SQL Group By +多个表计数

Question

I'm studying for an interview next week which has a small data analysis component. The recruiter gave me the following sample SQL question which I'm having trouble wrapping my mind around a solution. I'm hoping that I'm not biting off more than I can chew ;)..

SAMPLE QUESTION:

You are given two tables:

AdClick Table (columns: ClickID, AdvertiserID, UserID, and other fields) and AdConversion Table (columns: ClickID, UserID and other fields).

You have to find the total conversion rate (# of conversions/# of clicks) for users with 1 click, 2 click etc.

I've been playing with this for about an hour and keep hitting road blocks. I understand COUNT and GROUP BY but suspect I'm missing a simple SQL feature that I'm unaware of. This also makes it difficult for me to find any possible pointers/solutions via Google: not knowing the magic keywords to search on.

Example Input

dbo.AdConversion
----------------
ClickID UserID
1   1
2   1
4   1
5   3
6   2
7   2
12  1
9   4
10  4

dbo.AdClick
-----------
ClickID AdvertiserID    UserID
1   1   1
2   2   1
3   1   2
4   1   1
5   1   3
6   2   2
7   3   2
8   1   1
9   4   4
10  2   4
11  3   4
12  2   1

Expected Result:
----------------
UserClickCount  ConversionRate
4       80.00%
2       66.67%
1       100.00%     

Explanation/Clarification:

Users with 4 AdConversion.ClickIDs (aka Conversions) have an 80% conversation rate. Here there's just one user, UserID 1, which has 5 AdClicks with 4 AdConversions.

Users with 2 Conversions have a combined 6 Adclicks with 4 conversions for a conversion rate of 66.67%. Here, that'd be UserID 2 and 4.

Users with 1 Conversion, here only UserID 3, has 1 conversion against 1 AdClick for a 100% conversion rate.

---

Here's one possible solution I've come up with after some direction from Zack's comment. I can't imagine that it's the ideal solution or whether it has bugs in it or not:

DECLARE @Conversions TABLE
(
UserID int NOT NULL,
AdConversions int
)

INSERT INTO @Conversions (UserID, AdConversions)
SELECT adc.UserID, COUNT(adc.UserID)
FROM dbo.AdConversion adc
GROUP BY adc.UserID;

DECLARE @Clicks TABLE
(
UserID int NOT NULL,
AdClicks int
)
INSERT INTO @Clicks(UserID, AdClicks)
SELECT UserID, Count (ClickID)
FROM dbo.AdClick
GROUP BY UserID;


SELECT co.AdConversions, CONVERT(decimal(6,3), (CAST(SUM(co.AdConversions) AS float) / SUM(cl.AdClicks))) * 100
FROM @Conversions co
INNER JOIN @Clicks cl
ON co.UserID = cl.UserID
GROUP BY co.AdConversions;

Any advice would be greatly appreciated!

Thanks, Michael

Answer 1

Your logic seems good. Here is a version with common table expressions and a little update with the numeric conversion:

WITH tConversions as 

(SELECT UserID, COUNT(ClickID) as AdConversions
FROM AdConversion
GROUP BY UserID),

tClicks as

(SELECT UserID, COUNT(ClickID) as AdClicks
FROM AdClick
GROUP BY UserID)

SELECT co.AdConversions, CONVERT(decimal(10,2),CAST(SUM(co.AdConversions) as float) / SUM(cl.AdClicks) * 100) as ConversionRate
FROM tConversions co
INNER JOIN tClicks cl
ON co.UserID = cl.UserID
GROUP BY co.AdConversions

You can also use subqueries directly:

SELECT co.AdConversions, CONVERT(decimal(10,2),CAST(SUM(co.AdConversions) as float) / SUM(cl.AdClicks) * 100) as ConversionRate
FROM 

(SELECT UserID, COUNT(ClickID) as AdConversions
FROM AdConversion
GROUP BY UserID)

as co

INNER JOIN

(SELECT UserID, COUNT(ClickID) as AdClicks
FROM AdClick
GROUP BY UserID)

as cl

ON co.UserID = cl.UserID
GROUP BY co.AdConversions