[ 
https://issues.apache.org/jira/browse/LUCENE-9855?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17314518#comment-17314518
 ] 

Tomoko Uchida commented on LUCENE-9855:
---------------------------------------

Thanks [~julietibs], I think I understand your perspective.
{quote}What should be our strategy to support multiple NN implementations? The 
current design is to have a single format implementation like 
Lucene90VectorsFormat that can handle multiple approaches. But perhaps each 
approach should have its own format implementation instead, like 
Lucene90HnswVectorsFormat or Lucene90ClusterVectorsFormat, etc.
{quote}
I like the idea we have multiple format implementations for each search 
strategy (I suppose they are direct subclasses of the abstract vector format?), 
so that one format writer/reader only need to focus on single algorithm. To 
proceed this idea, in my understanding, current Codec API allows to assign only 
one implementation for one interface (abstract format class) so we may need to 
extend the Codec and perhaps IndexChain, or do you have any idea to switch the 
format variations without modifying the current Codec class...? I'm just 
curious since if we make the Codec API truly pluggable, developers could 
flexibly experiment their codec ideas without taking risks to patch to Lucene 
core; I think it's a really exciting improvement!

I think this issue should focus on the interface for the general vector format 
(to finish the first question). Please feel free to open issues for your second 
point if you'd like, I'll follow them.

> Reconsider codec name VectorFormat
> ----------------------------------
>
>                 Key: LUCENE-9855
>                 URL: https://issues.apache.org/jira/browse/LUCENE-9855
>             Project: Lucene - Core
>          Issue Type: Improvement
>          Components: core/codecs
>    Affects Versions: main (9.0)
>            Reporter: Tomoko Uchida
>            Assignee: Tomoko Uchida
>            Priority: Blocker
>
> There is some discussion about the codec name for ann search.
> https://lists.apache.org/thread.html/r3a6fa29810a1e85779de72562169e72d927d5a5dd2f9ea97705b8b2e%40%3Cdev.lucene.apache.org%3E
> Main points here are 1) use plural form for consistency, and 2) use more 
> specific name for ann search (second point could be optional).
> A few alternatives were proposed:
> - VectorsFormat
> - VectorValuesFormat
> - NeighborsFormat
> - DenseVectorsFormat



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org

Reply via email to