mikemccand commented on issue #14758: URL: https://github.com/apache/lucene/issues/14758#issuecomment-4253453513
What if we take the duplicate detection out of the picture and offer explicit labels specified by the user during indexing? So the Lucene user could optionally pass `String[]` or so labels to `KnnFloatVectorField`? With no labels (default) passed it does what it does today: the vector is only indexed into the single HNSW graph for the segment. But if user passes labels, it indexes into the single HNSW graph (for all vectors) and also into a separate graph, one per label. And then at search time we could fix the queries to optionally take a label/s? This is maybe more like @msokolov original idea, not sure, but wouldn't that remove some of the complexity of having to change `KnnVectorsWriter` API to iterate through all different vector fields for one document (row stride) vs the column stride the API offers today? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
