Question

原始数据表是

+--------+--------+--------+
| node_1 | node_2 | weight |
+--------+--------+--------+
|      1 |      2 |      5 |
|      1 |      3 |     10 |
|      2 |      1 |     21 |
|      1 |      4 |     15 |
+--------+--------+--------+

这是一个有向加权图。我想要的是将有向图转换为无向图，如

+--------+--------+--------+
| node_1 | node_2 | weight |
+--------+--------+--------+
|      1 |      2 |     26 |
|      1 |      3 |     10 |
|      1 |      4 |     15 |
+--------+--------+--------+

有边（1,3）并不意味着存在边（3,1）。

如何解决这个问题？

Answer 1

看起来您想要对无向节点的权重求和。你的例子不清楚相同节点之间是否有多个有向边（即1-> 2,1-> 2，1-> 2），但在这里我假设它们也应该合并

我没有一个PSQL数据库可以方便地测试它，但这是非常普通的SQL，所以你应该没问题。

免责声明：请先在测试台上使用ROLLBACK进行尝试。不要相信你的真实数据是陌生人在互联网上的建议，特别是如果他们有1个代表： - ）

这里的诀窍是很难对两个可互换的列（node_1，node_2）进行分组，这就是为什么我首先将它们重新排列为不可互换/有序的

BEGIN TRANSACTION

  -- 1a) Make a temporary place to hold our values
  CREATE TEMP TABLE tempnodes (int low_node, int high_node, int weight)
    ON COMMIT DROP;

  -- 1b) consistently write the nodes to temp
  -- making sure node_1 always <= node_2
  -- this ordering helps our grouping in the next step
  INSERT INTO temptable (low_node, high_node, weight)
    SELECT least(node_1, node_2), greatest(node_1, node_2), weight
    FROM oldtable;

  -- 2a) Create a place for our new data
  -- Alternatively, you could truncate the old table
  -- and write the new values back there
  -- however, this way we can go back if/when we make a mistake
  CREATE TABLE newtable(int low_node, int high_node, int summed_weight);

  -- 2b) merge the edges 
  INSERT INTO newtable( low_node, high_node, summed_weight)
    SELECT low_node, high_node, sum(weight)
    FROM temptable
    GROUP BY low_node, high_node;

ROLLBACK TRANSACTION;
--COMMIT TRANSACTION;

Answer 2

如果您不想（或者不需要）实际更改数据，可以通过一次选择来完成此操作：

select least(node_1, node_2) as node_1, 
       greatest(node_2, node_1) as node_2, 
       sum(weight) as weight
from graph
group by least(node_1, node_2), greatest(node_2, node_1)
order by 1,2;

如果您同时需要定向和无向图，则可以将上述内容转换为视图。

与pratik的答案类似，这可以合并为一个更改基础表的语句：

with directed_graph as (
  select least(node_1, node_2) as node_1, 
         greatest(node_2, node_1) as node_2, 
         sum(weight) as weight
  from graph
  group by least(node_1, node_2), greatest(node_2, node_1)
), 
new_graph as (
  update graph 
    set weight = dg.weight
  from directed_graph dg
  where (graph.node_1, graph.node_2) = (dg.node_1, dg.node_2)
  returning graph.*
)
delete from graph 
where not exists (select 1 
                  from new_graph ng
                  where ng.node_1 = graph.node_1 
                    and ng.node_2 = graph.node_2);

这是一个SQLFiddle演示：http://sqlfiddle.com/#!15/ad5b7/1

Answer 3

你需要使用update并删除这两个语句来解决这个问题。

UPDATE test_prit_1 t_1
   SET weight = weight + coalesce((SELECT t_2.weight
                               FROM test_prit_1 t_2
                              WHERE t_2.node_1 = t_1.node_2
                                AND t_2.node_2 = t_1.node_1
                                AND t_2.node_1 > t_2.node_2),
                             0)

然后删除语句以删除额外的行（在给定的示例中node_1 = 2和node_2 = 1）

DELETE FROM test_prit_1 t_1
 WHERE EXISTS (SELECT 1
          FROM test_prit_1 t_2
         WHERE t_2.node_1 = t_1.node_2
           AND t_2.node_2 = t_1.node_1
           AND t_2.node_2 > t_2.node_1)

使用SQL将有向边转换为无向边

3 个答案: