How can I count distinct multiple fields without r

问题:

I have a query with several groupings that returns a count per month. Something like this:

SELECT field1, field2, year(someDate), month(someDate), count(*) as myCount
FROM myTable
WHERE field5 = 'test'
GROUP BY field1, field2, year(someDate), month(someDate)

The problem is that I want the count to be distinct per day, based on an id field + the date field (without the time). As in, I want to get the distinct count of ids each day, per month. So I want something like this:

SELECT field1, field2, year(someDate), month(someDate), 
       count(distinct someID, someDate) as myCount
FROM myTable
WHERE field5 = 'test'
GROUP BY field1, field2, year(someDate), month(someDate)

There are 2 problems with this:

1. You can't list 2 distinct fields for a count aggregate
2. This will include the time of the date as well, so it won't filter anything out as it will almost always have a different time

I can take care of 2. easily by converting to a varchar of just the date, but I'm not sure how to deal with the multiple distinct fields issue. I can't use this solution, as I don't want to repeat the entire where clause and group by clause. This is what I've come up with:

SELECT field1, field2, year(someDate), month(someDate), 
       count(distinct someID + CONVERT(VARCHAR, someDate, 112)) as myCount
FROM myTable
WHERE field5 = 'test'
GROUP BY field1, field2, year(someDate), month(someDate)

Instead of listing the distinct fields in a comma-delimited list, I simply concatenated them. Is there any downside to this method that I should look out for? Can I count on it to be accurate? And - is there any better way to accomplish this?

Basically, I'm grouping per month, but the "distinct" count should be based on day. As in, if I have id 31 on Jan 3 and Jan 5, I want it to count as 2 for January, but if I have id 31 twice on Jan 3, I only want it to count once.

Some basic sample data & expected output (skipping field1 and field2 for this):

*Date*              *ID*
1/3/12 00:00:09     22
1/3/12 00:13:00     22
1/4/12 12:00:00     22
1/7/12 15:00:45     27
1/15/12 15:00:00    22
2/6/12 00:00:09     50
2/8/12 00:13:00     44
2/8/12 12:00:00     45
2/22/12 15:00:45    33
2/22/12 15:00:00    33
2/22/12 15:00:00    44

*Year*  *Month* *Count*
2012    Jan     4
2012    Feb     5

回答1:

UPDATED

Based on your sample data, this gives the required result:

Declare @Tab table ([Date] datetime,ID int)
insert into @Tab([Date],ID) values
('2012-01-03T00:00:09.000', 22),
('2012-01-03T00:13:00.000', 22),
('2012-01-04T12:00:00.000', 22),
('2012-01-07T15:00:45.000', 27),
('2012-01-15T15:00:00.000', 22),
('2012-02-06T00:00:09.000', 50),
('2012-02-08T00:13:00.000', 44),
('2012-02-08T12:00:00.000', 45),
('2012-02-22T15:00:45.000', 33),
('2012-02-22T15:00:00.000', 33),
('2012-02-22T15:00:00.000', 44)

select DATEADD(month,DATEDIFF(month,0,[Date]),0) as MonthStart,SUM(distinctDayIDs)
from
(
    SELECT DATEADD(day,DATEDIFF(day,0,[Date]),0) as [Date], 
           count(distinct ID) as distinctDayIDs
    FROM @Tab
    --WHERE field5 = 'test'
    GROUP BY DATEADD(day,DATEDIFF(day,0,[Date]),0)
) t
group by DATEADD(month,DATEDIFF(month,0,[Date]),0)

I think, because we have to do a count for each day, we have to do it as two separate grouping operations.

---

Older Answer

It sounds like the desired output would be field1, field2, the date, and the count of distinct IDs on that date?

If so, I think you're overcomplicating things:

SELECT field1, field2, DATEADD(day,DATEDIFF(day,0,someDate),0) as Date, 
       count(distinct someID) as myCount
FROM myTable
WHERE field5 = 'test'
GROUP BY ffield1, field2, DATEADD(day,DATEDIFF(day,0,someDate),0)

(I'm using DATEADD/DATEDIFF to strip the time portion, rather than converting to varchar)

回答2:

You can try with a count 'over partition':

SELECT 
   field1, field2, someID, someDate, 
   count(*) OVER(PARTITION BY someID, someDate) as myCount
FROM myTable
WHERE field5 = 'test'
GROUP BY field1, field2, someID, someDate

Or prepare a CTE select:

;with cte as (
   select someDate, count( someID) as myCount
   from myTable
   group by someDate)
 select m.field1, m.field2, m.someID, m.someDate, cte.myCount
 from myTable m inner join cte 
   on m.someDate = cte.someDate
 where ...