[
https://issues.apache.org/jira/browse/LUCENE-9583?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17220400#comment-17220400
]
Julie Tibshirani commented on LUCENE-9583:
------------------------------------------
> Without it, we would need to load all vectors into RAM while
> flushing/merging, as we currently do in BinaryDocValuesWriter.BinaryDVs. I
> wonder if it's worth paying this cost for the simpler API.
This made me notice that we ignore vectors in {{SortingCodecReader}}, which can
be used to sort an index after it's already been created. I opened
https://github.com/apache/lucene-solr/pull/2028 to address this.
I'm not an expert in this code, but to me the trade-off seems worth it for a
well-scoped API. Having a tighter set of methods makes it clear to callers how
vectors are intended to be used: for retrieving docs through kNN or as a
contributor to document scores. And there could still be room for future
optimizations to avoid reloading the vectors? For example, when flushing we
always work with {{BufferedVectorValues}} -- maybe {{SortingVectorValues}}
could take that in directly.
> Another thing I noticed while reviewing this is that I moved the KNN
> {{search(float[] target, int topK, int fanout)}} method from {{VectorValues}}
> to {{VectorValues.RandomAccess}}. This I think we could move back, and handle
> the HNSW requirements for search elsewhere.
This seems like a nice change, no matter what happens with {{RandomAccess}} !
> How should we expose VectorValues.RandomAccess?
> -----------------------------------------------
>
> Key: LUCENE-9583
> URL: https://issues.apache.org/jira/browse/LUCENE-9583
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Michael Sokolov
> Priority: Major
>
> In the newly-added {{VectorValues}} API, we have a {{RandomAccess}}
> sub-interface. [~jtibshirani] pointed out this is not needed by some
> vector-indexing strategies which can operate solely using a forward-iterator
> (it is needed by HNSW), and so in the interest of simplifying the public API
> we should not expose this internal detail (which by the way surfaces internal
> ordinals that are somewhat uninteresting outside the random access API).
> I looked into how to move this inside the HNSW-specific code and remembered
> that we do also currently make use of the RA API when merging vector fields
> over sorted indexes. Without it, we would need to load all vectors into RAM
> while flushing/merging, as we currently do in
> {{BinaryDocValuesWriter.BinaryDVs}}. I wonder if it's worth paying this cost
> for the simpler API.
> Another thing I noticed while reviewing this is that I moved the KNN
> {{search(float[] target, int topK, int fanout)}} method from {{VectorValues}}
> to {{VectorValues.RandomAccess}}. This I think we could move back, and
> handle the HNSW requirements for search elsewhere. I wonder if that would
> alleviate the major concern here?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]