检测不一致的数据

在 PostgreSQL 数据库中清理数据

Darryl Reeves, Ph.D.

Industry Assistant Professor, New York University

不一致数据

某些餐馆检查规则

score 对应违规数量 2_4_hazard

camis    |              name               |  score | ... 
---------+---------------------------------+--------+-----
...      | ...                             | ...    | ...                                   
41659848 | LA BRISA DEL CIBAO              | 20     | ...
40961447 | MESON SEVILLA RESTAURANT        | 50     | ...
50063071 | WA BAR                          | 15     | ...
...      | ...                             | ...    | ...

A（0–13），B（14–27），C（28+）
评级情形：
- 初检为 A
- 复检为 A、B 或 C

¹ https://www1.nyc.gov/assets/doh/downloads/pdf/rii/restaurant-grading-faq.pdf

用 SQL 校验规则

相互依赖会引入不一致
规则可用 SQL 编码
A 适用于得分 0–13

SELECT 
    camis, 
    grade, 
    grade_date, 
    score, 
    inspection_type 
FROM 
    restaurant_inspection 
WHERE 
    grade = 'A' AND 
    score NOT BETWEEN 0 AND 13;

用 SQL 校验规则

B 适用于得分 14–27

SELECT 
    camis, 
    grade, 
    grade_date, 
    score, 
    inspection_type 
FROM 
    restaurant_inspection 
WHERE 
    grade = 'B' AND 
    score NOT BETWEEN 14 AND 27;

 camis    | grade | grade_date | score |         inspection_type          
 ---------+-------+------------+-------+----------------------------------
 50034653 | B     | 12/06/2019 | -1    | Cycle Inspection / Re-inspection

用 SQL 校验规则

SELECT 
    camis, grade, grade_date, score, inspection_type FROM 
    restaurant_inspection 
WHERE 
    (grade = 'A' OR grade = 'B' OR grade = 'C') AND
    inspection_type LIKE '%Reopening%';

 camis    | grade | grade_date | score |                 inspection_type                 | ... 
 ---------+-------+------------+-------+-------------------------------------------------+-----
 ...      | ...   | ...        | ...   | ...                                             | ...
 50005784 | C     | 05/29/2019 | 14    | Cycle Inspection / Reopening Inspection         | ...
 50091190 | C     | 07/12/2019 | 7     | Pre-permit (Operational) / Reopening Inspection | ...
 40395023 | C     | 09/13/2019 | 8     | Cycle Inspection / Reopening Inspection         | ...
 50037770 | C     | 10/26/2018 | 11    | Cycle Inspection / Reopening Inspection         | ...
 50036406 | C     | 07/10/2018 | 20    | Cycle Inspection / Reopening Inspection         | ...
 ...      | ...   | ...        | ...   | ...                                             | ...

¹ https://www1.nyc.gov/assets/doh/downloads/pdf/rii/restaurant-grading-faq.pdf

数据清洗要点

方法多样
需仔细权衡
领域知识关键
- 哪些值有效
- 重复的原因
- 合适的填充值

Passons à la pratique !

在 PostgreSQL 数据库中清理数据