別名和組通過SAS PROC SQL語句

Question

我與PROC SQL在SAS的工作和我的PROC SQL查詢之一的行爲非常奇怪：

我有一個大的數據集（約1萬行），其看起來是這樣的：

apple_key profit price cost months date  
golden_d  0.03  12  4  3   01/12 
golden_d  0.03  8  0  2   01/12 
granny_s  0.05  15  5  5   02/12 
red_d  0.04  13  0  1   01/12 
golden_d  0.02  1  2  12   03/14 

在我運行下面的查詢這組數據：

%let picking_date = 01/12; /* I simplify here - this part of my code definitely works */ 

proc sql; 
    CREATE TABLE output AS 
    SELECT 
     (CASE apple_key 
       WHEN "golden_d" THEN 1 
       WHEN "granny_s" THEN 2 
       WHEN "red_d" THEN 3 
     END) AS apple_id, 
     apple_key AS apple_name, 
     (CASE WHEN cost= 0 THEN 0 
      ELSE 1 
     END) AS cost_flag, 
     (CASE 
      WHEN CEIL(months/2) < 5 THEN CEIL(months/2) 
      ELSE 5 
     END) AS age, 
     "McDonalds" as farm, 
     sum(profit*price)/sum(price) as price_weighted_profit 
    FROM input_table 
    WHERE date = "&picking_date."d 
     AND price > cost 
     AND cost >= 0 
     AND cost >= 0 
    GROUP BY apple_id, apple_name, cost_flag, age, farm 
    ; 
run; 

當我運行此我GROUP BY聲明不起作用。我爲單個組獲得了一堆條目 （其中apple_id，apple_name，cost_flag，年齡和農場都是相同的，但是我的聚合不起作用）。

但是，當我分別運行GROUP BY（如下所示）時，一切正常。我爲每個組獲得一個具有「價格加權利潤」的條目：

proc sql; 
    CREATE TABLE output_tmp AS 
    SELECT 
     (CASE apple_key 
       WHEN "golden_d" THEN 1 
       WHEN "granny_s" THEN 2 
       WHEN "red_d" THEN 3 
     END) AS apple_id, 
     apple_key AS apple_name, 
     (CASE WHEN cost= 0 THEN 0 
      ELSE 1 
     END) AS cost_flag, 
     (CASE 
      WHEN CEIL(months/2) < 5 THEN CEIL(months/2) 
      ELSE 5 
     END) AS age, 
     "McDonalds" as farm 
    FROM input_table 
    WHERE date = "&picking_date."d 
     AND price > cost 
     AND cost >= 0 
     AND cost >= 0 
    ; 

    CREATE TABLE output AS 
    SELECT 
     apple_id, 
     apple_name, 
     cost_flag, 
     age, 
     farm, 
     sum(profit*price)/sum(price) as price_weighted_profit 
    FROM output_tmp 
    GROUP BY apple_id, apple_name, cost_flag, age, farm 
    ; 
quit; 

爲什麼會發生這種情況？我該如何解決它？這是讓我有點瘋狂......謝謝前面的幫助

Answer 1

它不起作用，因爲group by沒有把總和（利潤*價格）/總和（價格）語句作爲聚合函數。它沒有做到這一點，因爲像年齡的別名，cost_flag等

反正下面是正確的查詢： -

Proc sql; 
    CREATE TABLE output AS 
    SELECT 
      apple_id, 
      apple_name, 
      cost_flag, 
      age, 
      farm, 
      sum(profit*price)/sum(price) as price_weighted_profit 
     FROM 
     (
     SELECT 
      (CASE apple_key 
        WHEN "golden_d" THEN 1 
        WHEN "granny_s" THEN 2 
        WHEN "red_d" THEN 3 
      END) AS apple_id, 
      apple_key AS apple_name, 
      (CASE WHEN cost= 0 THEN 0 
       ELSE 1 
      END) AS cost_flag, 
      (CASE 
       WHEN CEIL(months/2) < 5 THEN CEIL(months/2) 
       ELSE 5 
      END) AS age, 
      "McDonalds" as farm 
     FROM input_table 
     WHERE date = "&picking_date."d 
      AND price > cost 
      AND cost >= 0 
      AND cost >= 0 

     ) a 
     GROUP BY apple_id, apple_name, cost_flag, age, farm; 
     quit; 

讓我知道如果您有任何疑問

Answer 2

經驗法則： - 每當您在select子句中使用任何聚合函數，其餘列應該是group by的一部分。在您發佈的問題中，您發佈的申請是總和（利潤*價格）/總和（價格），但是沒有導致問題的組。

Proc sql; 
    CREATE TABLE output AS 
     SELECT 
      (CASE apple_key 
        WHEN "golden_d" THEN 1 
        WHEN "granny_s" THEN 2 
        WHEN "red_d" THEN 3 
      END) AS apple_id, 
      apple_key AS apple_name, 
      (CASE WHEN cost= 0 THEN 0 
       ELSE 1 
      END) AS cost_flag, 
      (CASE 
       WHEN CEIL(months/2) < 5 THEN CEIL(months/2) 
       ELSE 5 
      END) AS age, 
      "McDonalds" as farm, 
      sum(profit*price)/sum(price) as price_weighted_profit 
     FROM input_table 
     WHERE date = "&picking_date."d 
      AND price > cost 
      AND cost >= 0 
      AND cost >= 0  
     GROUP BY apple_id, apple_name, cost_flag, age, farm; 
     quit; 

Answer 3

我懷疑發生了什麼是remerging。 SAS PROC SQL接受這樣的代碼：

proc sql; 
    select a.*, count(*) 
    from a; 

這並不彙總數據。相反，它將整個計數放在每一行上。換句話說，如果select中的密鑰與group by不完全匹配，則根據group by密鑰計算聚合函數，但結果會放在單獨的行上。其他數據庫使用窗口函數的子集來完成此操作。

在你的情況下，淹沒並不明顯。我認爲存在關鍵的困惑，因爲您在select中使用與原始數據中相同的名稱。我的建議是更改別名，使其明確無誤，並確保group by中的密鑰是明確的。