Hive SQL 计数项目，如果它等于 1，则说明使用了哪个项目

Question

我正在查询使用桌面设备或移动设备的用户，我想知道他们在一天内是否只使用了其中一个或两个，如果他们只使用了一个类别，那是哪个类别。

我编译了下面的查询，但我在循环中运行因为在我计算 'dev_count'.

之前我不能在 GROUP BY 子句中使用 'dev2'

SELECT day, user,
COUNT(DISTINCT CASE 
  WHEN dev = "SmartPhone" OR dev = "Tablet" 
    THEN "Mobile"
  ELSE "Desktop" 
END) AS dev_count,
CASE 
  WHEN dev_count = 1
    THEN
      CASE 
        WHEN dev = "SmartPhone" OR dev = "Tablet" 
          THEN "Mobile"
        ELSE "Desktop" 
        END
  ELSE ""
END AS dev2
FROM table1
WHERE
    ...
GROUP BY day, user, dev2
ORDER BY day;

有办法摆脱它吗？

Answer 1

你能试试这个吗？

我使用 subqry 来计算使用的设备数量，然后在外部查询中我使用 case when to use it to determine its mobile or desktop.

SELECT day, user, 
dev_count,
CASE 
  WHEN dev_count = 1
    THEN
      CASE 
        WHEN dev = "SmartPhone" OR dev = "Tablet" 
          THEN "Mobile"
        ELSE "Desktop" 
        END
  ELSE ""
END AS dev2
FROM 
(SELECT day, user,
COUNT( CASE 
  WHEN dev = "SmartPhone" OR dev = "Tablet" 
    THEN "Mobile"
  ELSE "Desktop" 
END) AS dev_count
,dev
FROM table1
WHERE
    ...
GROUP BY day, user, dev) rs
ORDER BY day;

Answer 2

您可以使用 collect_set() 聚合函数获取 UNIQUE 使用设备的数组 types/etc，然后您可以使用 array_contains() 检查数组包含的内容和大小 ()检查它是否包含多个元素，如果它包含单个元素，则获取数组[0]。

select day, user, 
       size(device_type_used_array) as dev_count,
       case when size(device_type_used_array)=1 
                then device_type_used_array[0] 
           else NULL --more than one, put what you want instead of NULL
       end as one_category_used
from
(
SELECT day, user, 
collect_set(CASE WHEN dev = "SmartPhone" OR
                      dev = "Tablet" 
                THEN "Mobile"
                ELSE "Desktop" 
           END) AS device_type_used_array
FROM table1
WHERE
    ...
GROUP BY day, user
)s
order by day;

device_type_used_array contains unique device type used per day. For not unique array use collect_list(), you can use both if necessary

Hive SQL 计数项目，如果它等于 1，则说明使用了哪个项目

Hive SQL count items and if it equals one tell what item was used

sql

hive

group-by

case

hiveql