在 MySQL 5.6 中按类别计算滚动百分比变化

Question

我有一个这样的 SQL 数据表，我想计算滚动百分比变化（按行和类别）。所以结果如下所示

我使用的 SQL 查询真的很慢，当有数千个类别时，它需要很长时间才能计算出来，你知道发生了什么吗？或者有什么改进？

首先创建一个示例data_table：

CREATE TABLE IF NOT EXISTS data_table (
    id INT AUTO_INCREMENT,
    num INT,
    category VARCHAR(10),
    price FLOAT(20,2),
    PRIMARY KEY (id)
);

INSERT INTO data_table(num,category,price)
VALUES(1,"A","10"),
      (2,"A","20"),
      (3,"A","30"),
      (1,"B","20"),
      (2,"B","30"),
      (3,"B","40");

用于计算百分比变化的 SQL：

SELECT 
     A.*, 
     CASE WHEN (A.price IS NULL OR B.price IS NULL OR B.price=0) THEN 0 ELSE
        (A.price - B.price)/(B.price) *100 END AS perc
FROM (SELECT
    num,
    category,
    price
  FROM data_table
  ) A LEFT JOIN (SELECT
    num,
    category,
    price
  FROM data_table
  ) B
ON (A.num = B.num+1) AND A.category=B.category;

Answer 1

你可以使用用户变量 -

SELECT
    dt.*,
    IF(@prev_cat <> category, NULL, ROUND((price - @prev_price) / @prev_price * 100, 1)) AS perc,
    @prev_cat := category,
    @prev_price := price
FROM data_table dt, (SELECT @prev_cat := 0, @prev_price := 0) vars
ORDER BY category, num;

如果你想用这个 perc 值更新你的表，你可以使用 -

ALTER TABLE `data_table` 
  ADD COLUMN `perc` DECIMAL(5,2) NULL AFTER `price`;

UPDATE `test`.`data_table` dt
JOIN (
    SELECT
        dt.*,
        IF(@prev_cat <> category, NULL, ROUND((price - @prev_price) / @prev_price * 100, 1)) AS perc_calc,
        @prev_cat := category,
        @prev_price := price
    FROM data_table dt, (SELECT @prev_cat := 0, @prev_price := 0) vars
    ORDER BY category, num
) z ON dt.id = z.id
SET dt.perc = z.perc_calc;

如果您使用的是 MySQL 8，那么使用 LAG() 会更容易一些 -

SELECT dt.*,
    ROUND((price - LAG(price, 1) OVER (PARTITION BY category ORDER BY num ASC)) / LAG(price, 1) OVER (PARTITION BY category ORDER BY num ASC) * 100, 1) AS `prev1`,
    ROUND((price - LAG(price, 2) OVER (PARTITION BY category ORDER BY num ASC)) / LAG(price, 2) OVER (PARTITION BY category ORDER BY num ASC) * 100, 1) AS `prev2`,
    ROUND((price - LAG(price, 20) OVER (PARTITION BY category ORDER BY num ASC)) / LAG(price, 20) OVER (PARTITION BY category ORDER BY num ASC) * 100, 1) AS `prev20`
FROM data_table dt;