[
https://issues.apache.org/jira/browse/LUCENE-9322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17084204#comment-17084204
]
Tomoko Uchida commented on LUCENE-9322:
---------------------------------------
Hi [~jtibshirani],
thank you for hard working on this!
{code:java}
TopDocs findNearestVectors(float[] queryVector, int k, int recallFactor) throws
new IOException;
{code}
I like this interface, {{recallFactor}} might be an interface for further
flexibility, but it's just an idea.
{quote}Why do we have different implementations of `VectorsFormat`, couldn’t we
just add an enum to the field info like `Strategy.HNSW` and
`Strategy.COARSE_QUANTIZATION`?
{quote}
Personally I would prefer an unified file format for vectors since it is
(theoretically) independent from higher level ANN algorithms. Could we expose
just one "Lucene90VectorsFormat" and low-level I/O, and make only higher logic
(o.a.l.a.index/document/search) to be customizable? Forward iteration is
encouraged anyway...
{quote}What about different distance metrics like angular and L1 distance?
{quote}
JFYI I previously implemented switchable distance function on the HNSW branch,
if you have not noticed it:
[https://github.com/apache/lucene-solr/blob/jira/lucene-9004-aknn-2/lucene/core/src/java/org/apache/lucene/index/VectorValues.java].
It is implemented by enum with {{distance()}} function. Also, I think it would
be good to persist (in the codec) which distance metric we use for the field.
{quote}How exactly is this used in a search? Where are the `Query` classes?
This would be the next part of the API to design/ discuss.
{quote}
We could refer/follow o.a.l.a.index.PointValues's approach, in other words,
concrete field classes with newXXXQuery() methods?
[https://github.com/apache/lucene-solr/blob/master/lucene/core/src/java/org/apache/lucene/index/PointValues.java]
Query part would also need some abstraction and there are many things to be
well thought..., so could we discuss about it in another dedicated issue, to
keep the scope here small ?
> Discussing a unified vectors format API
> ---------------------------------------
>
> Key: LUCENE-9322
> URL: https://issues.apache.org/jira/browse/LUCENE-9322
> Project: Lucene - Core
> Issue Type: New Feature
> Reporter: Julie Tibshirani
> Priority: Major
>
> Two different approximate nearest neighbor approaches are currently being
> developed, one based on HNSW ([#LUCENE-9004]) and another based on coarse
> quantization ([#LUCENE-9136]). Each prototype proposes to add a new format to
> handle vectors. In LUCENE-9136 we discussed the possibility of a unified API
> that could support both approaches. The two ANN strategies give different
> trade-offs in terms of speed, memory, and complexity, and it’s likely that
> we’ll want to support both. Vector search is also an active research area,
> and it would be great to be able to prototype and incorporate new approaches
> without introducing more formats.
> To me it seems like a good time to begin discussing a unified API. The
> prototype for coarse quantization
> ([https://github.com/apache/lucene-solr/pull/1314]) could be ready to commit
> soon (this depends on everyone's feedback of course). The approach is simple
> and shows solid search performance, as seen
> [here|https://github.com/apache/lucene-solr/pull/1314#issuecomment-608645326].
> I think this API discussion is an important step in moving that
> implementation forward.
> The goals of the API would be
> # Support for storing and retrieving individual float vectors.
> # Support for approximate nearest neighbor search -- given a query vector,
> return the indexed vectors that are closest to it.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]