规范化 MySQL 数据的 DateTime

Question

我正在开发一个汇总财务数据观察的数据库（以货币对 USDCAD 为例）。

财务数据以观察值的形式出现。我没有设置观察的日期和时间，因为它是由数据提供者确定的。

我的桌子是这样的： CREATE TABLE observations (symbol varchar(32) not null, datetime datetime not null, value decimal(20, 10) not null);

当我需要在没有具体观察的情况下获取系列的值时，就会出现问题。例如，我可能想在上午 10:30:15 对我们以加元销售的某种产品重新定价。

但是，我可能不会在上午 10:30:15 准确地进行 USDCAD 观察。这将需要我做一个加权平均才能得到那个确切时间的观察结果。

这有点麻烦，但可能。更大的问题是 - 如果我有一张包含我们产品每秒的美元价格的表格，并且我想以 USDCAD 重新定价每个产品价格观察（每秒）（其中有随机定时的观察），我不能想办法在没有疯狂子查询的情况下做到这一点。

我最好的选择是什么？插入值时，是否也计算 USDCAD 系列的每秒观测值更好？

Answer 1

1. 通过应用linear interpolation 在@time 处获得加权平均观测值：

   SELECT p0.y + (UNIX_TIMESTAMP(@time)-p0.x) * (p1.y-p0.y) / (p1.x-p0.x)
   FROM
     (
       SELECT   `value`                     AS y
         ,       UNIX_TIMESTAMP(`datetime`) AS x
       FROM     `observations`
       WHERE    `datetime` <= @time
       ORDER BY `datetime` DESC
       LIMIT    1
     ) p0,
     (
       SELECT   `value`                     AS y
         ,       UNIX_TIMESTAMP(`datetime`) AS x
       FROM     `observations`
       WHERE    `datetime` >= @time
       ORDER BY `datetime` ASC
       LIMIT    1
     ) p1;
   

2. 使用最新汇率为您的产品创建一个 VIEW 加元价格：

   CREATE VIEW `productsCAD` AS
     SELECT `products`.*
       ,    `products`.`price` * `exchange`.`value` AS `priceCAD`
     FROM   `products`
       , (
         SELECT   `value`
         FROM     `observations`
         ORDER BY `datetime` DESC
         LIMIT    1
       ) AS `exchange`;
   

---

更新

要获取多个预定义时间的插值价格（假设在名为times 的表中名为time 的列中存储为DATETIME 值），您需要使用子查询首先获取时间紧接在之前和之后的观察然后再次将这些与观察表连接以获得相应的值：

SELECT time, p0.y + IFNULL((unixtime-p0.x) * (p1.y-p0.y) / (p1.x-p0.x), 0) AS value
FROM
  (
    SELECT
          time,
          UNIX_TIMESTAMP(`time`)               AS `unixtime`,
      MAX(UNIX_TIMESTAMP(`before`.`datetime`)) AS `prevTime`,
      MIN(UNIX_TIMESTAMP( `after`.`datetime`)) AS `nextTime`
    FROM
      `times`
      JOIN (SELECT `datetime` FROM `observations`)
        AS `before` ON (`before`.`datetime` <= `time`)
      JOIN (SELECT `datetime` FROM `observations`)
        AS `after`  ON ( `after`.`datetime` >= `time`)
    GROUP BY `time`
  ) AS `matches`
  JOIN (
    SELECT
      UNIX_TIMESTAMP(`datetime`) AS x,
      `value`                    AS y
    FROM `observations`
  ) AS p0 ON (p0.x = `matches`.`prevTime`)
  JOIN (
    SELECT
      UNIX_TIMESTAMP(`datetime`) AS x,
      `value`                    AS y
    FROM `observations`
  ) AS p1 ON (p1.x = `matches`.`nextTime`);

如果您想在查询中指定所需的时间（而不是使用预定义的表 times），请将对 times 的引用替换为使用 UNION 动态构建此类表的子查询：

SELECT time, p0.y + IFNULL((unixtime-p0.x) * (p1.y-p0.y) / (p1.x-p0.x), 0) AS value
FROM
  (
    SELECT
          time,
          UNIX_TIMESTAMP(`time`)               AS `unixtime`,
      MAX(UNIX_TIMESTAMP(`before`.`datetime`)) AS `prevTime`,
      MIN(UNIX_TIMESTAMP( `after`.`datetime`)) AS `nextTime`
    FROM
      (
                  SELECT '2012-05-02 19:20:00' AS `time`
        UNION ALL SELECT '2012-05-02 19:20:05'
        UNION ALL SELECT '2012-05-02 19:20:10'
        UNION ALL SELECT '2012-05-02 19:20:15'
        -- etc.
      ) AS `times`
      JOIN (SELECT `datetime` FROM `observations`)
        AS `before` ON (`before`.`datetime` <= `time`)
      JOIN (SELECT `datetime` FROM `observations`)
        AS `after`  ON ( `after`.`datetime` >= `time`)
    GROUP BY `time`
  ) AS `matches`
  JOIN (
    SELECT
      UNIX_TIMESTAMP(`datetime`) AS x,
      `value`                    AS y
    FROM `observations`
  ) AS p0 ON (p0.x = `matches`.`prevTime`)
  JOIN (
    SELECT
      UNIX_TIMESTAMP(`datetime`) AS x,
      `value`                    AS y
    FROM `observations`
  ) AS p1 ON (p1.x = `matches`.`nextTime`);