shishaochen commented on issue #24470: [SPARK-27577][MLlib] Correct thresholds downsampled in BinaryClassificationMetrics URL: https://github.com/apache/spark/pull/24470#issuecomment-489418353 @srowen Get your point! Actually, if we choose score of the last element in each chunk as threshold, the calculated Recall, Precision, FMeasure on each threshold are exactly the same as those when no sampling (`numBins=0`). In other words, **they are accurate metrics**. The only difference is the count of thresholds when printing precision/recall/f1 curve compared to downsampling. Thus, why not return the correct metrics of full data set but approximate values?
---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
