多重轉換策略

SQL 中的探索式資料分析

Christina Maimone

Data Scientist

多重轉換

SELECT * FROM naics;

   id   |                  category                   | businesses 
--------+---------------------------------------------+------------
 111110 | Agriculture: Soybean Farming                |       4788
 111130 | Agriculture | Dry Pea and Bean Farming      |       3606
 111140 | Agriculture: Wheat Farming                  |       6393
 111150 | Agriculture - Corn Farming                  |      26469
 111160 | Agriculture: Rice Farming                   |        949
 111199 | Agriculture - All Other Grain Farming       |      15035
 111211 | Agriculture | Potato Farming                |        617
 611110 | Education - Elementary and Secondary        |     187859
 611210 | Education | Junior Colleges                 |       3961
 611310 | Education: Colleges and Universities        |      29148

CASE WHEN

-- 分別處理 :, - 和 |

SELECT CASE WHEN category LIKE '%: %' THEN split_part(category, ': ', 1)

            WHEN category LIKE '% - %' THEN split_part(category, ' - ', 1)

            ELSE split_part(category, ' | ', 1)

        END AS major_category,  -- 為結果取別名

       sum(businesses)          -- 同時選出商家數量

  FROM naics

 GROUP BY major_category;       -- 依上面建立的類別分組

 major_category |  sum   
----------------+--------
 Education      | 220968
 Agriculture    |  57857

重編碼資料表

原始值：fruit 資料表

 customer | fav_fruit 
----------+-----------
      349 | apple
      874 | Apple
      703 | apple
      667 | bannana
      622 | banana
      387 | BANANA
      300 | APPLES
      313 |  apple
      499 |  banana
      418 | apple
      841 | BANANA
      300 | APPLE
      754 | apple

標準化後：recode 資料表

 original | standardized  
----------+--------------
 APPLES   | apple
 apple    | apple  
 Apple    | apple
 bannana  | banana
 apple    | apple
  banana  | banana
 banana   | banana
 APPLE    | apple
  apple   | apple
 BANANA   | banana

步驟 1：CREATE TEMP TABLE

CREATE TEMP TABLE recode AS

  SELECT DISTINCT fav_fruit AS original,  -- 原始且雜亂的值

         fav_fruit AS standardized        -- 新的標準化值

    FROM fruit;

初始資料表

SELECT * 
  FROM recode;

 original | standardized 
----------+--------------
 APPLES   | APPLES
 apple    | apple  
 Apple    | Apple
 bannana  | bannana
 apple    | apple
  banana  |  banana
 banana   | banana
 APPLE    | APPLE
  apple   |  apple
 BANANA   | BANANA
(10 rows)

步驟 2：UPDATE 值

UPDATE table_name
   SET column_name = new_value
 WHERE condition;

步驟 2：UPDATE 值

-- 全部列：轉小寫並去除首尾空白

UPDATE recode 
   SET standardized=trim(lower(original));

-- 特定列：修正拼字錯誤

UPDATE recode
   SET standardized='banana'
 WHERE standardized LIKE '%nn%';

-- 全部列：移除結尾的 s

UPDATE recode
   SET standardized=rtrim(standardized, 's');

重編碼結果

SELECT * 
  FROM recode;

 original | standardized 
----------+--------------
 APPLES   | apple
 apple    | apple
 Apple    | apple
 apple    | apple
  banana  | banana
 banana   | banana
 APPLE    | apple
  apple   | apple
 BANANA   | banana
 bannana  | banana
(10 rows)

步驟 3：JOIN 原始與重編碼表

僅原始值

SELECT fav_fruit, count(*) 
  FROM fruit 
 GROUP BY fav_fruit;

 fav_fruit | count 
-----------+-------
 APPLES    |     1
 apple     |     1
 apple     |     3
  banana   |     1
 BANANA    |     2
  apple    |     1
 APPLE     |     1
 bannana   |     1
 banana    |     1
 Apple     |     1
(10 rows)

含重編碼值

SELECT standardized, 
       count(*) 
  FROM fruit
       LEFT JOIN recode
       ON fav_fruit=original
 GROUP BY standardized;

 standardized | count 
--------------+-------
 apple        |     8
 banana       |     5
(2 rows)

重點回顧

使用 CREATE TEMP TABLE 儲存原始值
用 UPDATE 產生標準化值
將原始資料 JOIN 到標準化資料

清理 Evanston 311 資料！

SQL 中的探索式資料分析