[
https://issues.apache.org/jira/browse/DRILL-3429?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14963852#comment-14963852
]
Aman Sinha commented on DRILL-3429:
-----------------------------------
[~mehant] the code change looks good to me. The unit test verifies the window
statistical aggregate results by comparing with the regular aggregates. But if
there is a problem with the regular aggregates..I think it would be better to
compare directly with the expected results.
> DrillAvgVarianceConvertlet may produce wrong results while rewriting stddev,
> variance
> -------------------------------------------------------------------------------------
>
> Key: DRILL-3429
> URL: https://issues.apache.org/jira/browse/DRILL-3429
> Project: Apache Drill
> Issue Type: Bug
> Reporter: Mehant Baid
> Assignee: Aman Sinha
> Priority: Critical
> Fix For: 1.3.0
>
> Attachments: DRILL-3429.patch
>
>
> DrillAvgVarianceConvertlet currently rewrites aggregate functions like avg,
> stddev, variance to simple computations.
> Eg:
> Stddev( x ) => power(
> (sum(x * x) - sum( x ) * sum( x ) / count( x ))
> / count( x ),
> .5)
> Consider the case when the input is an integer. Now the rewrite contains
> multiplication and division, which will bind to functions that operate on
> integers however the expected result should be a double and since double has
> more precision than integer we should be operating on double during the
> multiplication and division.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
