VSIM

Syntax

VSIM key (ELE | FP32 | VALUES num) vector [WITHSCORES] [WITHATTRIBS] [COUNT num] [EF n] [NOTHREAD] [TRUTH]

Flags: readonly
ACL categories: @read, @slow, @vector
Arity: -2

Returns the elements of the vector set at key most similar to a vector, most similar first.

Options

Option	Effect
`ELE element`	Use an element's vector.
`VALUES n x1 ... xn`, `FP32 blob`	Use this vector.
`COUNT n`	Return at most `n` elements (10 by default).
`WITHSCORES`	Give each element's similarity: `1` is identical.
`WITHATTRIBS`	Give each element's attributes.
`EF n`, `NOTHREAD`, `TRUTH`	Accepted. Every search compares every vector, so results are exact.
`FILTER expr`	Accepted, and ignored in KiviDB 1.1: every element is returned. See kividb#132, fixed in 1.2.

Reply

An array of elements; with WITHSCORES or WITHATTRIBS, each followed by its score or attributes.
ERR element not found in vector set for an ELE that does not exist.

Examples

> VSIM points VALUES 3 1 0 0 WITHSCORES COUNT 2
1) "a"
2) "1.000000"
3) "b"
4) "0.993884"
> VSIM points ELE c WITHATTRIBS COUNT 1
1) "c"
2) "{\"color\":\"red\",\"size\":3}"

Known issue

WITHSCORES gives the cosine similarity itself, from -1 to 1. Redis gives it on a scale from 0 to 1, as (1 + cosine) / 2. Identical vectors score 1 in both, but vectors at right angles score 0 here and 0.5 in Redis, and opposite vectors -1 here and 0 in Redis. The order of the results is the same. Filed as kividb#171.