CSE655: Probabilistic Reasoning

Assignment # 5

Date: January 10, 2010

Due Date: January 15, 2010

Problem 1

  • Using the BN model you developed as part of Assignment # 3 (Part 2), generate a data set of 2000 records and learn the structure of a BN using BN Power Constructor. How much the learned model matches with the actual model (You can define your own criterion to match similarity between two models).
  • Repeat the exercise after generating data sets of 5000 and 10,000. Does increasing the sample size has any impact on the similarity level?

Problem 2

  • Test the sensitivity of node “H” on the root nodes {“A”, “B”, “C”, “D”} in the following Bayesian Network.
  • Use the following two techniques:
    • Entropy based calculations (Unit # 14, Slide # 3)
    • Sensitivity Analysis as done in Influence Nets (Unit # 14, Slide # 6)
  • Sort the root nodes in terms of their influences on node “H”.

Problem 3

  • Consider the above Timed Influence Net. TIN.jpg
  • Suppose action “A” is taken at time 4 and “B” at time 3. Draw the probability profile of node D. To simplify the calculations, CPTs are provided instead of the CAST logic parameters.
  • Using the transformation technique discussed in “From Dynamic Influence Nets to Dynamic Bayesian Networks: A Transformation Algorithm” paper, transform this Timed Influence Net into an equivalent Dynamic Bayesian Network.