Implement ANN & SVM

[cappadaan]

Proposal:

Request to implement alternatives to KNN, at least ANN (speed) & SVM (precision).

Checklist:

^{To be completed by the assignee. Check off tasks that have been completed or are not applicable.}

Details

• Implementation completed
• Tests developed
• Documentation updated
• Documentation reviewed
• OpenAPI YAML updated and issue created to rebuild clients

[sanikolaev]

@cappadaan Thanks for the suggestion! We'd like to clarify a few things to better understand the request.

• ANN: our KNN implementation already uses HNSW. Are you looking for a different ANN algorithm, more tuning options for HNSW (speed vs recall), or something else?
• SVM: could you clarify how you expect SVM to be used in Manticore Search (e.g. classification, re-ranking, etc.)? Since SVM isn’t a nearest-neighbor method, a bit more context would help.
• Use case: what problem are you trying to solve — performance, ranking quality, or something else?

[cappadaan]

The request is made because there is simply nothing to choose now. I want to see if my implementations perform better or faster with other algoritms.

• SVM is stated as better in matching (retrieval) then KNN, so I want to check if this is the case for me and if so use it, also if it is slower.
• I would also like to use vector search for autocomplete, but KNN is simply way too slow. ANN could be the solution for that problem.