You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Shengkai Fang (Jira)" <ji...@apache.org> on 2025/04/03 07:47:00 UTC

[jira] [Comment Edited] (FLINK-34992) FLIP-437: Support ML Models in Flink SQL

    [ https://issues.apache.org/jira/browse/FLINK-34992?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17940575#comment-17940575 ] 

Shengkai Fang edited comment on FLINK-34992 at 4/3/25 7:46 AM:
---------------------------------------------------------------

Hi [~lihaosky] .

Currently, I am working on a PoC (PoC) for the ML_PREDICT feature. However, I have encountered some challenges during the implementation process, particularly regarding the FLIP documentation. Could you please provide further clarification on the following aspects of the FLIP proposal?

1. The FLIP document does not provide guidance on how developers should implement the MODEL entity. It lacks foundational APIs or interfaces analogous to DynamicTableFactory, DynamicTableSource, or LookupRuntimeProvider, which are critical for execution. Could you elaborate on the expected approach for implementing this entity and the required APIs?

2. The FLIP does not clarify which configuration options are critical to the framework versus optional ones. For instance, all examples in the FLIP require defining the "task" option. Is this option mandatory when loading a model, even for remote models whose behavior is already defined in the model file? In my view, remote models might inherit their behavior from the model file itself, making certain framework-level options redundant. Could you clarify the necessity of such parameters and their intended use cases?

Your insights would greatly help me resolve these ambiguities and ensure the PoC aligns with the FLIP’s design intent.

Thank you for your support!


was (Author: fsk119):
[~lihaosky] 

Currently, I am working on a PoC (PoC) for the ML_PREDICT feature. However, I have encountered some challenges during the implementation process, particularly regarding the FLIP documentation. Could you please provide further clarification on the following aspects of the FLIP proposal?

1. The FLIP document does not provide guidance on how developers should implement the MODEL entity. It lacks foundational APIs or interfaces analogous to DynamicTableFactory, DynamicTableSource, or LookupRuntimeProvider, which are critical for execution. Could you elaborate on the expected approach for implementing this entity and the required APIs?


2. The FLIP does not clarify which configuration options are critical to the framework versus optional ones. For instance, all examples in the FLIP require defining the "task" option. Is this option mandatory when loading a model, even for remote models whose behavior is already defined in the model file? In my view, remote models might inherit their behavior from the model file itself, making certain framework-level options redundant. Could you clarify the necessity of such parameters and their intended use cases?

Your insights would greatly help me resolve these ambiguities and ensure the PoC aligns with the FLIP’s design intent.

Thank you for your support!

> FLIP-437: Support ML Models in Flink SQL
> ----------------------------------------
>
>                 Key: FLINK-34992
>                 URL: https://issues.apache.org/jira/browse/FLINK-34992
>             Project: Flink
>          Issue Type: New Feature
>          Components: Table SQL / API, Table SQL / Planner, Table SQL / Runtime
>            Reporter: Hao Li
>            Priority: Major
>
> This is an umbrella task for FLIP-437. FLIP-437: https://cwiki.apache.org/confluence/display/FLINK/FLIP-437%3A+Support+ML+Models+in+Flink+SQL



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