红移。我们如何将表（从列到列）转换（动态）？

Question

我们如何将Redshift表从列转换为行？

例如，如果我们有一个通用（未知）表，如下所示：

source table:

date        id      alfa                beta                gamma   ...                 omega
2018-08-03  1       1                   2                   3                           4
2018-08-03  2       4                   3                   2                           1
...
2018-09-04  1       3                   1                   2                           4
...

我们如何才能取得以下成果？

transposed table:

date        id      column_name     column_value
2018-08-03  1       alfa            1
2018-08-03  1       beta            2
...
2018-08-03  2       omega           1
...
2018-09-04  1       gamma           2
...

目标表，列数（alfa，beta，gamma，...，omega）都是动态的（所以我们正在寻找一个解决方案，不需要每列的case when映射，因为我们喜欢将其应用于几个不同的表格）。

但是我们将在所有目标表中具有date和date和id字段（或者最后是所有表中的主键或候选键）。

我们的Redshift版本是：

PostgreSQL 8.0.2, Redshift 1.0.3380

我们怎么做？

Answer 1

您需要将列名硬编码到查询中。

CREATE TABLE stack(date TEXT, id BIGINT, alpha INT, beta INT, gamma INT, omega INT);

INSERT INTO STACK VALUES('2018-08-03', 1, 1, 2, 3, 4);
INSERT INTO STACK VALUES('2018-08-03', 2, 4, 3, 2, 1);
INSERT INTO STACK VALUES('2018-08-04', 1, 3, 1, 2, 4);

SELECT
  date,
  id,
  col,
  col_value
FROM
(
SELECT date, id, alpha AS col_value, 'alpha' AS col FROM stack
UNION
SELECT date, id, beta  AS col_value, 'beta'  AS col FROM stack
UNION
SELECT date, id, gamma AS col_value, 'gamma' AS col FROM stack
UNION
SELECT date, id, omega AS col_value, 'omega' AS col FROM stack
) AS data
ORDER BY date, id, col

结果是：

2018-08-03  1   alpha   1
2018-08-03  1   beta    2
2018-08-03  1   gamma   3
2018-08-03  1   omega   4
2018-08-03  2   alpha   4
2018-08-03  2   beta    3
2018-08-03  2   gamma   2
2018-08-03  2   omega   1
2018-08-04  1   alpha   3
2018-08-04  1   beta    1
2018-08-04  1   gamma   2
2018-08-04  1   omega   4

Answer 2

代替不在评论中提供答案，这里是半伪代码来解释我是如何做到的，如果您需要更多信息/说明，请告诉我

# dictionary to define your target structure
target_d = {'date':'','id':'','column_name':'','column_value':''}

# dictionary for source structure
source_d = {'date':'date','id':'id','column_name1':'','column_name2':''....}

使用上面的这个dict你声明一个字段是否被映射它将不是动态的，所有其他字段/列将被旋转，你可以使用源表DDL将其增强为动态

# assuming you already read your source data
# your while loop to go thru the coming data
while <your code here>
    # create a dict to process an incoming row
    curr_d = target_d.copy()

    curr_d['date'] = date from incoming record
    curr_d['id'] = id from incoming record

    # since we are going to create a row for each column name/value combos 
    # we need a new dict to hold the values

    out_d = curr_d

上面这一行有两个目的，为输出行创建一个新的dict并保留输出行的持久部分（即date和id）

    # rest of the fields are going to be pivoted now
    for afield in source_d:
        if afield not in source_d.values():
            curr_d['column_name'] = afield
            curr_d['column_value'] = column value from incoming record

        create a 'row' from your out_d dict
        write to output/ append to output data frame (if you use a data frame)

虽然循环将通过源行，for循环将为目标的每个列名/值组合创建一个新行

如果这对您有用，请告诉我。