University of Waterloo

School of Computer Science

CS 486/686, SAMPLE Midterm Examination
Winter 2025

Name:

Waterloo Student ID:

• Instructor: Victor Zhong & Jesse Hoey

• Date: Any day before Feb 13th 700pm

• Location: Anywhere you want

• Time: Anytime before Feb 13th 700pm

• There are 4 questions on this exam

• There are 10 pages in this exam (including cover page and two additional
pages at the end for work that does not fit in the spaces provided)

• There are a total of 100 marks on the exam

• You have 110 minutes to complete the exam

• ONLY non-programmable calculators are allowed

Question 1 2 3 4 TOTAL
Marks 20 30 30 20 100

Score
 Question 1: Short Answers (20 marks out of a total of 100 marks)

WRITE ANSWERS IN THE SPACES PROVIDED AFTER EACH QUESTION

(1a) [4 marks ] Complete the following sentence: A* search uses a priority queue of nodes
ranked by , and so is a mixture of and best-first
search.

(1b) [4 marks ] A heuristic for a search problem is an estimate of the true cost to the goal.
Is the following statement true or false: an admissible heuristic is always greater than
the true cost. Briefly explain

(1c) [4 marks ] True or False: in a deterministic planner, causal rules specify when a fea-
ture keeps its value when not acted upon. Briefly explain why.

(1d) [4 marks ] Why is variable elimination for constraint satisfaction problems intractable?

(1e) [4 marks ] Give one reason why Heuristic Depth-First Search is often used in practice.
 Question 2: Constraint Satisfaction (30 marks marks out of a total of
100 marks)

The KoobeCaf™ social network consists of a number of members who can be connected (or
not) to each other. Connected members on KoobeCaf™ can be either both friends (f ) or
enemies (e) with each other. Members come in two types: positive (+) and negative (−).
While connected members of different types must be enemies, members of the same type
can be friends or enemies. A stable social network is one in which all members satisfy these
constraints. This can be represented as a constraint satisfaction problem (CSP) as follows:
• variables and domains: Each KoobeCaf™ member is represented with a tuple containing:
their type (+ or −) and whether they are friends (f ) or enemies (e) with each of their (up
to N − 1) connections. Thus, a member b of a network with three fully connected members
a, b, c (in that order), would be represented with a variable B = Tb Xa Xc where Tb ∈ {+, −}
is b’s type and Xa , Xc ∈ {f, e} are b’s relationship with a and c, respectively. Thus, B has
domain {+f f, +f e, +ef, +ee, −f f, −f e, −ef, −ee}, where −f e means b is type negative,
friends with a, but enemies with c. The order of the friend/enemy relations in each tuple is
alphabetic.
• constraints: A binary constraint between each pair of connected members that requires
the two members to (1) have the same relation (e.g. in the 3-member network A,B,C above,
if A = ∗∗f then C = ∗f ∗ where ∗ means either value); and (2) be of the same type OR
be enemies. Call this constraint H(·, ·). For example, with three fully connected members
represented by variables A, B, C: H(A, C) = T rue for
{A, C} ∈ {{+∗f, +f ∗}, {+∗e, +e∗}, {−∗f, −f ∗}, {−∗e, −e∗}, {+∗e, −e∗}, {−∗e, +e∗}}.

This graph shows a KoobeCaf™ network with three members: a, b and c, in which a is positive
(+), and b is negative (−) and domain values of the wrong type have been removed from
the domains of A and B.

A
{+f f , +f e
+ef , +ee}
⟨A, H(A, B)⟩
⟨A, H(A, C)⟩
H(A, B) H(A, C)
⟨B, H(A, B)⟩
⟨C, H(A, C)⟩
C
B
{+f f, + f e, +ef,
{−f f , −f e H(B, C)
⟨B, H(B, C)⟩ ⟨C, H(B, C)⟩ +ee, − f f, −f e
−ef , −ee}
−ef, − ee}

(2a) [25 marks ] Using AC-3, make the CSP shown above arc-consistent by filling in the
table on the facing page, in which each iteration is on a single line in the table and a
check-mark is under each constraint that is consistent (i.e. the arc is not in the
(Question 3 CONTINUES ON NEXT PAGE...)
 TDA queue), and the domain values remaining for each variable are shown under each
variable name. Always choose the left-most (in the table) inconsistent constraint to
make consistent next. You may not need all rows, but you will not need more rows.
Don’t add redundant arcs back on the TDA (after changing the domain of X for the
arc ⟨X, c(X, Y )⟩, you only add arcs ⟨Z, c′ (Z, X)⟩ to TDA where Z ̸= Y).

⟨B, H(B, C)⟩

⟨B, H(A, B)⟩
⟨A, H(A, B)⟩

⟨C, H(B, C)⟩

⟨A, H(A, C)⟩

⟨C, H(A, C)⟩
A = Ta Xb Xc B = Tb Xa Xc C = Tc Xa Xb
Ta = + Tb = − Tc ∈ {+, −}
Xb , Xc ∈ {f, e} Xa , Xc ∈ {f, e} Xa , Xb ∈ {f, e}
+f f +f e +ef +ee −f f −f e −ef −ee +f f +f e +ef +ee
−f f −f e −ef −ee
+ef +ee ✓

(2b) [5 marks ] Based only on your arc-consistent network (last line of the table), can you
guarantee that there is a solution (a stable network)? Briefly explain why or why not.
 Question 3: Search (30 marks marks out of a total of 100 marks)

WRITE ANSWERS IN THE SPACES PROVIDED AFTER EACH QUESTION

An artificial intelligence conference has Nc + Ns attendees, of which Nc do research on
connectionist AI (“connectionists”) and Ns do research on symbolic AI “symbolicists”. On
their way to the conference reception, they come across a river with no bridge that they
must cross. There is a boat that can carry only 2 people at a time. If more connectionists
than symbolicists are left on either side of the river, the connectionists will start a fight with
the symbolicists. At least one person must be in the boat on each trip (to paddle the boat).
The goal is to move everyone to the other side of the river in the shortest number of trips
without a fight starting. Assume Nc < Ns to start with.
The state space can be easily described by the number of connectionists and symbolicists
on each side, plus the location of the boat. The cost function, g(n) is simply 1 for each trip
across the river. A simple heuristic h(n) for this problem is to relax it by ignoring fights
that break out.

(3a) [10 marks ] Give a formal (mathematical) definition of a state, neighborhood function,
cost function and the heuristic function described above.

(3b) [5 marks ] Is the h(n) defined above admissible? Carefully explain why or why not.
(4c) [15 marks ] Show part of the search graph that results from applying algorithm A* for
the problem with Nc = Ns = 3 starting from the configuration where everyone is on one
side, with the heuristic h(n) and cost g(n) defined above. Specifically, continue until
you have expanded (generated the successors of) four nodes. Break ties arbitrarily.
Number the four nodes you expand to indicate the order in which they are expanded
(“1”,”2”,”3”,”4”). Then, number (with “5”) the next leaf that you would expand if
you were to continue with A* search on this graph. Label each node with its g(n)
value and its h(n) value. You should label all nodes, not just the ones you expand. Do
not add a node to the tree if it is already in the tree somewhere with a lower or equal
value of f (n) = g(n) + h(n). Branches of your search graph should terminate (have an
infinite heuristic value) if the connectionists outnumber the symbolicists on either side
of the river.
 Question 4: Decision Trees (20 marks out of a total of 100 marks)

You are trying to build a system to predict whether a stock price will rise or fall, using
the following dataset of historical measurements of four attributes: the season, whether an
election is taking place, whether the price of oil (oil price) has risen or fallen recently
whether the price of the stock (stock price) in question rose or fell.

election season oil price stock price

no winter rise rise
no summer fall fall
no summer rise rise
yes winter rise fall
no winter fall fall
yes summer rise rise
yes summer fall fall
yes winter fall fall

(4a) [15 marks ] Using information gain (weighted by fractions of documents) as your
feature selection mechanism, construct a decision tree that predicts if the stock price
will rise or fall based on the attributes above. Continue splitting until no more attributes
are available, or until all data agrees on the target attribute (stock price change). Show
your work and draw a graph of your final decision tree.
(4b) [3 marks ] What prediction would your decision tree give for the following test data

election season oil price stock price

no summer fall
no winter rise
yes summer rise

(4c) [2 marks ] Suppose now you received the following data points and added them to the
data in Q(1), and re-learned the decision tree using information gain to split.

election season oil price stock price

yes summer rise fall
yes winter rise rise

Would your decision tree change? If so, what would the change be? Would your
classification of the data in (b) change? If so, show the new classification. If not,
explain why not.
ADDITIONAL WORK - Clearly state which question you are working on
ADDITIONAL WORK - Clearly state which question you are working on