You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Yunfeng Zhou (Jira)" <ji...@apache.org> on 2022/05/23 11:54:00 UTC

[jira] [Updated] (FLINK-27742) Fix Compatibility Issues Between Flink ML Stages

     [ https://issues.apache.org/jira/browse/FLINK-27742?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Yunfeng Zhou updated FLINK-27742:
---------------------------------
    Summary: Fix Compatibility Issues Between Flink ML Stages  (was: Fix Compatibility Issues Between Flink ML Operators.)

> Fix Compatibility Issues Between Flink ML Stages
> ------------------------------------------------
>
>                 Key: FLINK-27742
>                 URL: https://issues.apache.org/jira/browse/FLINK-27742
>             Project: Flink
>          Issue Type: Bug
>          Components: Library / Machine Learning
>    Affects Versions: ml-2.0.0
>            Reporter: Yunfeng Zhou
>            Priority: Critical
>              Labels: pull-request-available
>
> It is discovered that StringIndexer and LogisticRegression in Flink ML cannot be connected in a pipeline. The reason is that the output label column of StringIndexer is integer, while LogisticRegression can only accept input data whose labels are doubles.
> In order to make Flink ML stages compatible with each other, the following changes need to be made.
> - For stages who can only accept double-typed inputs, update their implementation to accept any numerical type.
> - For stages that generates labels as integers, make them return labels as doubles.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)