insaf021 opened a new pull request, #354:
URL: https://github.com/apache/commons-math/pull/354

   On master, with `SummaryStatistics` partitions:
   
       aggregate([{1, 2}, {3}])      n=3 min=1.0 max=3.0 mean=2.0 variance=1.0 
sum=6.0
       aggregate([{1, 2}, {}, {3}])  n=3 min=NaN max=NaN mean=NaN variance=NaN 
sum=NaN
   
   `SummaryStatistics` reports NaN for every statistic while it holds no 
values, and `AggregateSummaryStatistics.aggregate` folds each summary in 
unconditionally: `JdkMath.min`/`max` propagate the NaN, and the sum, mean and 
second moment pick it up through the running totals. One empty partition 
anywhere in the collection therefore wipes out the combined result, although 
the javadoc says it should match a single summary over the combined data. Empty 
partitions are routine when work is split across threads or groups. The 
NaN-tolerant min/max removed in MATH-1671 masked part of this; the variance was 
already lost before that.
   
   Summaries with `getN() == 0` are now skipped, both when picking the starting 
summary and in the accumulation loop, and the javadoc says so. An all-empty 
collection still returns n = 0 with NaN statistics, as before.
   
   `AggregateSummaryStatisticsTest.testAggregateEmptySummaries` puts empty 
partitions at the start, middle and end. On master it fails with 
`expected:<5.0> but was:<NaN>`; it passes with the patch. `mvn clean verify 
javadoc:javadoc -pl commons-math-legacy -am` is green on JDK 21 (3264 tests in 
legacy, Checkstyle and SpotBugs clean).
   


-- 
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