[jira] [Commented] (FLINK-31029) KBinsDiscretizer gives wrong bin edges in 'quantile' strategy when input data contains only 2 distinct values

["Dong Lin (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-31029?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17707487#comment-17707487 ] 

Dong Lin commented on FLINK-31029:
----------------------------------

Merged to apache/flink-ml master branch 5dacbd97429a525b0f7e81931f55f3d87f79de57

> KBinsDiscretizer gives wrong bin edges in 'quantile' strategy when input data contains only 2 distinct values
> -------------------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-31029
>                 URL: https://issues.apache.org/jira/browse/FLINK-31029
>             Project: Flink
>          Issue Type: Bug
>          Components: Library / Machine Learning
>            Reporter: Fan Hong
>            Assignee: Zhipeng Zhang
>            Priority: Major
>              Labels: pull-request-available
>
> When one input column contains only 2 distinct values and their counts are same, KBinsDiscretizer transforms this column to all 0s using `quantile` strategy. An example of such column is `[0, 0, 0, 1, 1, 1]`.
> When the 2 distinct values have different counts, the transformed values are also all 0s, which cannot distinguish them.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)