HiveSQL——设计一张最近180天的注册、活跃留存表

0 问题描述

现有一个用户活跃表user_active(user_id,active_date)、用户注册表user_regist(user_id,regist_date)，表中分区字段都为dt(yyyy-MM-dd)，用户字段均为user_id; 设计一张 1-180天的注册活跃留存表；表结构如下：

1 数据分析

完整的代码如下：

select
    regist_date,
    diff,
    active_user_cnt,
    case
        when nvl(regis_cnt, 0) != 0 then active_user_cnt / regis_cnt
        end as retention_rate
from (
         select
             t1.regist_date,
             max(t1.regist_count)                     as regis_cnt,
             datediff(t2.active_date, t1.regist_date) as diff,
             count(t2.user_id)                        as active_user_cnt
         from (select
                   user_id,
                   to_date(regist_date)                                    as regist_date,
                   count(user_id) over (partition by to_date(regist_date)) as regist_count
               from user_regist
               where dt >= date_sub(current_date(), 180)) t1
                  left join
              (select
                   user_id,
                   to_date(active_date) as active_date
               from user_active
               where dt >= date_sub(current_date(), 180)
               group by user_id, to_date(active_date)) t2
              on t1.user_id = t2.user_id
         where datediff(active_date, regist_date) >= 1
           and datediff(active_date, regist_date) <= 180
         group by t1.regist_date, datediff(t2.active_date, t1.regist_date)
     ) t3
order by regist_date,
         diff;

上述代码解析：

步骤一：基于注册表，求出用户的注册日期regist_date、每日的用户注册数量regist_count

select
    user_id,
    to_date(regist_date)                                    as regist_date,
    count(user_id) over (partition by to_date(regist_date)) as regist_count
from user_regist
where dt >= date_sub(current_date(), 180);

步骤二：将用户注册表作为主表，关联活跃表，关联键为user_id，一对多的关系，形成笛卡尔积。需要注意：活跃用户表，每个用户每天可能会有多次活跃的情况，因此需要去重。

select
    t1.regist_date,
    t1.user_id,
    t1.regist_count,
    t2.user_id,
    t2.active_date,
    datediff(t2.active_date, t1.regist_date) as diff
from (select
          user_id,
          to_date(regist_date)                                    as regist_date,
          count(user_id) over (partition by to_date(regist_date)) as regist_count
      from user_regist
      where dt >= date_sub(current_date(), 180)) t1
  left join
     (select
          user_id,
          to_date(active_date) as active_date
      from user_active
      where dt >= date_sub(current_date(), 180)
      group by user_id, to_date(active_date)) t2
  on t1.user_id = t2.user_id;

步骤三：基于注册日期，留存周期分组（以“天”为单位），计算该留存周期下的活跃用户数

select
    t1.regist_date,
    max(t1.regist_count)                     as regis_cnt,
    datediff(t2.active_date, t1.regist_date) as diff,
    count(t2.user_id)                        as active_user_cnt

from (select
          user_id,
          to_date(regist_date)                                    as regist_date,
          count(user_id) over (partition by to_date(regist_date)) as regist_count
      from user_regist
      where dt >= date_sub(current_date(), 180)) t1
         left join
     (select
          user_id,
          to_date(active_date) as active_date
      from user_active
      where dt >= date_sub(current_date(), 180)
      group by user_id, to_date(active_date)) t2
     on t1.user_id = t2.user_id
where datediff(active_date, regist_date) >= 1
  and datediff(active_date, regist_date) <= 180
group by t1.regist_date, datediff(t2.active_date, t1.regist_date);

步骤四：计算留存率retention_rate

select
    regist_date,
    diff,
    active_user_cnt,
    case
        when nvl(regis_cnt, 0) != 0 then active_user_cnt / regis_cnt
        end as retention_rate
from (
         select
             t1.regist_date,
             max(t1.regist_count)                     as regis_cnt,
             datediff(t2.active_date, t1.regist_date) as diff,
             count(t2.user_id)                        as active_user_cnt
         from (select
                   user_id,
                   to_date(regist_date)                                    as regist_date,
                   count(user_id) over (partition by to_date(regist_date)) as regist_count
               from user_regist
               where dt >= date_sub(current_date(), 180)) t1
                  left join
              (select
                   user_id,
                   to_date(active_date) as active_date
               from user_active
               where dt >= date_sub(current_date(), 180)
               group by user_id, to_date(active_date)) t2
              on t1.user_id = t2.user_id
         where datediff(active_date, regist_date) >= 1
           and datediff(active_date, regist_date) <= 180
         group by t1.regist_date, datediff(t2.active_date, t1.regist_date)
     ) t3
order by regist_date,
         diff;