[
https://issues.apache.org/jira/browse/HADOOP-18426?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17597018#comment-17597018
]
ASF GitHub Bot commented on HADOOP-18426:
-----------------------------------------
Hexiaoqiao commented on code in PR #4811:
URL: https://github.com/apache/hadoop/pull/4811#discussion_r956979425
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/util/SampleStat.java:
##########
@@ -91,7 +91,7 @@ public SampleStat add(long nSamples, double x) {
}
else {
// The Welford method for numerical stability
- a1 = a0 + (x - a0) / numSamples;
+ a1 = a0 + (x - a0 * nSamples) / numSamples;
Review Comment:
Great catch here. It makes sense to me.
##########
hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/metrics2/util/SampleStat.java:
##########
@@ -117,7 +117,7 @@ public double total() {
* @return the arithmetic mean of the samples
*/
public double mean() {
- return numSamples > 0 ? (total / numSamples) : 0.0;
+ return numSamples > 0 ? a1 : 0.0;
Review Comment:
I am not sure why update here at HADOOP-13804.
@xkrogen would you mind to give another check?
> Improve the accuracy of MutableStat mean
> ----------------------------------------
>
> Key: HADOOP-18426
> URL: https://issues.apache.org/jira/browse/HADOOP-18426
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Shuyan Zhang
> Assignee: Shuyan Zhang
> Priority: Major
> Labels: pull-request-available
>
> The current MutableStat mean calculation method is more prone to loss
> accuracy because the sum of samples is too large. We can process each sample
> on its own to improve mean accuracy.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org
