Build a Bayesian network of your invention with six

PROJECTS ON
SUPERVISED AND
FACTORIZATION
BAYESIAN NETWORKS
Concha Bielza, Pedro Larrañaga
Universidad Politécnica de Madrid
Course 2007/2008
Hugin Lite 6.7
Factorization Exercise
1. Build a Bayesian network of your invention with six
nodes and binary variables
How to Build BNs (Hugin GUI Help)
2. Generate 50, 100, 200 and 400 cases from the
previously built Bayesian network
Case Generator (Hugin GUI Help)
3. Structure learning with PC and NPC algorithms with
two level of significance (0.05 and 0.10)
Structure Learning (Hugin GUI Help)
4. Parameter learning with the EM learning algorithm
EM learning (Hugin GUI Help)
Hugin Lite 6.7
-----------------------------------------------------------------------PC
NPC
-----------------------------------------------------------------------0.05
Simulation 50
0.10
-----------------------------------------------------------------------0.05
Simulation 100
0.10
----------------------------------------------------------------------0.05
Simulation 50
0.10
-----------------------------------------------------------------------0.05
Simulation 100
0.10
----------------------------------------------------------------------
Hamming distance between the structure of the original
Bayesian network, and the one obtained after learning
BAYESIA
Factorization Exercice
• Generate two data bases (50 and 500 instances
and different percentage of missing data)
from the “Asia.xbl” Bayesian network
2. Apply the following learning algoithms:
“EQ”, “SopLEQ”, “Tabo” and “TaboOrder”
to both data bases
3. Compare the induced Bayesian networks with the
“Asia.xbl”
4. Obtain information in Internet about the learning
algorithms
Weka
Factorization Exercice
1. Using the “tips-discrete-cfs9.arff” dataset
2. Learn Bayesian network structures with:
- Conditional independence tests
- Local search
- Global search
3. Estimate the parameters:
- Simple estimation
- BMA estimator
BAYESIA
Supervised Exercice
1. Generate 3 files (100, 200 and 400 cases) from the
“Asia.xbl” Bayesian network
2. Choose variable “Cancer” as the class (target) variable
3. Induce the following classifiers:
Naive Bayes
Augmented naive Bayes
Markov blanket
4. Compare the accuracies of the different models in the
3 files
Weka
Supervised Exercice
1. Open the file “tips-discrete-cfs9.arff”
2. Learn naive Bayes and TAN models
3. Obtain the corresponding accuracies
with a 10-fold cv validation method
4. Repeat the exercice with a FSS method
(Select Attributes in Weka)