Vector data type

Introduces the vector data type along with the functions and operators available for working with vectors

RisingWave supports a dedicated vector(n) data type to represent fixed-length numerical vectors, similar to the pgvector extension in PostgreSQL. Vectors are commonly used to store embeddings and other high-dimensional feature data.

Define a vector type

vector(n)

n is a positive integer specifying the number of dimensions (elements) in the vector. All elements are stored as real (float32) values, and they must be finite numbers,NULL, NaN, inf, and -inf are not allowed. Vectors are represented in SQL using a bracketed, comma-separated list format such as [1.0, 2.0, 3.0].

Examples

-- Create a table with vector columns
CREATE TABLE embeddings (
    id INT PRIMARY KEY,
    text_content VARCHAR,
    embedding vector(128),  -- 128-dimensional vector
    features vector(3)      -- 3-dimensional vector
);

-- Insert vector data
INSERT INTO embeddings VALUES 
(1, 'Sample text', '[0.1, 0.2, 0.3, ...]', '[1.0, 2.0, 3.0]'),
(2, 'Another text', '[0.4, 0.5, 0.6, ...]', '[4.0, 5.0, 6.0]');

Type casts

Vectors can be converted between other types using casts:

From	To	Cast type
vector(n)	varchar	assign
varchar	vector(n)	explicit
vector(n)	real[]	implicit
int[] / numeric[] / real[] / double precision[]	vector(n)	assign

Distance functions

RisingWave provides built-in operators and functions for measuring distance and similarity between vectors.

Operator / Function	Note
vector(n) <-> vector(n) → double precision / l2_distance(vector(n), vector(n)) → double precision
vector(n) <=> vector(n) → double precision / cosine_distance(vector(n), vector(n)) → double precision
vector(n) <+> vector(n) → double precision / l1_distance(vector(n), vector(n)) → double precision
vector(n) <#> vector(n) → double precision	Negative inner product
inner_product(vector(n), vector(n)) → double precision

Other vector operators

In addition to distance functions, RisingWave supports element-wise operations and vector transformations:

Operator / Function	Description
vector(n) + vector(n) → vector(n)	Element-wise addition
vector(n) - vector(n) → vector(n)	Element-wise subtraction
vector(n) * vector(n) → vector(n)	Element-wise multiplication
vector(m) || vector(n) → vector(m+n)	Concatenate two vectors
vector_norm(vector(n)) → double precision	Euclidean norm
l2_normalize(vector(n)) → vector(n)	Normalize with Euclidean norm