MySQL (MariaDB) 上的第 5 个百分位

Question

我正在尝试使用 ~300k 行表中的订单来查找商品价格的第 95 个百分位（也是最高购买量）。

我已经成功地找到了使用此代码的单件商品的第 95 个百分位和最高购买量：

 SELECT type_id,
       Max(price) AS buy,
       Min(price) AS '95th% buy'
FROM   (SELECT *,
               ( Row_number()
                   OVER (
                     partition BY type_id
                     ORDER BY price DESC) ) AS rownr
        FROM   orderbuffertest AS rownr
        WHERE  is_buy_order = 1
        ORDER  BY ( Row_number()
                      OVER (
                        partition BY type_id
                        ORDER BY price DESC) ) ASC) AS t1
WHERE  t1.type_id = 44992
       AND t1.rownr < (SELECT Count(*)
                       FROM   orderbuffertest
                       WHERE  is_buy_order = 1
                              AND type_id = 44992) * 0.05;  

但是，现在我正在尝试GROUP BY type_id，它破坏了我所有的价值观。

有人知道如何GROUP BY type_id 这个查询吗？甚至可以改进原来的方法？

提前谢谢你，

TheJazzle

附言。这是我的数据库的链接，如果您想弄乱/测试它：https://gofile.io/?c=Ga6ODr

Answer 1

此查询应为您提供所需的结果。它通过price 分配ROW_NUMBER 并计算CTE 中每个type_id 和订单类型（is_buy_order）的所有行，然后选择MAX 价格作为buy 价格（对于@987654330 @)，并且行的最低价格 >= 第 95 个百分位作为第 95 个百分位价格。如果第 95 个百分位数中除最高价格外没有任何行，则返回第二高价格。类似的逻辑适用于sell 和95th%sell 价格的生成：

WITH prices AS (
  SELECT type_id, price, is_buy_order,
         ROW_NUMBER() OVER (PARTITION BY type_id, is_buy_order ORDER BY price DESC) AS rownr,
         COUNT(*) OVER (PARTITION BY type_id, is_buy_order) AS num_rows
  FROM   orderbuffertest
)
SELECT type_id,
       MAX(CASE WHEN is_buy_order = 1 THEN price END) AS buy,
       COALESCE(MIN(CASE WHEN is_buy_order = 1 AND 100.0 * (rownr - 1) / num_rows <= 5 AND rownr != 1 THEN price END), 
                MAX(CASE WHEN is_buy_order = 1 AND rownr = 2 THEN price END)) AS `95th%buy`,
       MIN(CASE WHEN is_buy_order = 0 THEN price END) AS sell,
       COALESCE(MAX(CASE WHEN is_buy_order = 0 AND 100.0 * rownr / num_rows >= 95 AND rownr != num_rows THEN price END), 
                MAX(CASE WHEN is_buy_order = 0 AND rownr = num_rows - 1 THEN price END)) AS `95th%sell`
FROM prices
GROUP BY type_id

如果由于某种原因不能使用 CTE，可以将 CTE 编写为子查询：

SELECT type_id,
       MAX(CASE WHEN is_buy_order = 1 THEN price END) AS buy,
       COALESCE(MIN(CASE WHEN is_buy_order = 1 AND 100.0 * (rownr - 1) / num_rows <= 5 AND rownr != 1 THEN price END), 
                MAX(CASE WHEN is_buy_order = 1 AND rownr = 2 THEN price END)) AS `95th%buy`,
       MIN(CASE WHEN is_buy_order = 0 THEN price END) AS sell,
       COALESCE(MAX(CASE WHEN is_buy_order = 0 AND 100.0 * rownr / num_rows >= 95 AND rownr != num_rows THEN price END), 
                MAX(CASE WHEN is_buy_order = 0 AND rownr = num_rows - 1 THEN price END)) AS `95th%sell`
FROM (
  SELECT type_id, price, is_buy_order,
         ROW_NUMBER() OVER (PARTITION BY type_id, is_buy_order ORDER BY price DESC) AS rownr,
         COUNT(*) OVER (PARTITION BY type_id, is_buy_order) AS num_rows
  FROM   orderbuffertest
) prices
GROUP BY type_id

Demo on dbfiddle