mirror of
https://github.com/data-privacy-stack/presidio.git
synced 2026-09-21 13:38:05 -05:00
f55bf966ff
* fix(analyzer): omitted deny_list_score in YAML registry now defaults to 1.0 CustomRecognizerConfig.deny_list_score defaulted to 0.0 and was always present in model_dump() output, so RecognizerRegistryProvider (YAML/dict registry config) passed deny_list_score=0.0 into PatternRecognizer even when the config said nothing. Any positive score_threshold then filtered out every deny-list match loaded through the provider, while the same recognizer built via PatternRecognizer() directly or RecognizerRegistry.add_pattern_recognizer_from_dict() correctly defaulted to 1.0. Changes the field default to None and overrides model_dump() to exclude None values, matching the existing pattern used by HuggingFaceRecognizerConfig/GLiNERRecognizerConfig/LangExtractRecognizerConfig in the same file. Excluding the key entirely (not just setting it to None) matters here because PatternRecognizer.from_dict() forwards the dumped dict straight into the constructor via cls(**dict) - passing an explicit None would set the score to None outright rather than falling back to the constructor's own default. Explicit deny_list_score values are unaffected. Behavior change: deny-list matches from YAML registry configs without an explicit deny_list_score now score 1.0 instead of 0.0 - called out in docs/analyzer/recognizer_registry_provider.md. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * Simplify: default deny_list_score to 1.0 instead of overriding model_dump Per review feedback (omri374): defaulting the field itself to 1.0 - PatternRecognizer's own default - makes the value identical whether the key is passed explicitly or omitted, so there's no need for the exclude_none override at all. Drops the model_dump() override entirely. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> --------- Co-authored-by: Pradyoth P <pradyoth0@gmail.com> Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>