SQL窗口函数将选定距离内的值分组

Question

Is there a function in PostgreSQL, that groups rows with similar value? Best would be window function like ST_clusterDBSCAN, which puts together rows within a selected distance. Here is the example:

Group   Value    
A       1    
A       2     
A       2     
A       5    
A       6     
A       10
B       1
B       3

And I am looking for function, that would give me result like this.

SELECT group, value, 
       "FUNCTION"(value, 2) OVER (PARTITION BY group) cluster 
FROM mytable

Where second argument (2) means maximum range between values, that could be in one cluster.

Group   Value   Cluster   
A       1       1 
A       2       1
A       2       1
A       5       2
A       6       2
A       10      3
B       1       1
B       3       1

Answer 1

Try this; it is the approach dnoeth suggested. I'm calling your test data set "temp", and I renamed the group column "agroup". You can change the threshold (the difference you're looking for) by changing the right side of the inequality, and you may want to change the sorting in your real data. BTW, the range unbounded statement is the default for any window, so is not really necessary but I left it in for clarity.

WITH step1 AS (
SELECT t.*, CASE WHEN (value - lag(value,1) OVER w) > 1 THEN 1 ELSE 0 END AS aflag 
FROM temp t WINDOW w AS (ORDER BY agroup,value))
SELECT s.agroup,s.value,sum(aflag) OVER w2 + 1 AS cluster 
FROM step1 s WINDOW w2 AS (PARTITION BY agroup ORDER BY agroup,value RANGE UNBOUNDED PRECEDING)