getting a distinct count from with a date field - sql

Question

I have a piece of code that is looking for the distinct count of Kegs, the count of the distinct kegs that are tagged and ones that are untagged, what I have so far is:
with CTE as
(select UID_KEG, IS_TAGGED, movement_date
from MOVEMENT M
inner join Keg on M.UID_Keg = Keg.Unique_ID
where DATEPART(year,Movement_date) = '2019'
and UID_MOVEMENT_TYPE = 1
)
select COUNT(Distinct CTE.UID_KEG) as 'Kegs', datepart(week,movement_date)
as 'Week number',
SUM(case when Is_Tagged = 1 then 1 end) as 'tagged',
SUM(case when Is_Tagged = 0 then 1 end) as 'untagged'
from CTE
group by datepart(week,movement_date)
order by [Week number] asc
It currectly returns a distinct count of the kegs but the figures for tagged and un tagged are incorrect and I can only assume it because it's counting duplicate kegs.
Can any one advise how I can get round this or do a count on just the distinct kegs?

Answer 1

You want conditional aggregation using COUNT(DISTINCT). That would be:
SELECT COUNT(DISTINCT CTE.UID_KEG) as Kegs,
datepart(week, movement_date) as Week_number,
COUNT(DISTINCT CASE WHEN Is_Tagged = 1 THEN CTE.UID_KEG END) as tagged,
COUNT(DISTINCT(CASE WHEN Is_Tagged = 0 THEN CTE.UID_KEG END) as untagged
FROM CTE
GROUP BY datepart(week, movement_date)
ORDER BY MIN(movement_date);
Notes:
The tagged and untagged counts may still add up to more than the total count, assuming that kegs can be both tagged and untagged in a single week.
You should include the year() as well as the week, especially because you are not selecting data from a single year.
Only use single quotes for string and date constants. Do not use them for column aliases; that can lead to hard-to-debug errors.

Answer 2

If you remove the Distinct from your count, the sum of untapped and tapped should equal your total (if it is a binary 0 or 1). This indicates that you have duplicate UID_KEG values. Take some time to understand why. Part of your problem is that it seems you don't quite understand the shape of your dataset very well.
Take some time to look at the data to understand if there are duplicates (why? are they caused by the join, or are they in the base data?), look to see if they can appear as tagged and untagged.
EDIT: In response to your comment. If they can be scanned twice you will have to have the assumption that if Is_Tagged = 1 for any UID_KEG in that day, then all kegs with that UID_KEG are tagged.
In that case you will have to adapt the code to use this assumption.
WITH CTE
AS (
SELECT UID_KEG
,IS_TAGGED
,movement_date
FROM MOVEMENT M
INNER JOIN Keg ON M.UID_Keg = Keg.Unique_ID
WHERE DATEPART(year, Movement_date) = '2019'
AND UID_MOVEMENT_TYPE = 1
)
SELECT CTE.UID_KEG AS 'Kegs'
,datepart(week, movement_date) AS 'Week number'
,MAX(Is_Tagged) AS 'tagged'
FROM CTE
GROUP BY CTE.UID_KEG
,datepart(week, movement_date)
ORDER BY [Week number] ASC
This code might not be perfect, I couldn't test it, but it should get you a complete list of each keg, in each day, and if that keg was marked as tagged at least once, and if it was not marked as tagged at all.
The most important thing here is removing duplication of the kegs within each day, then it is possible to calculate.
I'm not great with CTE's but you will need to aggregate one level up to the daily level, now you will be able to count the distinct number of kegs and which ones were tagged and untagged.
Hope that makes sense.
EDIT: here is a subquery that should work
SELECT [Week number]
,count(1) [numKegs]
,sum(tagged) [numTagged]
FROM (
SELECT UID_KEG AS 'Kegs'
,datepart(week, movement_date) AS 'Week number'
,MAX(IS_TAGGED) AS 'tagged'
FROM MOVEMENT M
INNER JOIN Keg ON M.UID_Keg = Keg.Unique_ID
WHERE DATEPART(year, Movement_date) = '2019'
AND UID_MOVEMENT_TYPE = 1
GROUP BY UID_KEG
,datepart(week, movement_date)
) kegdailylevel
GROUP BY [Week number]
ORDER BY [Week number] ASC