liquidcarbon / sql_interview

Interactive SQL interview on Google Colaboratory

Interactive SQL interview on Google Colaboratory

Recently I had a technical interview on SQL performed over Webex, with tasks and solutions typed into a shared Google Doc. This is quite standard and works fine. But wouldn't it be nice if all participants of the process could interact with a real database, more closely approximating real work?

This solution uses Google Colaboratory, where a SQLite database is created at runtime using Python Pandas, and the SQL queries are written into templated solution cells.

Sample task

We have a table events that looks like this:

events (
  account STRING,
  event STRING,
  timestamp DATETIME
)

Provide counts of each event type for each account.

output (
  account STRING,
  event1_count INT,
  event2_count INT,
  event3_count INT
)

Create a database with sample data

import numpy as np
import pandas as pd
import secrets
import sqlite3
import time

accounts = [secrets.token_hex(2) for i in range(5)]
event_types = ['PLAY','STOP','ERROR']
time_now = int(time.time()*1e3)
N = 100
df = pd.DataFrame({
    'account': np.random.choice(accounts, size=N),
    'event': np.random.choice(event_types, p=[.5, .3, .2], size=N),
    'timestamp': [(time_now) - np.random.randint(0, 1e6) for i in range(N)],
}).sort_values(['timestamp'])
conn = sqlite3.connect('interview.sqlite')
df.to_sql('events', conn, if_exists='replace', index=False)

Solution

# solution - using CASE because SQLite does not have IF
q = '''
SELECT
  account
  ,SUM(CASE event WHEN 'PLAY' THEN 1 ELSE 0 END)  AS count_PLAY
  ,SUM(CASE event WHEN 'STOP' THEN 1 ELSE 0 END)  AS count_STOP
  ,SUM(CASE event WHEN 'ERROR' THEN 1 ELSE 0 END) AS count_ERROR
FROM events
GROUP BY account
'''
result = pd.read_sql(q, conn)
result

Output

	account	count_PLAY	count_STOP	count_ERROR
0	251a	12	5	3
1	e4b0	14	3	2
2	78a3	15	2	4
3	ba75	10	5	2
4	9c11	17	2	3