最终示例

用 SQL 进行数据驱动决策

Tim Verdonck

Professor Statistics and Data Science

业务场景

MovieNow 计划投资新影片。
上新片的成本高于老片。
数据分析第一步：
- 客户对新片的评分是否高于老片？
- 各国是否存在差异？

1. 连接数据

需要的信息：

renting 电影租赁记录及评分
customers 客户所属国家

movies 电影发行年份

SELECT *
FROM renting AS r
LEFT JOIN customers AS c
ON c.customer_id = r.customer_id
LEFT JOIN movies AS m
ON m.movie_id = r.movie_id;

2. 选择相关记录

仅使用至少有 4 条评分的电影记录
仅使用 2018-04-01 之后的租赁记录

SELECT *
FROM renting AS r
LEFT JOIN customers AS c
ON c.customer_id = r.customer_id
LEFT JOIN movies AS m
ON m.movie_id = r.movie_id
WHERE r.movie_id IN (
    SELECT movie_id
    FROM renting
    GROUP BY movie_id
    HAVING COUNT(rating) >= 4)
AND r.date_renting >= '2018-04-01';

3. 聚合

聚合类型：

统计租赁次数
统计不同电影数
计算平均评分

聚合层级：

总体聚合
按发行年份
按发行年份并按客户国家分别统计

3. 聚合

SELECT c.country,
       m.year_of_release,
       COUNT(*) AS n_rentals,
       COUNT(DISTINCT r.movie_id) AS n_movies,
       AVG(rating) AS avg_rating
FROM renting AS r
LEFT JOIN customers AS c
ON c.customer_id = r.customer_id
LEFT JOIN movies AS m
ON m.movie_id = r.movie_id
WHERE r.movie_id IN (
    SELECT movie_id
    FROM renting
    GROUP BY movie_id
    HAVING COUNT(rating) >= 4)
AND r.date_renting >= '2018-04-01'
GROUP BY ROLLUP (m.year_of_release, c.country)
ORDER BY c.country, m.year_of_release;

结果表

| year_of_release | country | n_rentals | n_movies | avg_rating         |
|-----------------|---------|-----------|----------|--------------------|
| 2009            | null    | 10        | 1        | 8.7500000000000000 | 
| 2010            | null    | 41        | 5        | 7.9629629629629630 | 
| 2011            | null    | 14        | 2        | 8.2222222222222222 | 
| 2012            | null    | 28        | 5        | 8.1111111111111111 | 
| 2013            | null    | 10        | 2        | 7.6000000000000000 | 
| 2014            | null    | 5         | 1        | 8.0000000000000000 | 
| null            | null    | 333       | 50       | 7.9024390243902439 |

Let's practice!

用 SQL 进行数据驱动决策