DATA

Spring 2022
8 Foundations of Data Science Midterm Solution

INSTRUCTIONS
You have 1 hour and 50 minutes to complete the exam.
• The exam is closed book, closed notes, closed computer, closed calculator, except the provided midterm study
guide.
• Mark your answers on the exam itself in the spaces provided. We will not grade answers written on scratch
paper or outside the designated answer spaces.
• If you need to use the restroom, bring your phone and exam to the front of the room.
For questions with circular bubbles, you should select exactly one choice.
# You must choose either this option
# Or this one, but not both!
For questions with square checkboxes, you may select multiple choices.
2 You could select this choice.
2 You could select this one too!
Preliminaries
You can complete and submit these questions before the exam starts.
(a) What is your full name?

(b) What is your student ID number?

(c) Who is your lab GSI? You may write self-service if you have no lab GSI.

(d) Sign here to confirm that all work on this exam is your own (or type your name if online).
 2

1. (41.0 points) Basketable

The teams table contains one row for each of the 30 teams in the National Basketball Association (NBA) league.
Columns exist for the team’s name, division, conference, and home arena capacity. Each team has its own
arena. The first five rows are:

name division conference arena

Celtics Atlantic Eastern 18642
Lakers Pacific Western 18997
Nets Atlantic Eastern 17732
Pistons Central Eastern 20491
Rockets Southwest Western 18055

The players table contains a row for each of the 528 players in the 2020 NBA season. Columns are the player’s
name, 2019 salary (2019), 2020 salary (2020), 2019 team name (19team), and 2020 team name (20team). For
players who joined in 2020, their 2019 value is 0 and their 19team value is No Team. The first three rows are:

name 2019 2020 19team 20team

Stephen Curry 37457154 40231758 Warriors Warriors
Dwight Howard 5337000 1620564 Wizards Lakers
Zion Williamson 0 9757440 No Team Pelicans

(a) (4.0 points)

This partially completed expression evaluates to the name of the team (a string) with the smallest arena
capacity. Assume no two arenas have the same capacity.
teams._________._________._________
(a) (b) (c)
i. (1.0 pt) Fill in blank (a).

sort('arena')

ii. (2.0 pt) Fill in blank (b).

column('name')

iii. (1.0 pt) Which of these could fill in blank (c)?

# min()
# max()
item(0)
# item(1)
 3

(b) (3.0 points)

This partially completed expression evaluates to a table with one row per division in the Eastern
conference that has two columns: the division and the count of the number of teams in that division.
This expression should evaluate to the following table.

division count
Atlantic 5
Central 5
Southwest 5

teams._________.group(_________)
(a) (b)
Reminders:
• The teams table has columns name, division, conference, and arena.
• The players table has columns name, 2019, 2020, 19team, and 20team.
i. (2.0 pt) Fill in blank (a).

where('conference', 'Eastern')

ii. (1.0 pt) Which of these could fill in blank (b)?

# 'count'
# 'conference'
'division'
# 'name'
 4

(c) (6.0 points)

This partially completed expression evaluates to a table with one row per division that has two
columns: the division and the total 2020 salary for all players in that division. Any label for the second
column is acceptable.
teams._________.select('division', '2020')._________
(a) (b)
Reminders:
• The teams table has columns name, division, conference, and arena.
• The players table has columns name, 2019, 2020, 19team, and 20team.
i. (4.0 pt) Fill in blank (a).

join('name', players, '20team')

ii. (2.0 pt) Fill in blank (b).

group('division', sum)
 5

(d) (12.0 points)

Write an expression that correctly computes each of the following quantities.
You may use t for teams, p for players, a for the arena column of the teams table, and np for NumPy.
import numpy as np
t = teams
p = players
a = teams.column('arena')
Reminders:
• The teams table has columns name, division, conference, and arena.
• The players table has columns name, 2019, 2020, 19team, and 20team.
i. (3.0 pt) The largest increase in salary from 2019 to 2020 (an integer) of any player.

max(players.column('2020') - players.column('2019'))

ii. (3.0 pt) The number of players in 2020 (an integer) who played for the same team in 2019 and
2020.

sum(players.column('19team') == players.column('20team'))

iii. (2.0 pt) The number of teams (an integer) that have an arena size that is above average.

sum(a > np.average(a))

iv. (4.0 pt) Select all of the quantities below that can be computed from these two tables.
 The number of divisions that had at least 5 players paid more than $20,000,000 in 2020
 The name of the team that paid the most player salary per seat in its arena in 2020 (Note: The
number of seats in an arena is its capacity.)
2 The number of players who retired after the 2019 season.
2 The name of the player that made the most additional salary by changing teams in 2020 compared
to the amount they would have made staying at their 2019 team
2 None of these
 6

(e) (16.0 points)

The two histograms below displaying 2020 salaries were generated from data in the players table. The
first histogram only includes players who had the same 19team and 20team. The second only includes
players who had a different 19team from their 20team and played in both 2019 and 2020.
The bins are make_array(0, 5, 10, 15, 20, 30, 40).
Additional information: Among the 440 players who played in both 2019 and 2020, 60% played on the
same team and 40% played on different teams.

i. (2.0 pt) About what percentage of the players who had the same 19team and 20team had a salary
between $10 million and $20 million in 2020?
15%
# 20%
# 25%
# 30%
# 35%

ii. (2.0 pt) About what percentage of the players who had the same 19team and 20team had a salary
of $10 million or more in 2020?
# 10%
30%
# 50%
# 70%
# 90%

iii. (4.0 pt) About how many players played on different teams in 2019 and 2020 and made between $5
million and $10 million in 2020?
Please express your answer as a Python expression (e.g., 0.1 * 0.2 + 0.3) rather than simplifying it
to a single number.

440 * 0.4 * 3 * 5 / 100

 7

iv. (4.0 pt) Select all of the quantities below that can be determined from only these two histograms
and the additional information that appears just above the histograms.
Reminder: The additional information was that among the 440 players who played in both 2019 and
2020, 60% played on the same team and 40% played on different teams.
2 The total number of players who played in 2019 and had a 2020 salary below $2 million
 The total number of players who played in 2019 and had a 2020 salary below $20 million
 Among all players who played in both 2019 and 2020, the proportion who had a salary of $20
million or more
 Among all players who played in 2019 and had a 2020 salary of $20 million or more, the proportion
who played on the same team in 2019 and 2020
2 None of these.

v. (2.0 pt) How would you use these histograms to determine whether the 2020 salary distribution was
different for players with a different team than for players with the same team?
Compare the two histograms visually and look for differences.
# Use the two histograms to perform an A/B test.
# Use the histograms to compute the average salary for both groups and compare those averages.
# Use the histograms to compute the total salary for both groups and compare those totals.

vi. (2.0 pt) The $30-$40 million bin is slightly taller for players with a different team (right histogram)
than for players with the same team (left histogram). What can we conclude from this difference?
# Players who switch teams are paid more.
# Players who switch teams are more likely to end up with a salary of $30-$40 million.
Within that bin, the density among players with a different team is higher than the density among
players with the same team.
# Within that bin, the number of players with a different team is higher than the number of players
with the same team.
 8

2. (11.0 points) Sus

In the mobile game Among Us, Crewmates on a spaceship work together to complete tasks while a few
randomly-selected Imposters secretly try to eliminate crewmates. If all Crewmates complete their tasks, the
Crewmates win; if the Imposters eliminate all but one of the crewmates, the Imposters win.
Matty made a games table listing each game they played in 2021, ordered chronologically. The first three rows:

team outcome length completed

Crewmate Win 981 7
Imposter Loss 840 8
Crewmate Loss 520 3

The columns include:

• team: which team Matty was on in the game.
• outcome: whether Matty’s team won or lost.
• length: the duration of the game in seconds.
• completed: the number of tasks completed by all crewmates before the game ended.
(a) (3.0 points)
Choose which type of visualization would be most useful for investigating each of the following.
i. (1.0 pt) The distribution of game lengths.
# Bar Chart
Histogram
# Line Plot
# Scatter Plot

ii. (1.0 pt) The association between game length and number of tasks completed.
# Bar Chart
# Histogram
# Line Plot
Scatter Plot

iii. (1.0 pt) The average game length for each outcome.
Bar Chart
# Histogram
# Line Plot
# Scatter Plot
 9

(b) (8.0 points)

i. (4.0 pt) The result of which of the following expressions contains in one of its cells the total number
of games in which Matty won? Select all that apply.
2 games.pivot('outcome', 'team')
 games.group('outcome')
2 games.group('team')
2 games.group(['team', 'outcome'])
2 None of these

ii. (4.0 pt) The result of which of the following expressions contains in one of its cells the total number
of tasks completed in all games for which Matty was a Crewmate and lost? Select all that apply.
2 games.pivot('completed', 'team', 'outcome', collect=sum)
 games.pivot('team', 'outcome', 'completed', collect=sum)
2 games.group('team').group('outcome').group('completed', collect=sum)
 games.group(['team', 'outcome'], collect=sum)
2 None of these
 10

3. (24.0 points) Chances

Each pet photo at the end of a lab is chosen from a collection of 20 pets with 10 cats, 9 dogs, and 1 bird.
For each event below, choose the Python expression that evaluates to the probability of that event.
(a) (8.0 points)
i. (2.0 pt) When one pet is chosen at random, the probability that it is either a cat or a bird.
# (9 / 20) ** 2
# (10 / 20) * (1 / 20)
(10 / 20) + (1 / 20)
# 1 - (9 / 20) ** 2
# 1 - (10 / 20) * (1 / 20)
# 1 - ((10 / 20) + (1 / 20))

ii. (2.0 pt) When two pets are chosen at random with replacement, the probability that they are both
dogs.
(9 / 20) ** 2
# (10 / 20) * (1 / 20)
# (10 / 20) + (1 / 20)
# 1 - (9 / 20) ** 2
# 1 - (10 / 20) * (1 / 20)
# 1 - (10 / 20) + (1 / 20)

iii. (2.0 pt) When two pets are chosen at random with replacement, the probability that the first is a
cat and the second is not.
# 10 / 20 + 10 / 20
(10 / 20) * (10 / 20)
# (10 / 20) * (9 / 20) * (1 / 20)
# 1 - (10 / 20) * (10 / 20)
# 1 - (10 / 20 + 10 / 20)
# 1 - (10 / 20) * (9 / 20) * (1 / 20)

iv. (2.0 pt) When two pets are chosen at random with replacement, the probability that the first chases
the second. Assume dogs only chase cats, cats only chase birds, and birds don’t chase.
(10 / 20) * (10 / 20)
# (19 / 20) * (10 / 20)
(10 / 20) * (1 / 20) + (9 / 20) * (10 / 20)
# 1 - ((9 / 20) * (1 / 20) + (10 / 20) * (9 / 20))
# 1 - ((10 / 20) ** 2 + (9 / 20) ** 2 + (1 / 20) ** 2)
# 1 - ((10 / 20) ** 2 + (9 / 20) ** 2 + (1 / 20))
 11

(b) (8.0 points)

The pygmy hippo is a small, reclusive (and cute) hippopotamid type that is native to the forests and
swamps of West Africa. Two teams of zoologists set out to estimate the proportion that are male by
sampling at random from the population. The first team samples 100 hippos and finds the proportion of
males in their sample to be A. The second team samples 40 hippos and finds the proportion of males in
their sample to be B. The full population has all 2,500 wild pygmy hippos; the proportion P of males in
the population is 50% (but unknown to the zoologists).
i. (4.0 pt) Which of the following are more likely than not? Select all that apply.
2 A is smaller than B.
2 A is larger than B.
 P is closer to A than B.
2 P is closer to B than A.
2 None of these.

ii. (2.0 pt) Which of the following is largest?

# The chance that A is above 55%
The chance that B is above 55%
# The chance that A is above 60%
# The chance that B is above 60%

iii. (2.0 pt) Which Python expression evaluates to the probability that B is not 0 and not 1, but instead
a proportion between 0 and 1?
# 0
# 1
# 0.5 ** 40
# 1 - (0.5 ** 40)
# 0.5 ** 40 + 0.5 ** 40
1 - (0.5 ** 40 + 0.5 ** 40)
 12

(c) (8.0 points)

Complete the code below that uses a simulation repeated 10,000 times to estimate the chance that the
average dice outcome when rolling 5 fair 6-sided dice is within 0.5 of 3.5. (That is, larger than 3 and smaller
than 4.) For example, the average dice outcome of rolling (3, 2, 2, 6, 4) from the 5 dice is (3+2+2+6+4)/5
= 3.2, which is within 0.5 of 3.5.
def within(x, y, z):
"Return whether z is strictly within x of y."
return _________
(a)
count = 0
for i in np.arange(10000):
if within(0.5, 3.5, np.average(_________(_________))):
(b) (c)
count = count + 1

estimate = count / _________

(d)
i. (3.0 pt) Fill in blank (a). You may call the built-in function abs to compute the absolute value of a
number.

abs(z-y) < x or y-x < z < y + x or y-x < z and z < y + x

ii. (1.0 pt) Which of these could fill in blank (b)?

# sample_proportions
np.random.choice
# Table.sample
# within
# max
# min

iii. (3.0 pt) Fill in blank (c). You may include one or more commas.

np.arange(1, 7), 5

iv. (1.0 pt) Which of these could fill in blank (d)?

# counts
# trials
# len(count)
10000
 13

4. (24.0 points) Wordle

In the game of Wordle, a player guesses up to 6 words until they correctly guess the secret word of the day or
run out of guesses. Their guess count is either the guess number that was correct, 1 through 6, or X if all 6
guesses were incorrect.
For all 1,000 UC Berkeley students who played Wordle yesterday, we have collected the proportion of students
with each guess count. These proportions appear in the table below and an array called berkeley.

1 2 3 4 5 6 X
0.0 0.17 0.33 0.27 0.20 0.02 0.01

berkeley = make_array(0.00, 0.17, 0.33, 0.27, 0.20, 0.02, 0.01)

Wordle’s creator, Josh Wardle, sent us the proportion of guess counts for all players who tried to guess yesterday’s
word in an array called everyone.

1 2 3 4 5 6 X
0.0 0.09 0.25 0.32 0.28 0.03 0.03

everyone = make_array(0.00, 0.09, 0.25, 0.32, 0.28, 0.03, 0.03)

(a) (12.0 points)
Let’s investigate whether the distribution of guess counts for UC Berkeley students differs from the distri-
bution for all players on yesterday’s Wordle. Describe a hypothesis test that would aid this investigation.
i. (2.0 pt) Complete the null hypothesis: The distribution of guess counts for UC Berkeley students is
...
# uniform with a 1/7 chance for each possible guess count.
# like a random sample from a uniform distribution with a 1/7 chance for each possible guess count.
# different from a uniform distribution with a 1/7 chance for each possible guess count.
# the population of guess counts for all Wordle players.
like a random sample from the population of guess counts for all Wordle players.
# different from the population of guess counts for all Wordle players.

ii. (2.0 pt) Complete the alternative hypothesis: The distribution of guess counts for UC Berkeley
students is . . .
# the same as the distribution of guess counts for all Wordle players.
different from the distribution of guess counts for all Wordle players.
# the same as the uniform distribution.
# different from the uniform distribution.
 14

iii. (2.0 pt) Which test statistic is best for choosing between the null and alternative hypotheses?
# total guess count
# most common guess count
# guess count
total variation distance
# observed average

iv. (2.0 pt) Which line of code simulates a distribution of proportions for 1000 Berkeley students under
the null hypothesis?
# sample_proportions(1000, berkeley)
sample_proportions(1000, everyone)
# sample_proportions(1000, make_array('1', '2', '3', '4', '5', '6', 'X'))
# sample_proportions(1000, make_array(1/7, 1/7, 1/7, 1/7, 1/7, 1/7, 1/7))

v. (2.0 pt) How does increasing the number of times a distribution is simulated under the null hypothesis
affect the outcome of the hypothesis test?
# The probability that the null hypothesis is false will increase.
# The probability that the null hypothesis is true will increase.
# The observed distribution of guess counts for Berkeley students will be more similar to the
distribution for all players.
# The observed test statistic for Berkeley students will be more similar to the test statistic for all
players.
The empirical distribution of the test statistic under the null hypothesis will be more similar to its
theoretical distribution.

vi. (2.0 pt) If the null hypothesis is rejected because the p-value of this hypothesis test is very small,
what can we conclude? Select all that apply.
2 Attending Berkeley improves most people’s Wordle performance.
2 Attending Berkeley changes most people’s Wordle performance.
2 Attending Berkeley does not improve most people’s Wordle performance.
2 Attending Berkeley does not change most people’s Wordle performance.
 None of these.
 15

(b) (4.0 points)

Assume the observed test statistic is assigned to obs. We simulate under the null hypothesis 10,000 times
and append each simulated test statistics to an array sim. Complete this Python expression that computes
the p-value for this hypothesis test.
_________ ( _________ >= _________ ) / len( _________ )
(a) (b) (c) (d)
i. (1.0 pt) Fill in blank (a).

sum or np.count_nonzero

ii. (1.0 pt) Fill in blank (b).

sim

iii. (1.0 pt) Fill in blank (c).

obs

iv. (1.0 pt) Fill in blank (d).

sim
 16

(c) Define reading more as spending an extra two hours a day reading The New York Times, and a good game
of Wordle as one in which the player guesses the word in 3 or fewer tries. We want to test if reading more
leads to a higher proportion of good games.
Among the 1000 Berkeley students who played Wordle yesterday, 500 were selected at random (without
replacement) one month ago and asked to read more. All 1000 played yesterday’s Wordle, and the number
of guesses each student took was recorded.
i. (2.0 pt) How would a permutation test be used to investigate whether reading more leads to a higher
proportion of good games?
Repeatedly, all 1000 students would be partitioned at random without replacement into two groups
of 500, and the proportion of good games in those two groups would be compared for simulating a
null distribution.
# Repeatedly, all 1000 students would be partitioned at random without replacement into two groups
of 500, and within each group the proportion of good games for students who read more would be
compared to that of the students who didn’t.
# Repeatedly, the proportion of good games for students who read more would be compared with
the proportion of good games of a random permutation of those who didn’t.
# Repeatedly, the proportion of good games for students who read more would be compared with
the proportion of good games of a random permutation of all 1000 students.

ii. (2.0 pt) Suppose we consider the following alternative hypothesis: Among the 1000 students, the
proportion of good games would be higher if they all read more than if none of them read more.
Complete this null hypothesis: Among the 1000 students, the proportion of good games . . .
# would be lower for students who read more than for those who didn’t.
# for the 500 students who were selected to read more is the same as for the other 500 students.
# for students who read more would be 50%.
would be the same whether they all read more or none of them read more.

iii. (2.0 pt) Which of the following test statistics is best for choosing between the null and alternative
hypotheses above?
The difference between the proportion of good games in each group.
# The absolute difference between the proportion of good games in each group.
# The difference between the proportion of good games in the “read more” group and 0.5.
# The difference between the proportion of good games in the “didn’t read more” group and 0.5.

iv. (2.0 pt) When we conduct this permutation test, we compute a p-value of 0.002. Assume we had
chosen a p-value cut-off of 0.05. Which of the following can we conclude about the 1000 Berkeley
students based on this result? Select all that apply.
 Reading more increases the proportion of good games.
 There is an association between reading more and the proportion of good games.
2 Being a Berkeley student is a confounding factor for the association between reading more and the
proportion of good games.
2 None of these