LanguageValues and typesStandard typesMaths and data
Vector
On this page
A list of numbers, as an AI model's embedding holds them.
Use it for similarity search. To keep the numbers with the model that made them, use Embedding. For a fixed length, build your own type with dimensions:.
Shape
A list of Decimal.
Example
[0.5, -0.25, 1.0]Good to know
- It holds from 1 to 16,000 numbers.
==tells you whether two are the same.<andlist.sort_bydon't take it.
In JSON
An array of its items, in a request, a response or json.write:
[0.5,-0.25,1][0.0123,-0.00456,0.789][1250.5]
Details
Text form, storage, how it is shown, and operations
Vector is in the group Maths and data, and has been in Jadpo since 0.1.0.
Text form
It has no single text form, so text.from and text.parse don't take it.
How it is stored
It is stored in one column named for the field, of type vector. [0.0123,-0.00456,0.789] is stored as '[0.0123,-0.00456,0.789]'::vector.
How it is shown to people
text.format(value, locale:) shows it to people in their locale.
With locale: Locale.en_gb:
| Value | Shown |
|---|---|
[0.5,-0.25,1] |
[0.5; -0.25; 1] |
[0.0123,-0.00456,0.789] |
[0.0123; -0.00456; 0.789] |
[1250.5] |
[1,250.5] |
Operations
ai.distance(a: Vector, b: Vector) -> Decimal: How far apartaandbare, two vectors of one type, as that type'sdistance:compares them and pgvector works it out: smaller is nearer. Cosine, the default, is from 0 to 2; euclidean from 0; inner product is the negative inner product.
It also takes text.format, json.read, json.write and ==. How every standard type behaves says how they work.
Markdown for agents: /docs/types/type-vector.md.Built from Jadpo main at aee2e8f.