A way to create vector index using SQL query ?

[ExtReMLapin]

Hello, back from my long sleep.

As of today the docs shows a way to create a HNSW using SQL import and java, but nothing else.

I gave a try with :

CREATE INDEX embedding_index ON EMBEDDING(vector)
ENGINE HNSW WITH {
  "dimensions": 1024,
  "distanceFunction": "Cosine"
}

but it fails with

{
    "error": "Error on transaction commit",
    "detail": "CREATE INDEX embedding_index ON EMBEDDING(vector) ENGINE HNSW WITH {   \"dimensions\": 1024,   \"distanceFunction\": \"Cosine\" }",
    "exception": "com.arcadedb.exception.CommandSQLParsingException"
}

What am I supposed to do to create an index in SQL ?

Thanks

not going to lie, i shoved the whole documentation and arcadedb to google gemini and told him to find me a way to create a vector, that's what it came up with

[odysseaspenta]

The syntax that should work according to the documentation is:

CREATE INDEX on Type(name, vector) HNSW;

This currently does not work. I am working on a fix to provide a temporary solution so that you can create a vector index without importing data from a file. I am seeing this as a temporary solution since the current syntax of the CREATE INDEX SQL statement doesn't provide a way to specify some of the parameters that are needed to build a vector index.

Someone with more knowledge of the code can hopefully chime in.

[ExtReMLapin]

ref

#2529

#2530

[lvca]

This issue is not relevant anymore because of the new LSM Vector implementation: #2529