Graphical Models: Assignment 7
Due April 24, 2017
- This problem uses the data set from Problem 6. Agent Hydziz
I. Dennity collected data random sample of Depravians on PersonType,
Gender, and HairLength. The data he collected is provided here. Agent Dennity is considering these models for the relationship between PersonType, Gender and HairLength:
- A fully connected Baysian network, as shown below, with no restrictions on the local distributions.
- A Bayesian network in which HairLength depends on PersonType and Gender, but Gender does not depend on PersonType.
Assume uniform prior distributions on
all local distributions, as in the K2 algorithm. What is the
probability of the data given each of these structures?
- Find the probability of data given each of the following three structures:
- Fully connected Bayesian network in which the gender
distribution for government agents is the same as the gender
distribution for dissidents;
- Fully connected Bayesian network in which the hair length distribution is the
same for all women and the hair length distributions are the same
for male government agents, male government supporters, and apolitical
- Fully connected Bayesian network in which the restrictions for both part a and part b hold.
- If all five structures -- two from Problem 1 and three from
Problem 2 -- have equal prior probability, what is the posterior
probability of each of the structures?
- This problem concerns the example network from Unit 6, shown
below. We are interested in using likelihood weighting to
estimate the probability distribution of E given B=b2. A sample
of 250 cases was generated as follows:
- A was drawn from P(A)
- B was set to B=b2
- C was drawn from P(C|A), with A set to its sampled value.
- D was drawn from P(D|B,C), with B=b2 and C set to its sampled value.
- E was drawn from P(E|C), with C set to its sampled value.
The sampled values are given here
Use likelihood weighting to estimate the
probability distribution of E given B=b2. Compare with the exact
probability distribution. Comment.