Skip to content

CLUSTER_CENTERS

函数说明

CLUSTER_CENTERS() 函数可用于确定向量列的 K 个聚类中心。返回一行 JSON 数组字符串,包含所有聚类中心。

语法结构

SELECT cluster_centers(col kmeans 'k, op_type, init_type, normalize')  FROM tbl;

参数释义

参数 说明
col 必需的。要确定聚类中心的向量列。
k 必需的。要将数据集分成的簇的数量,大于 0,小于等于总行数。
op_type 必需的。在聚类计算过程中使用的距离函数。目前支持 vector_l2_ops。
init_type 必需的。要使用的初始化聚类中心算法。目前我们支持 random 和 kmeanspluspus(K-means++)。
normalize 必需的。布尔值,要使用的聚类算法,true 代表 Spherical Kmeans,false 代表 Regular Kmeans。

示例

drop table if exists points;
CREATE TABLE points (id int auto_increment PRIMARY KEY,coordinate vecf32(2));
insert into points(coordinate) VALUES
 ("[-7.68905443,6.62034649]"),
 ("[-9.57651383,-6.93440446]"),
 ("[6.82968177,1.1648714]"),
 ("[-2.90130578,7.55077118]"),
 ("[-5.67841327,-7.28818497]"),
 ("[-6.04929137,-7.73619342]"),
 ("[-6.27824322,7.22746302]");
SET GLOBAL experimental_ivf_index = 1;--需设置参数 experimental_ivf_index 值为 1(默认 0)才能使用向量索引
--create index idx_t1 using ivfflat on points(coordinate)  lists=1 op_type "vector_l2_ops";

-- 每个点代表其在 x 和 y 轴上的坐标,查询聚类中心,使用 Regular Kmeans
--K-means++
mysql>  SELECT cluster_centers(coordinate kmeans '2,vector_l2_ops,kmeansplusplus,false') AS centers FROM points;
+----------------------------------------------------+
| centers                                            |
+----------------------------------------------------+
| [ [-2.5097303, 5.640863],[-7.101406, -7.3195944] ] |
+----------------------------------------------------+
1 row in set (0.01 sec)

--KMeans
mysql>  SELECT cluster_centers(coordinate kmeans '2,vector_l2_ops,random,false') AS centers FROM points;
+----------------------------------------------------+
| centers                                            |
+----------------------------------------------------+
| [ [-6.362137, -0.09336702],[6.829682, 1.1648715] ] |
+----------------------------------------------------+
1 row in set (0.00 sec)

-- 每个点代表经纬度坐标,查询聚类中心,使用 Spherical Kmeans
mysql> SELECT cluster_centers(coordinate kmeans '2,vector_l2_ops,kmeansplusplus,true') AS centers FROM points;
+------------------------------------------------------+
| centers                                              |
+------------------------------------------------------+
| [ [0.70710677, 0.70710677],[0.83512634, 0.5500581] ] |
+------------------------------------------------------+
1 row in set (0.00 sec)

--可结合 CROSS JOIN 和 UNNEST 语法将 JSON 类型数据内的聚类中心取出来。
mysql> SELECT value FROM  (
    -> SELECT cluster_centers(coordinate kmeans '2,vector_l2_ops,kmeansplusplus,false') AS centers FROM  points
    -> ) AS subquery 
    -> CROSS JOIN  UNNEST(subquery.centers) AS u;
+-------------------------+
| value                   |
+-------------------------+
| [-2.5097303, 5.640863]  |
| [-7.101406, -7.3195944] |
+-------------------------+
2 rows in set (0.00 sec)