Tutorial-III | Machine Learning | CSO504 | Monsoon 2022-2023 | 15/11/2022

1. Consider a scenario where grade obtained by a student depends on the difficulty of the
course and how intelligent the student is. Students got admission on GRE based on his
intelligence. Finally, instructor issues the recommendation letter based on the grade
obtained in the subject. Draw the causal graph.
2. Here is a graphical model that models the grades obtained by a student in a course and its
implications. Difficulty represents the difficulty of the course and intelligence is an
indicator of how intelligent the student is, GRE represents the GRE scores of the student
and Letter presents the event of the student receiving a letter of recommendation from the
faculty teaching the course. The random variables given in the model are modeled as
discrete variables

(Difficulty = D, Intelligence = I, Grade = G, GRE = E, Letter = L) and the corresponding

Conditional Probability Distributions are given in the table below. There are two levels of
difficulty D0 and D1, the student may be intelligent (I0) or not so intelligent (I1), there are
three grades (G0, G1, G2), the student may or may not get a letter (L0 and L1) and student
may or may not get a good GRE score (E0 and E1).

P(D0)=0.6 P(L1|G0) = 0.9 P(D1) = 0.4 P(I0) = 0.7 P(I1) = 0.3

P(G0|I0,D0) = 0.3 P(G0|I0,D1) = 0.05 P(G0|I1,D0) = 0.9 P(G0|I1,D1) = 0.5 P(G1|I0,D0)

= 0.4

P(E0|I0) = 0.95 P(E0|I1) = 0.2 P(E1|I0) = 0.05 P(E1|I1) = 0.8 P(L0|G0) = 0.1 P(L1|G0) =
0.9 P(L0|G1) = 0.4 P(L1|G1) = 0.6 P(L0|G2) = 0.99 P(L1|G2) = 0.01

P(G1|I0,D1) = 0.25 P(G1|I1,D0) = 0.08 P(G1|I1,D1) = 0.3 P(G2|I0,D0) = 0.3 P(G2|I0,D1)

= 0.7 P(G2|I1,D0) = 0.02 P(G2|I1,D1) = 0.2
 Based on the above data

(i). Find P(I1, D0, G2, E1, L0)

(ii). Find P(L0|I0)

(iii). Find P(I0|L0)

3. Consider the following figure for clusters found after performing flat clustering:(k-Means)
on a set of documents. The gold standard for each document is produced by human judges.
Each document belongs to one of the three gold standard classes (x, o and +). Calculate
the NMI and Purity quality measures for the above clustering.

4. Consider a perceptron network with one hidden layer having three units. The output layer
consists of one unit. Each input is represented via four features. Learning rate is 0.01 and
it optimizes the squared error loss function. Now apply it over following dataset and run
for six rounds using gradient descent and stochastic gradient descent. Show the updated
parameters after each round. Assume initial values of parameters are 0.5.

Sr. X0 X1 X2 X3 Price
No

P1 1 0.9 0.6 0.7 1

P2 1 0.6 0.9 0.7 1

P3 1 0.2 0.1 0.3 0

P4 1 0.4 0.1 0.1 0

5. Consider an environment having six grids [Fig1]. Starting from one of the grids an agent
has to reach the goal state (G). If an agent moves to a goal state from another state, it will
get a reward of 100. In all other movements return a reward of 0. The 𝛾 = 0.85 Compute
maximum discounted cumulative reward that any agent can obtain starting from any state
to the goal state (G).

6. Consider a distribution of points (1,2), (1,3), (2,5), (6,-1),(7,-3),(4,-3),(5,-1),(2,6),(3,4),(5,-

5). Assume initial centroids (2,5) and (5,-5). What would be cluster centroids after three
rounds of iterations? What would happen if we assume initial centroids as (5,-5) and (7,-
3).

7. Consider a two class classification problem [+1,-1] over a sample of males and females.
The confusion matrix is given below:

True Label
Predicted Label +1 -1
+1 70 30
-1 40 60
The confusion matrix for the male class is as follows:

True Label
Predicted Label +1 -1
+1 50 10
-1 35 55
Does this satisfy equal opportunity property?