[
https://issues.apache.org/jira/browse/CALCITE-5182?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17876911#comment-17876911
]
Mihai Budiu commented on CALCITE-5182:
--------------------------------------
This is a duplicate of [CALCITE-6080]
> STDDEV returning NaN values
> ---------------------------
>
> Key: CALCITE-5182
> URL: https://issues.apache.org/jira/browse/CALCITE-5182
> Project: Calcite
> Issue Type: Bug
> Affects Versions: 1.30.0
> Reporter: Matt P
> Priority: Major
>
> Example w/ Calcite 1.30:
> {code:java}
> 0: jdbc:calcite:model=src/test/resources/mode> SELECT STDDEV(x) FROM (VALUES
> 1+power(2,-52), 1+power(2,-51)) x;
> +--------+
> | EXPR$0 |
> +--------+
> | NaN |
> +--------+
> 1 row selected (0.577 seconds){code}
> This is due to the numerical issues in the formula that is being used:
> {code:java}
> variance = avg(x*x) - avg(x)*avg(x){code}
> Though mathematically this should always be non-negative, due to numerical
> errors it can be negative and thus when the square root is taken to get the
> STDDEV it results in NaN.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
