每組LIMIT-Google BigQuery /標准SQL

Question

我有一個類似以下的表格（ 此處為示例）：

CREATE TABLE topics (
  name varchar(64),
  url varchar(253),
  statistic integer,
  pubdate timestamp
);

INSERT INTO topics VALUES
('a',  'b',  100,  TIMESTAMP '2011-05-16 15:36:38'),  
('a',  'c',  110,  TIMESTAMP '2014-04-01 00:00:00'),  
('a',  'd',  120,  TIMESTAMP '2014-04-01 00:00:00'),  
('a',  'e',  90,   TIMESTAMP '2011-05-16 15:36:38'), 
('a',  'f',  80,   TIMESTAMP '2014-04-01 00:00:00'), 
('a',  'g',  70,   TIMESTAMP '2011-05-16 15:36:38'), 
('a',  'h',  150,  TIMESTAMP '2014-04-01 00:00:00'),  
('a',  'i',  50,   TIMESTAMP '2011-05-16 15:36:38'), 
('b',  'j',  10,   TIMESTAMP '2014-04-01 00:00:00'), 
('b',  'k',  11,   TIMESTAMP '2011-05-16 15:36:38'), 
('b',  'l',  12,   TIMESTAMP '2014-04-01 00:00:00'), 
('b',  'm',  9,    TIMESTAMP '2011-05-16 15:36:38'),
('b',  'n',  8,    TIMESTAMP '2014-04-01 00:00:00'),
('b',  'o',  7,    TIMESTAMP '2011-05-16 15:36:38'),
('b',  'p',  15,   TIMESTAMP '2014-04-01 00:00:00'), 
('b',  'q',  5,    TIMESTAMP '2011-05-16 15:36:38'),
('b',  'r',  2,    TIMESTAMP '2014-04-01 00:00:00')

我想根據它們的statistic值_from每個（ name, date(pubdate) ）組合采用最上面的兩行。

換句話說，我想使用GROUP BY name, date(pubdate) ，但不使用聚合函數，而是根據每個組的statistic簡單地獲取前兩行。 （因此，我知道這實際上不是GROUP BY ，而是“ greatest-n-per-group 。）

我正在將Google Big Query與標准SQL配合使用。 我研究了許多其他解決方案，但不確定在這種情況下如何實現結果。

所需結果：

name    url     statistic   date

a       b       100         2011-05-16
a       e       90          2011-05-16

a       h       150         2014-04-01
a       d       120         2014-04-01

b       m       9           2011-05-16
b       k       11          2011-05-16

b       l       12          2014-04-01
b       p       15          2014-04-01

Answer 1

使用ARRAY_AGG函數：

SELECT
  name,
  DATE(pubdate) AS pubdate,
  ARRAY_AGG(STRUCT(url, statistic) ORDER BY statistic DESC LIMIT 2) AS top_urls
FROM dataset.table
GROUP BY name, pubdate

您可以將子查詢與UNNEST一起使用，以獲取不帶數組的行作為輸出：

SELECT name, pubdate, url, statistic
FROM (
  SELECT
    name,
    DATE(pubdate) AS pubdate,
    ARRAY_AGG(STRUCT(url, statistic) ORDER BY statistic DESC LIMIT 2) AS top_urls
  FROM dataset.table
  GROUP BY name, pubdate
), UNNEST(top_urls)

Answer 2

以下是BigQuery標准SQL

#standardSQL
SELECT * EXCEPT(arr) FROM (
  SELECT name, DATE(pubdate) day, 
    ARRAY_AGG(STRUCT(url, statistic) ORDER BY statistic DESC LIMIT 2) arr
  FROM `project.dataset.table`   
  GROUP BY name, day
), UNNEST(arr)
-- ORDER BY name, day  

您可以使用問題中的示例數據來測試，玩游戲，如以下示例所示

#standardSQL
WITH `project.dataset.table` AS (
  SELECT 'a' name, 'b' url,  100 statistic,  TIMESTAMP '2011-05-16 15:36:38' pubdate UNION ALL  
  SELECT 'a', 'c',  110,  '2014-04-01 00:00:00' UNION ALL  
  SELECT 'a', 'd',  120,  '2014-04-01 00:00:00' UNION ALL  
  SELECT 'a', 'e',  90,   '2011-05-16 15:36:38' UNION ALL 
  SELECT 'a', 'f',  80,   '2014-04-01 00:00:00' UNION ALL 
  SELECT 'a', 'g',  70,   '2011-05-16 15:36:38' UNION ALL 
  SELECT 'a', 'h',  150,  '2014-04-01 00:00:00' UNION ALL  
  SELECT 'a', 'i',  50,   '2011-05-16 15:36:38' UNION ALL 
  SELECT 'b', 'j',  10,   '2014-04-01 00:00:00' UNION ALL 
  SELECT 'b', 'k',  11,   '2011-05-16 15:36:38' UNION ALL 
  SELECT 'b', 'l',  12,   '2014-04-01 00:00:00' UNION ALL 
  SELECT 'b', 'm',  9,    '2011-05-16 15:36:38' UNION ALL
  SELECT 'b', 'n',  8,    '2014-04-01 00:00:00' UNION ALL
  SELECT 'b', 'o',  7,    '2011-05-16 15:36:38' UNION ALL
  SELECT 'b', 'p',  15,   '2014-04-01 00:00:00' UNION ALL 
  SELECT 'b', 'q',  5,    '2011-05-16 15:36:38' UNION ALL
  SELECT 'b', 'r',  2,    '2014-04-01 00:00:00' 
)
SELECT * EXCEPT(arr) FROM (
  SELECT name, DATE(pubdate) day, 
    ARRAY_AGG(STRUCT(url, statistic) ORDER BY statistic DESC LIMIT 2) arr
  FROM `project.dataset.table`  
  GROUP BY name, day
), UNNEST(arr)
ORDER BY name, day   

結果

Row name    day         url statistic    
1   a       2011-05-16  b   100  
2   a       2011-05-16  e   90   
3   a       2014-04-01  h   150  
4   a       2014-04-01  d   120  
5   b       2011-05-16  k   11   
6   b       2011-05-16  m   9    
7   b       2014-04-01  p   15   
8   b       2014-04-01  l   12   

Answer 3

    with xx as(
      select name, url, statistic, pubdate, row_number() over(partition by name , url order by statistic desc) rn 
      from topics)
select * except(rn) 
from xx 
where rn <= 2;