Intro

Things to keep in mind regarding SQL shared screen coding challenges:

Make sure you are familiar with window functions.

As always with shared screen interviews, break down the problem as much as possible.
This is typically easily doable with SQL. And then solve each small chunk. The goal of
shared screen interviews is to check that you can code. Focus on sending that message
more than trying to look super smart.

Some companies will ask you to solve the problem with just one query, i.e. you are not
allowed to create temporary tables. Even in that case, break down the problem into
smaller pieces and solve them independently. Then put these parts together into just one
query via subqueries.

The data sets attached to this pdf are just for you to practice and verify that your query
works. Almost never in a shared screen interview you will have to work with actual data.
Here there are no wrong data to find, nothing related to the preprocessing steps you
usually have to take care of in a takehome challenge or insights to find. Datasets are just
to make sure the query is good.

Solutions below are in Hive, but they work for pretty much all SQL-like languages
commonly used by data scientists.
Query 1

Goal

For each user_id, find the difference between the last action and the second last action. Action
here is defined as visiting a page. If the user has just one action, you can either remove her from
the final results or keep that user_id and have NULL as time difference between the two actions.

The table below shows for each user all the pages she visited and the corresponding
timestamp.

Data

You can download the data set here

head (query_one,1)

Column Name Value Description

user_id 6684 this is id of the user

page home_page the page visited
unix_timestamp 1451640067 unix timestamp in seconds
Query 1 Solution

SELECT user_id,
unix_timestamp - previous_time AS Delta_SecondLast0ne_LastOne
FROM
(SELECT user_id,
unix_timestamp,
LAG(unix_timestamp, 1) OVER (PARTITION BY user_id ORDER BY
unix_timestamp) AS previous_time,
ROW_NUMBER() OVER (PARTITION BY user_id ORDER BY
unix_timestamp DESC) AS order_desc
FROM query_one
) tmp
WHERE order_desc = 1
ORDER BY user_id
LIMIT 5;

user_id Delta_SecondLast0ne_LastOne
2
3 2
4 4
5
7
Query 2

Goal

We have two tables. One table has all mobile actions, i.e. all pages visited by the users on
mobile. The other table has all web actions, i.e. all pages visited on web by the users.
Write a query that returns the percentage of users who only visited mobile, only web and both.
That is, the percentage of users who are only in the mobile table, only in the web table and in
both tables. The sum of the percentages should return 1.

Data

You can download the data sets here.

head (data_mobile,1)

Column
Value Description
Name
this is id of the user who visited a given page on
user_id 128
mobile
page page_5_mobile page visited by that user on mobile

head (data_web,1)

Column Name Value Description

user_id 1210 this is id of the user who visited a given page on web
page page_1_web page visited by that user on web
Query 2 Solution

SELECT 100*SUM(CASE WHEN m.user_id IS null THEN 1 ELSE 0 END)/COUNT(*) as

WEB_ONLY,
100*SUM(CASE WHEN w.user_id IS null THEN 1 ELSE 0 END)/COUNT(*) as
MOBILE_ONLY,
100*SUM(CASE WHEN m.user_id IS NOT null AND w.user_id IS NOT null
THEN 1 ELSE 0 END)/COUNT(*) as BOTH
FROM
(SELECT distinct user_id FROM query_two_web ) w
FULL OUTER JOIN
(SELECT distinct user_id FROM query_two_mobile ) m
ON m.user_id = w.user_id;

WEB_ONLY MOBILE_ONLY BOTH

16 31 52
Query 3

Goal

We define as power users those users who bought at least 10 products. Write a query that
returns for each user on which day they became a power user. That is, for each user, on which
day they bought the 10th item.

The table below represents transactions. That is, each row means that the corresponding user
has bought something on that date.

Data

You can download the data set here.

head (data,1)

Column
Value Description
Name

user_id 675 this is id of the user

2014-12-31 user 675 bought something on Dec 31, 2014 at
date
16:16:12 4:16:12 PM
Query 3 Solution

SELECT user_id,
date
FROM
(SELECT *, ROW_NUMBER() over(PARTITION BY user_id ORDER BY date) row_num FROM
query_three ) tmp
WHERE row_num = 10
LIMIT 5;

user_id date

1 2015-10-21 06:20:14
3 2015-10-29 22:41:00
4 2015-09-25 12:36:10
5 2015-09-18 06:25:40
6 2015-12-30 00:53:59
Query 4

Goal

We have two tables. One table has all $ transactions from users during the month of March and
one for the month of April.

Write a query that returns the total amount of money spent by each user. That is, the sum
of the column transaction_amount for each user over both tables.

Write a query that returns day by day the cumulative sum of money spent by each user.
That is, each day a user had a transcation, we should have how much money she has
spent in total until that day. Obviously, the last day cumulative sum should match the
numbers from the previous bullet point.

Data

You can download the data sets here

head (data_march,1)

Column Name Value Description

this is id of the user who had the corresponding

user_id 13399
transaction
2015-03-
date the transaction happened on March 1st.
01
transaction_amount 18 the user spent 18$ in that transaction

head (data_april,1)
Column Name Value Description

this is id of the user who had the corresponding

user_id 15895
transaction
2015-04-
date the transaction happened on April 1st.
01
transaction_amount 66 the user spent 66$ in that transaction
Query 4 Solution

Query 4.1 Solution

SELECT user_id,
SUM(transaction_amount) as total_amount
FROM
(
SELECT * FROM query_four_march
UNION ALL
SELECT * FROM query_four_april
) tmp
GROUP BY user_id
ORDER BY user_id
LIMIT 5;

user_id total_amount

2 67
3 26
4 156
5 45
6 32

Query 4.2 Solution

SELECT user_id,
date,
SUM(amount) over(PARTITION BY user_id ORDER BY date) as total_amount
FROM
(
SELECT user_id, date, SUM(transaction_amount) as amount
FROM query_four_march
GROUP BY user_id, date
UNION ALL
SELECT user_id, date, SUM(transaction_amount) as amount
 FROM query_four_april
GROUP BY user_id, date
) tmp
ORDER BY user_id, date
LIMIT 5;

user_id date total_amount

2 2015-03-13 67
3 2015-03-31 26
4 2015-03-28 63
4 2015-04-20 156
5 2015-03-01 45
Query 5

Goal

We have two tables. One is user id and their signup date. The other one shows all transactions
done by those users, when the transaction happens and its corresponding dollar amount.

Find the average and median transaction amount only considering those transactions that
happen on the same date as that user signed-up.

Data

You can download the data sets here

head (user,1)

Column Name Value Description

user_id 121 this is id of the user

sign_up_date 2015-01-02 user_id 121 signed up on Jan, 2.

head (transaction_table,1)

Column Name Value Description

user_id 856898 this is id of the user who had that transaction

2015-08-02 transaction happened on Aug, 2 at almost
transaction_date
03:56:08 4AM.
transaction_amount 49 transaction amount was 49$.
Query 5 Solution

Note: there are percentile built-in functions that can be used to estimate the median. However,
estimating the median is simple enough that it is a good exercise to try to implement it without
using the built-in function.

Solution

SELECT AVG(transaction_amount) AS average,

AVG(CASE WHEN row_num_asc BETWEEN row_num_desc-1 and row_num_desc+1
THEN transaction_amount ELSE NULL END ) AS median
FROM
(
SELECT transaction_amount,
ROW_NUMBER() OVER(ORDER BY transaction_amount) row_num_asc,
COUNT(*) OVER() - ROW_NUMBER() OVER(ORDER BY transaction_amount)
+ 1 AS row_num_desc -- need row number for median. there are many other ways
to do this
FROM query_five_users a
JOIN (SELECT *, to_date(transaction_date) AS date_only FROM
query_five_transactions) b
ON a.user_id = b.user_id AND a.sign_up_date = b.date_only
) tmp;

average median

49.25 49
Query 6

Goal

We have a table with users, their country and when they created the account. We want to find:

The country with the largest and smallest number of users

A query that returns for each country the first and the last user who signed up (if that
country has just one user, it should just return that single user)

Data

You can download the data set here

head (data,1)

Column
Value Description
Name

user_id 2 this is id of the user

2015-02-28 user 2 created her account on Feb, 2 around
created_at
16:00:40 4PM
country China She is based in China
Query 6 Solution

Query 6.1 Solution

SELECT country,
user_count
FROM
(
SELECT *,
ROW_NUMBER() OVER (ORDER BY user_count) count_asc,
ROW_NUMBER() OVER (ORDER BY user_count desc) count_desc
FROM (
SELECT country, COUNT(distinct user_id) as user_count
FROM query_six
GROUP BY country
) a

) tmp
WHERE count_asc = 1 or count_desc = 1;

country user_count

China 18350
Vietnam 1

Query 6.2 Solution

SELECT user_id,
created_at,
country
FROM
(
SELECT *,
ROW_NUMBER() OVER (PARTITION BY country ORDER BY created_at) count_asc,
ROW_NUMBER() OVER (PARTITION BY country ORDER BY created_at desc)
count_desc
FROM query_six
) tmp
WHERE count_asc = 1 or count_desc = 1
LIMIT 5;

user_id date total_amount

999103 2015-09-29 13:32:07 Bangladesh
155 2015-02-28 16:59:45 Bangladesh
999671 22015-09-29 16:12:39 Brazil
234 2015-02-28 17:39:27 Brazil
999806 2015-09-29 16:34:59 China