Categories:

Vector functions , Aggregate functions

VECTOR_AVG

Computes the element-wise average of vectors in an aggregate. Returns a vector where each element is the average of the corresponding elements across all input vectors. The output is always VECTOR(FLOAT, N) regardless of input type.

See also:

VECTOR_SUM , VECTOR_MIN , VECTOR_MAX , AVG, Vector Embeddings

Syntax

VECTOR_AVG( <vector_column> )

Arguments

vector_column

A column containing VECTOR values. All vectors in the column must have the same element type and dimension.

Returns

Returns a VECTOR(FLOAT, N) value where N is the dimension of the input vectors. Each element in the result vector is the average of the corresponding elements across all input vectors.

Usage notes

  • NULL values are ignored in the aggregation.
  • If all values in the group are NULL, the function returns NULL.
  • All input vectors in the column must have the same dimension and element type.
  • The output is always VECTOR(FLOAT, N) regardless of the input’s type. For information on floating-point numbers in Snowflake, see Floating-point data types.
  • Vector functions are optimized in a way that can reduce floating point precision. This function’s results have a margin of error up to 1e-4.

Examples

This example demonstrates computing the element-wise average of vectors:

CREATE OR REPLACE TABLE vector_data (
  id INT,
  category VARCHAR,
  embedding VECTOR(FLOAT, 3)
);

INSERT INTO vector_data
SELECT 1, 'A', [2.0, 4.0, 6.0]::VECTOR(FLOAT, 3)
UNION ALL SELECT 2, 'A', [4.0, 8.0, 12.0]::VECTOR(FLOAT, 3)
UNION ALL SELECT 3, 'B', [1.0, 2.0, 3.0]::VECTOR(FLOAT, 3)
UNION ALL SELECT 4, 'B', [3.0, 6.0, 9.0]::VECTOR(FLOAT, 3);

-- Compute average for each category
SELECT category, VECTOR_AVG(embedding) AS avg_vector
  FROM vector_data
  GROUP BY category
  ORDER BY category;
+----------+------------------+
| CATEGORY | AVG_VECTOR       |
+----------+------------------+
| A        | [3.0, 6.0, 9.0] |
| B        | [2.0, 4.0, 6.0] |
+----------+------------------+

This example shows scalar aggregation (no GROUP BY):

SELECT VECTOR_AVG(embedding) AS overall_avg
  FROM vector_data;
+------------------+
| OVERALL_AVG      |
+------------------+
| [2.5, 5.0, 7.5]  |
+------------------+

This example shows how integer vectors are converted to float output:

CREATE OR REPLACE TABLE int_vector_data (
  id INT,
  vec VECTOR(INT, 2)
);

INSERT INTO int_vector_data
SELECT 1, [1, 3]::VECTOR(INT, 2)
UNION ALL SELECT 2, [2, 4]::VECTOR(INT, 2);

SELECT VECTOR_AVG(vec) AS avg_result
  FROM int_vector_data;
+-------------+
| AVG_RESULT  |
+-------------+
| [1.5, 3.5]  |
+-------------+