How do you know the score range of each recognizer?

[lifepillar]

Is there a way to query what scores a recognizer can produce with and without context enhancement? Suppose I want to write my own specialized recognizer, and I want it to beat a given Presidio's recognizer R, except when R's score is boosted by context. For that, I need to know what scores R assigns (base score and, at least, the minimal enhanced score). Is there a way to get such info other than browsing through the code?

[omri374]

Each recognizer has a different way of assigning scores. For regex recognizers (PatternRecognizer) the score is determined by the Pattern score, the context score, and a validation score.
For the pattern score, see the list of patterns used by the recognizer, for example:

presidio/presidio-analyzer/presidio_analyzer/predefined_recognizers/uk_nhs_recognizer.py

Line 19 in d8541e9

PATTERNS = [

For context score, the value is determined by the ContextAwareEnhancer. In the default case, the value is set by context_similarity_factor which is 0.35.

presidio/presidio-analyzer/presidio_analyzer/context_aware_enhancers/lemma_context_aware_enhancer.py

Line 138 in d8541e9

result.score += self.context_similarity_factor

For checksum validation, the score is determined by EntityRecognizer.MAX_SCORE, which is set to 1.0 by default.
All these could be changed for a specific implementation.

To get more insights for every decision, you can return the decision_process output which will contain the score before and after each step :https://microsoft.github.io/presidio/analyzer/decision_process/