Aggregate query - column for querying affects aggregation - sql-server

Question

I have a table "Scores" with fields as follows:
UserId
LessonId
ExerciseId
Score
Timestamp
I want to setup a view, "vw_AggregateScoreForUser" that will aggregate data from that table, as follows:
SELECT UserId,
LessonId,
COUNT(ExerciseId) AS TotalExercises,
SUM(Score) AS TotalScore,
COUNT(DISTINCT CONVERT(date, Timestamp)) AS StudyDays
FROM Scores
GROUP BY UserId, LessonId
The tricky bit is StudyDays, where I'm counting the unique dates that the user has at least one entry here on - that gives me the days that they "studied", i.e. completed at least one exercise.
Now, say that I want to execute this view for lessons 1 to 5.
SELECT FROM vw_AggregateScoreForUser WHERE UserId = 1 AND LessonId BETWEEN 1 AND 5;
What I want, is one record returned that aggregates the data for those 5 lessons. But with the above setup, the data is grouped by LessonId, so I will get 5 records back.
The issue is that StudyDays may now be incorrect as it's computed per lesson. E.g. with the following data:
UserId LessonId ExerciseId ... Timestamp
1 1 1 2019-11-21 09:00
1 1 2 2019-11-22 10:00
1 2 1 2019-11-22 11:00
I would get the result
UserId LessonId TotalExercises ... StudyDays
1 1 2 2
1 2 1 1
I can't simply add StudyDays to get the number of days studied. That would give me 3, but the distinct count for StudyDays overall should be 2.
The issue is that I need LessonId in the view in order to be able to use it in the WHERE clause, but having it in the view will group my data by lesson causing the aggregate to be incorrect.
How do you include a field in a view so that you can filter on it, without having it affect the aggregation that occurs in that view?

Answer 1

Some grouping aggregates can't be stacked in multiple levels, as they give different result. A count-distinct from a count-distinct isn't the same as applying a count-distinct from the original set. The same happens with averages, which take into account the number of rows.
The problem in your case is the GROUP BY LessonID with a COUNT DISTINCT inside the view. You are already computing values by LessonID when you want (later on) multiple LessonID values to be computed together as a set.
As long as you keep your GROUP BY inside the view, you will have this problem. A solution would be changing the view for a table-valued function, which allows a range of lessons to be supplied:
CREATE FUNCTION dbo.ufnUserLessonSummary (
#UserID INT,
#LessonIDFrom INT,
#LessonIDTo INT)
RETURNS TABLE
AS RETURN
SELECT
UserId,
LessonId,
COUNT(ExerciseId) AS TotalExercises,
SUM(Score) AS TotalScore,
COUNT(DISTINCT CONVERT(date, Timestamp)) AS StudyDays
FROM
Scores AS S
WHERE
S.UserID = #UserID AND
S.LessonID BETWEEN #LessonIDFrom AND #LessonIDTo
GROUP BY
UserId,
LessonId
You can query it like the following:
SELECT
S.*
FROM
dbo.ufnUserLessonSummary(1, 1, 5) AS S
However, this is limited to a range of lessons. What happens if you want only lessons 1, 3 and 5? Another more complex, but more versatile option is to use an SP with a pre-loaded input table:
CREATE PROCEDURE dbo.uspUserLessonSummary
AS
BEGIN
SELECT
UserId,
LessonId,
COUNT(ExerciseId) AS TotalExercises,
SUM(Score) AS TotalScore,
COUNT(DISTINCT CONVERT(date, Timestamp)) AS StudyDays
FROM
Scores AS S
INNER JOIN #UserLesson AS U ON
S.UserID = U.UserID AND
S.LessonID = U.LessonID
GROUP BY
UserId,
LessonId
END
You can supply which records you want by loading the temporary table before executing:
IF OBJECT_ID('tempdb..#UserLesson') IS NOT NULL
DROP TABLE #UserLesson
CREATE TABLE #UserLesson (
UserID INT,
LessonID INT)
INSERT INTO #UserLesson (
UserID,
LessonID)
VALUES
(1, 1),
(1, 2),
(1, 3),
(1, 4),
(1, 5)
EXEC dbo.uspUserLessonSummary
You can also use variable tables with this approach.