[
https://issues.apache.org/jira/browse/SPARK-55939?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18119761#comment-18119761
]
Szehon Ho commented on SPARK-55939:
-----------------------------------
Removing Fix Version/s 4.2.1 during release cleanup: linked PR #54745 was
closed without merging, and no replacement PR under SPARK-55939 is merged.
Leaving this issue open for future work.
> Add built-in DataSketches ItemsSketch (Frequent Items) functions to Spark SQL
> -----------------------------------------------------------------------------
>
> Key: SPARK-55939
> URL: https://issues.apache.org/jira/browse/SPARK-55939
> Project: Spark
> Issue Type: New Feature
> Components: SQL
> Affects Versions: 4.2.0
> Reporter: Bo Xiong
> Assignee: Bo Xiong
> Priority: Major
> Labels: datasketches, most-frequent, pull-request-available,
> top-k
> Original Estimate: 4h
> Remaining Estimate: 4h
>
> h3. Motivation
> Spark SQL already provides built-in support for several Apache DataSketches
> algorithms:
> * HyperLogLog (HLL) sketches for approximate count distinct
> * Theta sketches for count distinct with set operations (contributed by my
> coworker)
> * Tuple sketches for count distinct with aggregated summaries (contributed
> by my coworker)
> * KLL sketches for approximate quantiles
> This request adds built-in support for the *DataSketches ItemsSketch*
> (Frequent Items), which tracks the approximate frequency of items in a data
> stream. It provides the following features:
> * Frequency estimates for any item (not just the top-K)
> * Configurable error guarantees ({{{}NO_FALSE_POSITIVES{}}} vs
> {{{}NO_FALSE_NEGATIVES{}}})
> * Mergeable binary representations for multi-level rollup aggregation
> h3. Use Cases
> * {*}Frequency estimation in streaming{*}: Maintain running frequency
> sketches that can be merged across micro-batches.
> * {*}Multi-dimensional rollup{*}: Build sketches at child-level dimensions
> and merge up to parent dimensions.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
