Hi Edward, thank you for the detailed description, I'm going through it to understand what to do next. The "rough" survey written by Trey Grainger is really precious, I'm using this as guidance because starting from today I have a week or two to spend developing on this, this could be useful to help the community. And I'm evaluating also https://github.com/o19s/hangry But given the lack of time (at end there are less than ten days), I need to concentrate my efforts only on one project. Again, any further suggestion is appreciated.
On Sat, Jun 20, 2020 at 2:34 PM Edward Ribeiro <edward.ribe...@gmail.com> wrote: > Hi Vincenzo, > > The vector search support in Solr is a work-in-progress with a lot of > discussions scattered among some JIRA issues. > > Start here: https://issues.apache.org/jira/plugins/servlet/mobile#issue > <https://issues.apache.org/jira/plugins/servlet/mobile#issue/SOLR-12890> > /SOLR-12890 > <https://issues.apache.org/jira/plugins/servlet/mobile#issue/SOLR-12890> > > Other discussions (some outdated): > > https://issues.apache.org/jira/plugins/servlet/mobile#issue > <https://issues.apache.org/jira/plugins/servlet/mobile#issue/SOLR-13500> > /SOLR-13500 > <https://issues.apache.org/jira/plugins/servlet/mobile#issue/SOLR-13500> > > https://issues.apache.org/jira/plugins/servlet/mobile#issue > <https://issues.apache.org/jira/plugins/servlet/issue/SOLR-14397 > >/SOLR-14397 > <https://issues.apache.org/jira/plugins/servlet/issue/SOLR-14397> > > Even though cosine similarity can be used to check similarity among > vectors, this solution is not scalable. The state of the art approach for a > large number of vectors has been to use Approximate Nearest Neighbors (ANN) > algorithms like HNSW graph. So, the related support to ANN algorithms in > Lucene has also been discussed: > > https://issues.apache.org/jira/plugins/servlet/mobile#issue > <https://issues.apache.org/jira/plugins/servlet/issue/LUCENE-9004> > /LUCENE-9004 > <https://issues.apache.org/jira/plugins/servlet/issue/LUCENE-9004> > > https://issues.apache.org/jira/plugins/servlet/mobile#issue > <https://issues.apache.org/jira/plugins/servlet/issue/LUCENE-9136> > /LUCENE-9136 > <https://issues.apache.org/jira/plugins/servlet/issue/LUCENE-9136> > > https://issues.apache.org/jira/plugins/servlet/mobile#issue > <https://issues.apache.org/jira/plugins/servlet/issue/LUCENE-9322> > /LUCENE-9322 > <https://issues.apache.org/jira/plugins/servlet/issue/LUCENE-9322> > > > Best, > Edward > > Em sex, 19 de jun de 2020 21:02, Vincenzo D'Amore <v.dam...@gmail.com> > escreveu: > > > Hi all, > > > > I've started to look at image similarity. For each image in documents I > > have a couple of vectors that represent the color and shape similarity. > > > > As pointer to begin a colleague suggested me to start with this project: > > > > https://github.com/moshebla/solr-vector-scoring > > > > but the project seems old and not maintained... On the other hand, > looking > > at Solr Documentation I see there is support for dot product and cosine > > Similarity > > > > > > > https://lucene.apache.org/solr/guide/7_5/vector-math.html#dot-product-and-cosine-similarity > > > > So Solr has the capability to calculate the similarity between two > vectors. > > But it is not clear how to use this native feature when searching. Am I > > missing something? Any help or even suggestion would be appreciated. > > > > Best regards, > > Vincenzo > > > > > > -- > > Vincenzo D'Amore > > > -- Vincenzo D'Amore
