Name
<-> — Renvoie la distance en 2D entre A et B.
Synopsis
double precision <->( geometry A , geometry B );
double precision <->( geography A , geography B );
Description
The <-> operator returns the 2D distance between two geometries. Used in the "ORDER BY" clause provides index-assisted nearest-neighbor result sets with true KNN distance search for geometries, and distance on the sphere for geographies.
![[Note]](../images/note.png)
|
|
|
Cet opérande utilise les index 2D GiST qui peuvent être disponibles sur les géométries. Il est différent des autres opérateurs qui utilisent des index spatiaux en ce sens que l'index spatial n'est utilisé que lorsque l'opérateur est dans la clause ORDER BY. |
|
|
|
|
L'index n'intervient que si l'une des géométries est une constante (pas dans une sous-requête/cte). Par exemple, 'SRID=3005;POINT(1011102 450541)'::geometry au lieu de a.geom |
Reportez-vous à l'atelier PostGIS : La recherche du plus proche voisin pour un exemple détaillé.
Enhanced: 2.2.0 -- True KNN ("K nearest neighbor") behavior for geometry and geography. Note for geography KNN is based on sphere rather than spheroid.
Changed: 2.2.0 -- Old hybrid-syntax workarounds may be slower once true KNN is available. See examples below.
Availability: 2.0.0 -- Weak KNN provides nearest neighbors based on geometry centroid distances instead of true distances. Exact results for points, inexact for all other types.
Exemples
SELECT ST_Distance(geom, 'SRID=3005;POINT(1011102 450541)'::geometry) as d,edabbr, vaabbr
FROM va2005
ORDER BY d limit 10;
d | edabbr | vaabbr
------------------+--------+--------
0 | ALQ | 128
5541.57712511724 | ALQ | 129A
5579.67450712005 | ALQ | 001
6083.4207708641 | ALQ | 131
7691.2205404848 | ALQ | 003
7900.75451037313 | ALQ | 122
8694.20710669982 | ALQ | 129B
9564.24289057111 | ALQ | 130
12089.665931705 | ALQ | 127
18472.5531479404 | ALQ | 002
(10 rows)
Puis la réponse brute KNN :
SELECT st_distance(geom, 'SRID=3005;POINT(1011102 450541)'::geometry) as d,edabbr, vaabbr
FROM va2005
ORDER BY geom <-> 'SRID=3005;POINT(1011102 450541)'::geometry limit 10;
d | edabbr | vaabbr
------------------+--------+--------
0 | ALQ | 128
5541.57712511724 | ALQ | 129A
5579.67450712005 | ALQ | 001
6083.4207708641 | ALQ | 131
7691.2205404848 | ALQ | 003
7900.75451037313 | ALQ | 122
8694.20710669982 | ALQ | 129B
9564.24289057111 | ALQ | 130
12089.665931705 | ALQ | 127
18472.5531479404 | ALQ | 002
(10 rows)
Si vous exécutez "EXPLAIN ANALYZE" sur les deux requêtes, vous constaterez une amélioration des performances pour la seconde.
When a plan cannot produce exact distances directly, use a hybrid query to find the true nearest neighbors. First a CTE query using the index-assisted KNN, then an exact query to get correct ordering:
WITH index_query AS (
SELECT ST_Distance(geom, 'SRID=3005;POINT(1011102 450541)'::geometry) as d,edabbr, vaabbr
FROM va2005
ORDER BY geom <-> 'SRID=3005;POINT(1011102 450541)'::geometry LIMIT 100)
SELECT *
FROM index_query
ORDER BY d limit 10;
d | edabbr | vaabbr
------------------+--------+--------
0 | ALQ | 128
5541.57712511724 | ALQ | 129A
5579.67450712005 | ALQ | 001
6083.4207708641 | ALQ | 131
7691.2205404848 | ALQ | 003
7900.75451037313 | ALQ | 122
8694.20710669982 | ALQ | 129B
9564.24289057111 | ALQ | 130
12089.665931705 | ALQ | 127
18472.5531479404 | ALQ | 002
(10 rows)