spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Reynold Xin (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-15598) Change Aggregator.zero to Aggregator.init
Date Fri, 27 May 2016 05:33:12 GMT

    [ https://issues.apache.org/jira/browse/SPARK-15598?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15303528#comment-15303528
] 

Reynold Xin commented on SPARK-15598:
-------------------------------------

Yup I think we'd need to refactor the internals.


> Change Aggregator.zero to Aggregator.init
> -----------------------------------------
>
>                 Key: SPARK-15598
>                 URL: https://issues.apache.org/jira/browse/SPARK-15598
>             Project: Spark
>          Issue Type: Sub-task
>          Components: SQL
>            Reporter: Reynold Xin
>
> org.apache.spark.sql.expressions.Aggregator currently requires defining the zero value
for an aggregator. This is actually a limitation making it difficult to implement APIs such
as reduce. In reduce (or reduceByKey), a single associative and commutative reduce function
is specified by the user, and there is no definition of zero value.
> A small tweak to the API is to change zero to init, taking an input, similar to the following:
> {code}
> abstract class Aggregator[-IN, BUF, OUT] extends Serializable {
>   def init(a: IN): BUF
>   def reduce(b: BUF, a: IN): BUF
>   def merge(b1: BUF, b2: BUF): BUF
>   def finish(reduction: BUF): OUT
> }
> {code}
> Then reduce can be implemented using:
> {code}
> f: (T, T) => T
> new Aggregator[T, T, T] {
>   override def init(a: T): T = identify
>   override def reduce(b: T, a: T): T = f(b, a)
>   override def merge(b1: T, b2: T): T = f(b1, b2)
>   override def finish(reduction: T): T = identify
> }
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org


Mime
View raw message