[
https://issues.apache.org/jira/browse/HIVE-165?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12656222#action_12656222
]
Adam Kramer commented on HIVE-165:
----------------------------------
I agree, and have been annoyed by the inconsistency between POP and SAMP
versions of var().
I have used a similar workaround, but since it (currently) takes two mapreduce
steps, it's faster to write my own single-reducer script. If I was using huge
data sets, the above would be faster, though...but I do worry a bit that for
really huge data sets, SUM(x*x) might overflow.
> var(col) built-in to go with avg(col) and count(col)
> ----------------------------------------------------
>
> Key: HIVE-165
> URL: https://issues.apache.org/jira/browse/HIVE-165
> Project: Hadoop Hive
> Issue Type: Wish
> Reporter: Adam Kramer
> Assignee: David Phillips
> Priority: Minor
>
> The last step in the unholy triumvirate of statistical built-ins is the
> variance. We already have the n (count) and the mean (avg). I currently have
> a job or two that filters all of the data into a single reducer which just
> computes mean/n/variance and writes it to a table...so my guess is that this
> would be a pretty big speed increase. Not a huge deal though, as computing
> the variance myself is trivial.
> (Average, variance, and n can be co-computed in one pass, so if you're doing
> var() you can basically have avg() and count() for free.)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
