Bigquery:在ST_GEOGPOINT数组中使用ST_CLUSTERDBSCAN

Question

我想使用ST_CLUSTERDBSCAN对地理位置进行聚类。Bigquery页面中的示例如下：

WITH Geos as
  (SELECT 1 as row_id, st_geogfromtext('point empty') as geo UNION ALL
    SELECT 2, st_geogfromtext('multipoint(1 1, 2 2, 4 4, 5 2)') UNION ALL
    SELECT 3, st_geogfromtext('point(14 15)') UNION ALL
    SELECT 4, st_geogfromtext('linestring(40 1, 42 34, 44 39)') UNION ALL
    SELECT 5, st_geogfromtext('polygon((40 2, 40 1, 41 2, 40 2))'))
SELECT row_id, geo, ST_CLUSTERDBSCAN(geo, 1e5, 1) OVER () AS cluster_num FROM
Geos ORDER BY row_id
+--------+-----------------------------------+-------------+
| row_id |                geo                | cluster_num |
+--------+-----------------------------------+-------------+
|      1 |          GEOMETRYCOLLECTION EMPTY |        NULL |
|      2 |    MULTIPOINT(1 1, 2 2, 5 2, 4 4) |           0 |
|      3 |                      POINT(14 15) |           1 |
|      4 |    LINESTRING(40 1, 42 34, 44 39) |           2 |
|      5 | POLYGON((40 2, 40 1, 41 2, 40 2)) |           2 |
+--------+-----------------------------------+-------------+

在我的代码中，我有一个聚合在一起的点数组。然而，它似乎没有工作，即使我在结果中看到的多点。

我的代码：

    ST_CLUSTERDBSCAN(ST_UNION_AGG(buyer_geo_point), 1e4, 2) OVER () AS cluster_num ,
    ST_UNION_AGG(buyer_geo_point)

结果要么为空，要么具有完全错误的值：

null
POINT(-41.5320687976469 -20.3600487114797)

null
MULTIPOINT(-39.0833794 -5.9597183, -39.00682744 -5.73228798)

null
POINT(-40.224447061747 -17.3677128083793)

null
POINT(-40.10711168 -18.08920528)
32  
null
POINT(-41.10854564 -21.47675214)

null
POINT(-51.11207578 -20.64520046)

117
MULTIPOINT(-38.08106136 -11.94490164, -38.06814822 -11.94196154)
    
117
MULTIPOINT(-38.07860266 -11.94484066, -38.0786308 -11.9448231, -38.0787098 -11.9447567, -38.0786912 -11.9447861, -38.0676091 -11.9453678)

null
MULTIPOINT(-39.98731268 -14.8426174, -39.98782804 -14.84623434)

更新:我想出了一个解决方案，让每个集群上的点都有标签。

WITH  merchant_cluster as (SELECT  
        pl_gl.merchant_id, 
        ST_CLUSTERDBSCAN(buyer_geo_point, 1e3, 1) OVER (Partition by merchant_id) as clusters ,
        buyer_geo_point
    FROM `geo-info-table` as geo
    LEFT JOIN `merchants-table` as m on geo.merchant_id = m.user_id
    LEFT JOIN `adresses-table` as add on m.user_id = add.user_id
)
SELECT merchant_id, STRUCT(ARRAY_AGG(IFNULL(clusters,-1)) as cluster_id, ARRAY_AGG(buyer_geo_point) as point) FROM merchant_cluster 
GROUP BY merchant_id

Answer 1

请在下面尝试

select cluster_num, ST_UNION_AGG(buyer_geo_point) geo_cluster 
from (
  select buyer_geo_point,
    ST_CLUSTERDBSCAN(buyer_geo_point, 1e4, 2) OVER () AS cluster_num
  from `project.dataset.table`
)
group by cluster_num

我试图使用你问题中暴露的点来模拟你的数据，并使用上面的代码得到了下面的结果(注意-我使用了ST_CLUSTERDBSCAN(buyer_geo_point，200000，1)，因为点的集合非常小)

​

​

下面是此结果的可视化结果-每个聚类分配了单独的颜色

​

​