> SQL vector database

Cool, but does it actually return the correct results for these SQL statements, especially when ORDER BY is concerned?

I.e. does it somehow have a way to get a recall of 100% from its indexes?

This page has a fairly helpful breakdown of the various types of indexes available, but from what I can tell it doesn’t explicitly mention the correctness trade offs made when using an index vs a pure table scan. I know in pgvector HNSW for instance is inexact. But it doesn’t really matter in a RAG context - even the embeddings themselves are stochastic.

https://myscale.com/blog/everything-about-vector-indexing/

I'm not sure how much of the underlying Clickhouse feature set is available, but the Clickhouse documentation about index types is thorough and informative.

https://clickhouse.com/docs/en/optimize/sparse-primary-index...

MyScaleDB utilizes approximate nearest neighbors (ANN) algorithms such as ScaNN, HNSW, and IVF. As a result, it may not achieve a 100% recall rate. However, depending on the search parameters used, it can attain recall rates of up to 95% or even 99%.

Considering that embedding vectors represent a lossy compression of the original text or images, is achieving a 100% recall necessary? I am interested in understanding its practical implications.

Disclaimer: I am an employee at MyScale.

> Considering that embedding vectors represent a lossy compression of the original text or images, is achieving a 100% recall necessary?

For the app, maybe not. But as a database absolutist, I think you must be able to dump all rows of a table with

    WITH
      limit_result AS (SELECT *, {similarity} AS metric FROM table ORDER BY {similarity} ASC LIMIT 10),
      dist AS (SELECT MAX(metric) AS max_m FROM limit_result)
    SELECT *, {similarity} AS metric FROM table, dist WHERE {similarity} > dist.max_m
    UNION ALL
    SELECT * FROM limit_result

... assuming that the ordered values are unique across the table and fully sortable

A recall of <100% may skip some rows in the limit_result, which then also won't show up in the main table's scan result, thus potentially corrupting a data dump process that uses sorted output.

I added it to my catalog of RAG retrieval tools. https://shelbyjenkins.github.io/blog/retrieval-is-all-you-ne...

A database that scales to billions is cool. Feels like k8s though. I'm curious who needs it.

Right now I'm interested in in-process retrieval options. We're going to use our own document databases anyways, so having the retrieval database in it's own server just adds an extra layer of complexity.