This preview shows pages 1–3. Sign up to view the full content.
CS195f Homework 2
Mark Johnson and Erik Sudderth
Homework due at 2pm, 1st October 2009
The frst question asks you to analyse the Following naive Bayes model that describes the
weather in a mythical country.
Y
=
{
night
,
day
}
X
1
=
{
cold
,
hot
}
X
2
=
{
rain
,
dry
}
P(
X
1
,X
2
,Y
) = P(
Y
)P(
X
1

Y
)P(
X
2

Y
)
P(
Y
=
day
) = 0
.
5
P(
X
1
=
hot

Y
=
day
) = 0
.
9
P(
X
1
=
hot

Y
=
night
) = 0
.
2
P(
X
2
=
dry

Y
=
day
) = 0
.
75
P(
X
2
=
dry

Y
=
night
) = 0
.
4
Question 1:
For each of the following formulae except the ±rst, write an equation which de±nes it in
terms of formulae that appear earlier in the list. (For example, you should give a formula
for
P(
x
1
,x
2
)
in terms of
P(
x
1
,x
2
,y
)
). Then given the model above, calculate and write out
the value of the formula for possible each combination of values of the variables that appear
in it.
a)
P(
x
1
,x
2
,y
)
.
b)
P(
x
1
,x
2
)
.
c)
P(
y

x
1
,x
2
)
.
d)
P(
x
1
)
.
e)
P(
x
2
)
.
f)
P(
x
1

x
2
)
.
g)
P(
x
2

x
1
)
.
1
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Documenth)
P(
x
1

x
2
,y
)
.
i)
P(
x
2

x
1
,y
)
.
Are
X
1
and
X
2
conditionally independent given
Y
? Are
X
1
and
X
2
marginally independent,
integrating over
Y
? Provide a short proof for both answers.
Consider a binary categorization problem, and let
p
(
y
i

x
i
) denote the posterior distri
bution of the latent class label
y
i
∈ {
0
,
1
}
given observation
x
i
. Suppose that the classiFer
ˆ
y
(
x
i
) is allowed to make one of three decisions: choose class 0, choose class 1, or “reject”
this data (refuse to make a decision). We can use a Bayesian decision theoretic approach to
tradeo± the losses incurred by incorrect decisions and rejections.
Question 2:
Suppose that the classiFer incurs a loss of
0
whenever it chooses the correct class, a loss of
1
whenever it chooses the wrong class, and a loss of
λ
whenever it selects the reject option.
Express the optimal decision rule
This is the end of the preview. Sign up
to
access the rest of the document.
 Spring '09
 johnson
 Machine Learning

Click to edit the document details