在日期字段的帮助下应用滑动 window 函数
Applying sliding window function with the help of a date field
我正在尝试在一天内计算不同的 ID，这很简单，但是如何在 window 中从当前日期算起 7 天后，在当前 window 中计算 30 天后.
我从当前行向后滑动 7/30 行，但我认为它可能 more/less 比 7/30 天不准确，因为我认为它只是按行数滑动那么你有什么处理这个的想法吗？
--mockup data SnowSQL
with test (date,id,a,b,c,d) as(
select as date, as id, as a, as b, as c, as d
from values
('2019-07-15','x_1','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','CA','NA'),
('2019-07-16','x_2','ps','e','CA','NA'),
('2019-07-16','x_3','c','xb','CH','AS'),
('2019-07-17','x_4','ps','e','US','NA'),
('2019-07-17','x_5','c','ps4','CH','AS'),
('2019-07-17','x_6','c','ps4','CH','AS'),
('2019-07-17','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-19','x_8','c','ps','CH','AS'),
('2019-07-20','x_8','c','ps','CH','AS'),
('2019-07-21','x_8','c','ps','CH','AS'),
('2019-07-22','x_8','c','ps','CH','AS'),
('2019-07-29','x_8','c','ps','CH','AS'),
('2019-08-01','x_8','c','ps','CH','AS'),
('2019-08-02','x_9','c','ps','CH','AS'),
('2019-08-03','y_1','c','ps','CH','AS'),
('2019-08-04','y_8','c','ps','CH','AS'),
('2019-08-05','z_8','c','ps','CH','AS'),
('2019-08-06','a_8','c','ps','CH','AS'),
('2019-08-07','b_8','c','ps','CH','AS'),
('2019-08-07','c_8','c','ns','CH','AS'),
('2019-08-07','e_8','ps','e','US','NA'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-10','gx_8','c','ps','CH','AS'),
('2019-08-11','v_8','c','ps','CH','AS') )
--subquery
select 1 as part,
(
select sum(x)
from (
select count(distinct id) as x
from test
where
date = '2019-07-16'
group by a,b,c,d) ) as dau,
(
select sum(x)
from (
select count(distinct id) as x
from test
where
date between '2019-07-16'::date-interval '7 days' and '2019-07-16'
// <= '2019-07-15'
//and date >= '2019-07-15'::date-interval '7 days'
group by a,b,c,d)) as w,
(select sum(x)
from (
select count(distinct id) as x
from test
where
date <= '2019-07-16'
and date >= '2019-07-16'::date-interval '30 days'
group by a,b,c,d)) as m
union
--window function
select 2,sum(dau),sum(w),sum(m)
from(
select
date,
a,
b,
c,
d,
count(distinct id) over (Partition by date,a,b,c,d Order by date)as dau,
count(distinct id) over (Partition by a,b,c,d Order by date rows between 7 preceding and current row) as w,
count(distinct id) over (Partition by a,b,c,d Order by date rows between 30 preceding and current row) as m
from test
group by
date,
a,
b,
c,
d,
id)
where date='2019-07-16'
group by date
;
第 1 部分通过子查询在时间范围内进行计数
第二部分使用滑动函数进行计数
预期结果：两者应该在相同的时间范围内获得相同的计数。
如果我对问题的理解正确，您可以像这样在 window 函数中使用范围参数（为简单起见，我将您的 table 值转储到临时 table 中）：
select distinct [date], id
from #test
order by [date] desc
select [date],
count(*) over (order by date desc rows between current row and 7 following ) count_7_day,
count(*) over (order by date desc rows between current row and 30 following ) count_30_day
from (select distinct [date], id
from #test )x
order by [date] desc
希望这对您有所帮助？
日期在 table
create table test as(
select as date, as id, as a, as b, as c, as d
from values
('2019-07-15','x_1','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','CA','NA'),
('2019-07-16','x_2','ps','e','CA','NA'),
('2019-07-16','x_3','c','xb','CH','AS'),
('2019-07-17','x_4','ps','e','US','NA'),
('2019-07-17','x_5','c','ps4','CH','AS'),
('2019-07-17','x_6','c','ps4','CH','AS'),
('2019-07-17','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-19','x_8','c','ps','CH','AS'),
('2019-07-20','x_8','c','ps','CH','AS'),
('2019-07-21','x_8','c','ps','CH','AS'),
('2019-07-22','x_8','c','ps','CH','AS'),
('2019-07-29','x_8','c','ps','CH','AS'),
('2019-08-01','x_8','c','ps','CH','AS'),
('2019-08-02','x_9','c','ps','CH','AS'),
('2019-08-03','y_1','c','ps','CH','AS'),
('2019-08-04','y_8','c','ps','CH','AS'),
('2019-08-05','z_8','c','ps','CH','AS'),
('2019-08-06','a_8','c','ps','CH','AS'),
('2019-08-07','b_8','c','ps','CH','AS'),
('2019-08-07','c_8','c','ns','CH','AS'),
('2019-08-07','e_8','ps','e','US','NA'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-10','gx_8','c','ps','CH','AS'),
('2019-08-11','v_8','c','ps','CH','AS') );
我过去这样做的方式是，我想将不同的一天绘制成行，从而将结果作为不同的行，以便按天数列按颜色显示我这样解决这个问题：
with distinct_data as (
select distinct date, id
from test
), days as ( -- set of days we want to range over
select column1 as days from (values (1),(7),(30))
), windows_ends as ( -- last 60 days of date
select dateadd('day',-SEQ8(),current_date) as win_date_end
from table(generator(rowcount => 60))
), windows as (
select d.days
,w.win_date_end
,dateadd('day',-d.days, w.win_date_end) as win_date_start
from days as d
join windows_ends as w
)
select w.days
,w.win_date_start
,w.win_date_end
,count(distinct(d.id)) as c_id
from distinct_data as d
join windows as w on d.date > w.win_date_Start and d.date <= w.win_date_end
group by 1,2,3
order by 3,1;
对于我的结果，我删除了 win_date_start 在我的数据开始之前的行，所以我不会得到数据集开始后 2 天开始的 30 天。
以上允许动态天数（我发现这非常有用）但是在具有三个固定结果的情况下，我只是添加了一个枢轴，因此..
with distinct_data as (
select distinct date, id
from test
), days as ( -- set of days we want to range over
select column1 as days from (values (1),(7),(30))
), windows_ends as ( -- last 60 days of date
select dateadd('day',-SEQ8(),current_date) as win_date_end
from table(generator(rowcount => 60))
), windows as (
select d.days
,w.win_date_end
,dateadd('day',-d.days, w.win_date_end) as win_date_start
from days as d
join windows_ends as w
), manyrows as (
select w.days
,w.win_date_end
,count(distinct(d.id)) as c_id
from distinct_data as d
join windows as w on d.date > w.win_date_Start and d.date <= w.win_date_end
group by 1,2
)
select win_date_end as date, "1" as d1, "7" as d7, "30" as d30
from manyrows
pivot(sum(c_id) for days in (1,7,30)) as p
order by win_date_end;
但您可以看到结果如何对最后日期产生尾随效应...我倾向于过滤掉...
这也需要更改 60 from now 过滤器
鉴于您固定的 30 天 window，我觉得这可以更有效地完成，就像这样
with distinct_data as (
select distinct date
,id
from test
), distinct_dates as (
select distinct date
,dateadd('day',-30,date) as d30_s
from distinct_data
), mixed_data as (
select d.date
,b.id
,datediff('days',b.date, d.date) as d_diff
,iff(d_diff < 1, id, NULL) as d1_id
,iff(d_diff < 7, id, NULL) as d7_id
,iff(d_diff < 30, id, NULL) as d30_id
from distinct_dates as d
join distinct_data as b on b.date >= d.d30_s and b.date <= d.date
)
select date
,count(distinct(d1_id)) as d1
,count(distinct(d7_id)) as d7
,count(distinct(d30_id)) as d30
from mixed_data
group by 1 order by 1;
但这只给出了 window 天数的总和，因此前两种方法为 2019-7-23 给出了 7 天和 30 天，因为后者没有那一天。所以这取决于您对数据的需求。
如果我们使用当前行，除非您的情况只有 1 行，否则我们将不会获得异常输出，因此如果该功能可用，我们应该使用按范围滑动。如果没有将您的数据与日期维度结合起来并为您的滑动 window 填充数据，并使用案例语句来获取您的 windows（如果您有多个 windows）。现在我们有了所需的数据，在案例上使用常规 window 函数并对其应用计数。
我正在尝试在一天内计算不同的 ID，这很简单，但是如何在 window 中从当前日期算起 7 天后，在当前 window 中计算 30 天后.
我从当前行向后滑动 7/30 行，但我认为它可能 more/less 比 7/30 天不准确，因为我认为它只是按行数滑动那么你有什么处理这个的想法吗？
--mockup data SnowSQL
with test (date,id,a,b,c,d) as(
select as date, as id, as a, as b, as c, as d
from values
('2019-07-15','x_1','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','CA','NA'),
('2019-07-16','x_2','ps','e','CA','NA'),
('2019-07-16','x_3','c','xb','CH','AS'),
('2019-07-17','x_4','ps','e','US','NA'),
('2019-07-17','x_5','c','ps4','CH','AS'),
('2019-07-17','x_6','c','ps4','CH','AS'),
('2019-07-17','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-19','x_8','c','ps','CH','AS'),
('2019-07-20','x_8','c','ps','CH','AS'),
('2019-07-21','x_8','c','ps','CH','AS'),
('2019-07-22','x_8','c','ps','CH','AS'),
('2019-07-29','x_8','c','ps','CH','AS'),
('2019-08-01','x_8','c','ps','CH','AS'),
('2019-08-02','x_9','c','ps','CH','AS'),
('2019-08-03','y_1','c','ps','CH','AS'),
('2019-08-04','y_8','c','ps','CH','AS'),
('2019-08-05','z_8','c','ps','CH','AS'),
('2019-08-06','a_8','c','ps','CH','AS'),
('2019-08-07','b_8','c','ps','CH','AS'),
('2019-08-07','c_8','c','ns','CH','AS'),
('2019-08-07','e_8','ps','e','US','NA'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-10','gx_8','c','ps','CH','AS'),
('2019-08-11','v_8','c','ps','CH','AS') )
--subquery
select 1 as part,
(
select sum(x)
from (
select count(distinct id) as x
from test
where
date = '2019-07-16'
group by a,b,c,d) ) as dau,
(
select sum(x)
from (
select count(distinct id) as x
from test
where
date between '2019-07-16'::date-interval '7 days' and '2019-07-16'
// <= '2019-07-15'
//and date >= '2019-07-15'::date-interval '7 days'
group by a,b,c,d)) as w,
(select sum(x)
from (
select count(distinct id) as x
from test
where
date <= '2019-07-16'
and date >= '2019-07-16'::date-interval '30 days'
group by a,b,c,d)) as m
union
--window function
select 2,sum(dau),sum(w),sum(m)
from(
select
date,
a,
b,
c,
d,
count(distinct id) over (Partition by date,a,b,c,d Order by date)as dau,
count(distinct id) over (Partition by a,b,c,d Order by date rows between 7 preceding and current row) as w,
count(distinct id) over (Partition by a,b,c,d Order by date rows between 30 preceding and current row) as m
from test
group by
date,
a,
b,
c,
d,
id)
where date='2019-07-16'
group by date
;
第 1 部分通过子查询在时间范围内进行计数
第二部分使用滑动函数进行计数
预期结果：两者应该在相同的时间范围内获得相同的计数。
如果我对问题的理解正确，您可以像这样在 window 函数中使用范围参数（为简单起见，我将您的 table 值转储到临时 table 中）：
select distinct [date], id
from #test
order by [date] desc
select [date],
count(*) over (order by date desc rows between current row and 7 following ) count_7_day,
count(*) over (order by date desc rows between current row and 30 following ) count_30_day
from (select distinct [date], id
from #test )x
order by [date] desc
希望这对您有所帮助？
日期在 table
create table test as(
select as date, as id, as a, as b, as c, as d
from values
('2019-07-15','x_1','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','US','NA'),
('2019-07-15','x_2','ps','e','CA','NA'),
('2019-07-16','x_2','ps','e','CA','NA'),
('2019-07-16','x_3','c','xb','CH','AS'),
('2019-07-17','x_4','ps','e','US','NA'),
('2019-07-17','x_5','c','ps4','CH','AS'),
('2019-07-17','x_6','c','ps4','CH','AS'),
('2019-07-17','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-18','x_7','c','ns','CH','AS'),
('2019-07-19','x_8','c','ps','CH','AS'),
('2019-07-20','x_8','c','ps','CH','AS'),
('2019-07-21','x_8','c','ps','CH','AS'),
('2019-07-22','x_8','c','ps','CH','AS'),
('2019-07-29','x_8','c','ps','CH','AS'),
('2019-08-01','x_8','c','ps','CH','AS'),
('2019-08-02','x_9','c','ps','CH','AS'),
('2019-08-03','y_1','c','ps','CH','AS'),
('2019-08-04','y_8','c','ps','CH','AS'),
('2019-08-05','z_8','c','ps','CH','AS'),
('2019-08-06','a_8','c','ps','CH','AS'),
('2019-08-07','b_8','c','ps','CH','AS'),
('2019-08-07','c_8','c','ns','CH','AS'),
('2019-08-07','e_8','ps','e','US','NA'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-08','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-09','f_8','c','xb','CH','AS'),
('2019-08-10','gx_8','c','ps','CH','AS'),
('2019-08-11','v_8','c','ps','CH','AS') );
我过去这样做的方式是，我想将不同的一天绘制成行，从而将结果作为不同的行，以便按天数列按颜色显示我这样解决这个问题：
with distinct_data as (
select distinct date, id
from test
), days as ( -- set of days we want to range over
select column1 as days from (values (1),(7),(30))
), windows_ends as ( -- last 60 days of date
select dateadd('day',-SEQ8(),current_date) as win_date_end
from table(generator(rowcount => 60))
), windows as (
select d.days
,w.win_date_end
,dateadd('day',-d.days, w.win_date_end) as win_date_start
from days as d
join windows_ends as w
)
select w.days
,w.win_date_start
,w.win_date_end
,count(distinct(d.id)) as c_id
from distinct_data as d
join windows as w on d.date > w.win_date_Start and d.date <= w.win_date_end
group by 1,2,3
order by 3,1;
对于我的结果，我删除了 win_date_start 在我的数据开始之前的行，所以我不会得到数据集开始后 2 天开始的 30 天。
以上允许动态天数（我发现这非常有用）但是在具有三个固定结果的情况下，我只是添加了一个枢轴，因此..
with distinct_data as (
select distinct date, id
from test
), days as ( -- set of days we want to range over
select column1 as days from (values (1),(7),(30))
), windows_ends as ( -- last 60 days of date
select dateadd('day',-SEQ8(),current_date) as win_date_end
from table(generator(rowcount => 60))
), windows as (
select d.days
,w.win_date_end
,dateadd('day',-d.days, w.win_date_end) as win_date_start
from days as d
join windows_ends as w
), manyrows as (
select w.days
,w.win_date_end
,count(distinct(d.id)) as c_id
from distinct_data as d
join windows as w on d.date > w.win_date_Start and d.date <= w.win_date_end
group by 1,2
)
select win_date_end as date, "1" as d1, "7" as d7, "30" as d30
from manyrows
pivot(sum(c_id) for days in (1,7,30)) as p
order by win_date_end;
但您可以看到结果如何对最后日期产生尾随效应...我倾向于过滤掉...
这也需要更改 60 from now 过滤器
鉴于您固定的 30 天 window，我觉得这可以更有效地完成，就像这样
with distinct_data as (
select distinct date
,id
from test
), distinct_dates as (
select distinct date
,dateadd('day',-30,date) as d30_s
from distinct_data
), mixed_data as (
select d.date
,b.id
,datediff('days',b.date, d.date) as d_diff
,iff(d_diff < 1, id, NULL) as d1_id
,iff(d_diff < 7, id, NULL) as d7_id
,iff(d_diff < 30, id, NULL) as d30_id
from distinct_dates as d
join distinct_data as b on b.date >= d.d30_s and b.date <= d.date
)
select date
,count(distinct(d1_id)) as d1
,count(distinct(d7_id)) as d7
,count(distinct(d30_id)) as d30
from mixed_data
group by 1 order by 1;
但这只给出了 window 天数的总和，因此前两种方法为 2019-7-23 给出了 7 天和 30 天，因为后者没有那一天。所以这取决于您对数据的需求。
如果我们使用当前行，除非您的情况只有 1 行，否则我们将不会获得异常输出，因此如果该功能可用，我们应该使用按范围滑动。如果没有将您的数据与日期维度结合起来并为您的滑动 window 填充数据，并使用案例语句来获取您的 windows（如果您有多个 windows）。现在我们有了所需的数据，在案例上使用常规 window 函数并对其应用计数。