Open pixelogik opened 11 years ago
This would allow to use just the similarity matrix (nearest N for each vector), that can easily be stored in databases (NoSQL, SQL) for lookup. This way applications must not compute nearest neighbours on every request.
What about a graph database output? (Titan, Neo4j, etc)
This would allow to use just the similarity matrix (nearest N for each vector), that can easily be stored in databases (NoSQL, SQL) for lookup. This way applications must not compute nearest neighbours on every request.