COSINE_SIMILARITY_APPROXIMATE

Description

Calculates cosine similarity between two vectors. NULL arrays or NULL elements raise an error.

Available in Doris 4.2 builds that include this function.Eligible Lance Top-N queries can use native approximate retrieval; other queries execute scalar kernels.

Syntax

COSINE_SIMILARITY_APPROXIMATE(<array1>, <array2>)

Parameters

Parameter	Description
<array1>	Numeric array, implicitly converted to ARRAY; conversion can affect precision.
<array2>	Numeric array, implicitly converted to ARRAY; conversion can affect precision. The two arrays must have the same length.

Return Value

Returns FLOAT; larger means more similar. Scalar evaluation computes dot(x,y) / (norm(x) * norm(y)). If either norm is zero, including two empty arrays, it returns 0.Unequal lengths, NULL arrays, or NULL elements raise an error rather than return NULL.Native candidate eligibility, numerical precision, and NULL/NaN behavior can differ from scalar execution; see automatic search.

Example

SELECT COSINE_SIMILARITY_APPROXIMATE([1,0], [0,1]) AS value;

value
0

SELECT COSINE_SIMILARITY_APPROXIMATE([0,0], [1,0]) AS value;

value
0

SELECT COSINE_SIMILARITY_APPROXIMATE(NULL, [0,1]);

ERROR: First argument for function cosine_similarity_approximate cannot be null

SELECT COSINE_SIMILARITY_APPROXIMATE([0,NULL], [0,1]);

ERROR: First argument for function cosine_similarity_approximate cannot have null