JingsongLi opened a new pull request, #11326:
URL: https://github.com/apache/paimon/pull/11326

   ### Purpose
   
   The `weighted_score` hybrid ranker min-max normalizes each route in single 
precision. A route with large finite scores of opposite sign can overflow when 
subtracting the minimum from the maximum. Data Evolution can select the wrong 
Top-K, and primary-key searches can reject the resulting NaN score.
   
   Compute the normalization in double precision when a finite route's span 
overflows. Keep the existing single-precision calculation for representable 
spans.
   
   ### Tests
   
   - Added regressions for Data Evolution and primary-key ranking in Java, and 
for the Python ranker. Each reproduces the incorrect result or exception on the 
baseline.
   - JDK 8 Maven verification passed 110 targeted ranking, vector filtering, 
and full-text cases. All 10 Python hybrid-ranking tests passed with runtime 
warnings treated as errors.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to