[
https://issues.apache.org/jira/browse/FLINK-1901?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14704214#comment-14704214
]
ASF GitHub Bot commented on FLINK-1901:
---------------------------------------
Github user ChengXiangLi commented on the pull request:
https://github.com/apache/flink/pull/949#issuecomment-132864761
Thanks for the detail explanation, @tillrohrmann . As the owner of this
issue, make it work correctly and efficiently is my top desire, so i would
never say no to a reasonable and proper suggestion(although i did expect it
happens earlier :-)). `OutOfMemoryError` is one of my concern during the
implementation as well, unlimited usage of Java heap is unacceptable, so i
limit the reservoir size(`PriorityQueue` size in implementation) same as
`numSamples` parameter, which give the user full control of the memory used in
java heap for `sample` operator. It may lead to `OutOfMemoryError` if the
`numSamples` is too large, but other operators may lead to `OutOfMemoryError`
as well if user cache too much objects inside UDF. Anyway, it's just my first
thought, a fully managed memory usage is a better solution, and i would like to
implement that.
As mentioned in the previous comment, is this the next step you prefer?
1. move the `sample` and the `sampleWithSize` methods from the `DataSet` to
the `DataSetUtils` class, and merge this PR.
2. implement `topK` operator in a separate PR.
3. update `sampleWithSize` implementation, and move back to `DataSet`.
> Create sample operator for Dataset
> ----------------------------------
>
> Key: FLINK-1901
> URL: https://issues.apache.org/jira/browse/FLINK-1901
> Project: Flink
> Issue Type: Improvement
> Components: Core
> Reporter: Theodore Vasiloudis
> Assignee: Chengxiang Li
>
> In order to be able to implement Stochastic Gradient Descent and a number of
> other machine learning algorithms we need to have a way to take a random
> sample from a Dataset.
> We need to be able to sample with or without replacement from the Dataset,
> choose the relative or exact size of the sample, set a seed for
> reproducibility, and support sampling within iterations.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
