feat(weave): Implement PresidioEntityRecognitionGuardrail
#3575
Conversation
|
New dependencies detected. Learn more about Socket for GitHub ↗︎
|
|
Preview this PR with FeatureBee: https://beta.wandb.ai/?betaVersion=344b359812816bcf5cdb671a02248483b8c5706b |
|
Hi @tssweeney |
| from presidio_anonymizer import AnonymizerEngine | ||
|
|
||
|
|
||
| class PresidioEntityRecognitionResponse(BaseModel): |
There was a problem hiding this comment.
@tcapelle @soumik12345 are you going with TypedDict or BaseModel? Either is fine, but I would pick 1 and be consistent for all the scorers. Maybe you should also have a test to enforce this property
There was a problem hiding this comment.
nit: Prefer the new annotations syntax here
| _analyzer: "AnalyzerEngine" | ||
| _anonymizer: "AnonymizerEngine" | ||
|
|
||
| def __init__( |
There was a problem hiding this comment.
nit: Isn't Scorer a BaseModel? Can you use the pydantic-style init, or do you need to do it this way?
| deny_lists: Optional[dict[str, list[str]]] = None, | ||
| regex_patterns: Optional[dict[str, list[dict[str, str]]]] = None, | ||
| custom_recognizers: Optional[list[Any]] = None, | ||
| show_available_entities: bool = False, |
There was a problem hiding this comment.
This seems weird. Why not just have this as a classmethod or a docs page?
| selected_entities = self.get_available_entities() | ||
|
|
||
| # Get available entities dynamically | ||
| available_entities = self.get_available_entities() |
There was a problem hiding this comment.
there's some duplication with the above available_entities
| self._anonymizer = anonymizer | ||
|
|
||
| @weave.op | ||
| def group_analyzer_results_by_entity_type( |
There was a problem hiding this comment.
is this something that should be an op?
| return "\n".join(explanation_parts) | ||
|
|
||
| @weave.op | ||
| def anonymize_text( |
There was a problem hiding this comment.
not sure if these should all be ops. They seem to be internal helpers?
| return anonymized_text | ||
|
|
||
| @weave.op | ||
| def score(self, output: str) -> PresidioEntityRecognitionResponse: |
There was a problem hiding this comment.
I don't think this signature is correct if you are returning the model dump of it.
| @weave.op | ||
| def score(self, output: str) -> PresidioEntityRecognitionResponse: | ||
| analyzer_results = self._analyzer.analyze( | ||
| text=str(output), entities=self.selected_entities, language=self.language |
There was a problem hiding this comment.
isn't output already str?
Description
Implement the
PresidioEntityRecognitionGuardrailbased on the PresidioEntityRecognitionGuardrail from the safeguards library originally contributed by @ash0ts .Sample Trace