for third-party outputs; minor refactor.
pythainlp/transliterate/thaig2p_v2.py	Replace ignore with cast and structured pipeline output typing.
pythainlp/transliterate/thaig2p.py	Add type: ignore[misc] on torch modules and narrow NDArray generics.
pythainlp/transliterate/thai2rom_onnx.py	Improve IO encoding, NDArray typing, and fix array end-token comparison.
pythainlp/transliterate/thai2rom.py	Add type: ignore[misc] on torch module classes.
pythainlp/transliterate/lookup.py	Replace return ignores with casts for typed optionals/strings.
pythainlp/transliterate/ipa.py	Replace ignores with casts for epitran outputs.
pythainlp/translate/word2word_translate.py	Replace ignore with cast for optional list return.
pythainlp/translate/tokenization_small100.py	Replace ignores with casts; narrow state/vocab typing.
pythainlp/tokenize/newmm.py	Add type parameters to defaultdict graph used in BFS.
pythainlp/tokenize/han_solo.py	Narrow featurizer return types to list[Any] payloads.
pythainlp/tokenize/core.py	Replace return ignore with cast for paragraph tokenization typing.
pythainlp/tag/wangchanberta_onnx.py	Tighten NDArray typing, providers defaulting, and SentencePiece API usage.
pythainlp/tag/tltk.py	Replace ignore with cast for POS tagging output typing.
pythainlp/tag/thainer.py	Add explicit feature dict typing for NER feature extraction.
pythainlp/tag/thai_nner.py	Narrow entity dict typing and expand docstrings/exception chaining.
pythainlp/tag/crfchunk.py	Add a blank line for module formatting consistency.
pythainlp/tag/chunk.py	Add a blank line for module formatting consistency.
pythainlp/tag/_tag_perceptron.py	Simplify saved JSON payload typing to Any.
pythainlp/summarize/keybert.py	Add NDArray typing, float32 conversions, and exception chaining.
pythainlp/summarize/freq.py	Narrow ranking/frequency typing and adjust ranking implementation.
pythainlp/spell/words_spelling_correction.py	Replace import checks with import_module, add docstrings, and add cache.
pythainlp/spell/wanchanberta_thai_grammarly.py	Add type: ignore[misc] on torch module class.
pythainlp/spell/tltk.py	Replace ignore with cast for spell candidate outputs.
pythainlp/spell/phunspell.py	Replace ignore with cast for correction output typing.
pythainlp/soundex/sound.py	Replace ignore with cast for panphon output typing.
pythainlp/soundex/complete_soundex.py	Narrow internal helper signatures and return tuple typing.
pythainlp/phayathaibert/core.py	Tighten callable typing and replace ignore with cast for tokenizer output.
pythainlp/parse/ud_goeswith.py	Rename intermediate variables and reformat vectorized operations for clarity.
pythainlp/generate/wangchanglm.py	Narrow regex pattern typing and replace ignore with cast for decode.
pythainlp/el/core.py	Tighten return types for entity linking results and fix docstring param name.
pythainlp/el/_multiel.py	Tighten EL output typing and replace ignore with cast.
pythainlp/corpus/wordnet.py	Replace ignores with casts and align custom_lemmas return behavior.
pythainlp/corpus/core.py	Replace ignores with casts and adjust minor formatting.
pythainlp/corpus/common.py	Reformat complex condition and dict construction for readability.
pythainlp/coref/core.py	Replace singleton with cache keyed by (model_name, device); tighten return typing.
pythainlp/chunk/crfchunk.py	Replace ignore with cast for tagger outputs.
pythainlp/chunk/init.py	Add a blank line for module formatting consistency.
pythainlp/benchmarks/word_tokenization.py	Tighten stats typing, improve exception chaining, and type NDArray helpers.
pythainlp/augment/wordnet.py	Narrow WordNet augmentation internal list typings.
pythainlp/augment/word2vec/core.py	Minor formatting for long call readability.
pythainlp/augment/word2vec/bpemb_wv.py	Replace ignore with cast for BPEmb tokenizer output.
pythainlp/ancient/currency.py	Narrow return type and fix minor comment formatting.
pyproject.toml	Reformat extras/tox entries; enable mypy strictness; expand ignore-missing-imports modules.

[Copilot]

The bahttext() signature now accepts Optional[float], but when number is None the function returns an empty string. That’s a behavioral change from the previous contract and isn’t documented in the docstring; consider either rejecting None (raise) or explicitly documenting the None -> "" behavior to avoid surprising callers.

[Copilot]

Enabling global mypy strict = true can introduce many new type-checking failures across the project and CI (the repo runs mypy pythainlp in PRs). If this PR isn’t also updating the codebase to be strict-clean, consider scoping strictness via per-module overrides or enabling the stricter flags incrementally.

[Copilot]

__compute_frequencies() declares defaultdict[str, float] but initializes it with defaultdict(int), which produces int defaults and can break type-checking (especially with mypy strict). Consider using a float default factory (e.g., defaultdict(float)) or otherwise ensuring the declared value type matches the factory.

[sonarqubecloud]

Quality Gate passed

Issues
3 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
1.3% Duplication on New Code

See analysis details on SonarQube Cloud