See log for assignment and due dates.
HW1 1.1, 1.2. 10 pts.
HW2 1.3, 1.4, 1.5 15 pts---5 pts per part.
HW3 1.6, 1.7, 1.8 15 pts--5pts per part. 1.6: 1 pt off per wrong table;
1.7: (i): 1pt off per wrong table; (ii): take off at most 2pts for errors;
1.8: (i) take off at most 2 pts; 1pt if "admittance theorem" is used; (ii):
take off at most two points for errors.
HW4 1.9, 1.10 10 pts---5 pts per part. 1.9: 2 points taken off per wrong
(or missing) subpart; 2 points per wrong (or missing) subpart.
HW5 Complete the table for the two "coins and bell" networks. 8 points total.
One point off for each wrong entry. If the tables for the two networks are
different, take off 5 points.
HW6 (Hugin) Exercise 1.11 and 1.12. 5 points in all. (1.11) 2 point
if the student reports
installation of Hugin (any version) on his or her PC; or, if the student
announces his or her intention of using the version in the lab.
(1.12) 3 points for a screen dump. 2 points for a net file instead.
One point off with the wrong DAG.
Commendation for two screen dumps or a screen dump showing two nets.
Extra commendation for screen dumps
indicating that evidence was entered.
HW7 (Hugin) Exercises 1.13, 1.14, and 1.15. 15 points.
Check that three nets are
submitted. Check the structure. If wrong, take out up to three points.
HW8 Complete the variable (bucket) elimination exercise on the dyspnea
network. 5 points. Forgetting to remove eliminated variables after a sum: -2
points; removing the wrong variable (i.e., not the one that is in the sum):
-2.
HW9 Ex.2.2: 3 points. Check structure. If wrong, take off two points.
Ex. 2.3: 3 points. If wrong answer, take off two points.
Ex. 2.4: 4 points. If wrong answer, take off two or three points.
Ex. 2.6, 6 points. 3 for part (i), 3 for part (ii). Check especially that the
numbers computed in part(i) are used in part (ii).
Ex. 2.7, 1 point. Is the second model there?
HW10 Ex.2.5: 5 points. One point off for each wrong table, up to 4 points
off.
Ex.2,8: (i) 3 points
The probability of infected milk is the same for each one of the
days, with the probabilities in the book. In particular, P(Inf_{i+1}=yes |
Inf_i = no) = 0.0002. Some students incorrectly used 0.0005 instead,
and they noticed instability: P(Inf1) = 0.07, and P(Inf7) = 0.13.
Take 2 points
off for this.
(ii) 2 points. Full credit for "logical" argument; full credit for algebraic
proof.
(iii) 2 points. Full credit for "logical" argument; full credit for algebraic
proof.
2.9 2 points.
2.10 (i) 2 points
(ii) 2 points
(iii) 2 points
HW11 Ex. 3.15 [J96; breakfast]: 10 points. 1 point off for not modeling
the cutlery
dependence situation properly. 3 points off for edges backwards. 1 point off
for separate nodes for the two types of breakfast (i.e., for not enforcing the
condition that only one type of breakfast is possible). 3 points off for not
distinguishing observation from actual value.
Ex. 2.13 [J00; Monty Hall]: 5 points. 2 points off for wrong answer.
HW12 Ex. 3.1 5 points. Minor algebraic errors will cost 1-2 points.
Ex. 3.2(i) 3 points. Minor algebraic errors will cost 1-2 point.
Ex. 3.2(ii) 2 points.
HW13 (a---counts as homework)
Ex. 4.7 5 points. 1 point off if sum is not done at chance node; 1 point
off if max is not taken at decision node. 1 point off if utilities are not
multiplied by the probabilities. 1 point off if correct policy is not
highlighted, or the policy is not indicated in some way.
Ex. 4.8(i) 5 points--similar to 4.7. Wrong tree: 1 point off.
HW13 (b---counts as Hugin assignement)
Ex. 4.8(ii) 5 points. Two points off for wrong answer.
Ex. 4.8(iii) 5 points. Up to two points off for the wrong network.
Two groups of
students appear to have shared a solution for the whole assignment:
they have made the same mistakes. In several cases, students misread a "7" as
a ">" sign! In any case, it appears that these students did not understand
what they were doing, and they will receive a grade of 0 on this assignment.
Project (Bucket Elimination)
Parsing of net file to get the CPTs in a data structure: 25%
Putting CPTs in buckets, initially: 35%
Computing conformal tables correctly: 15%
Updating buckets: 20%
Getting the right probability: 5%.