Hi,

I am trying to learn the parameters for the following test BN using
OpenBaye:

$python2.4 bayesian_network_test.py
BNet Test Bayesian Network
Vertices:
hod (discrete, 24)
b (discrete, 5)
os (discrete, 4)
top (discrete, 7)
ur (discrete, 5)
r (discrete, 5)
cr (discrete, 5)
c1 (discrete, 15)
dr (discrete, 5)
dow (discrete, 7)

Edges:
0: dow (discrete, 7) -> c1 (discrete, 15)
1: hod (discrete, 24) -> c1 (discrete, 15)
2: r (discrete, 5) -> c1 (discrete, 15)
3: cr (discrete, 5) -> c1 (discrete, 15)
4: dr (discrete, 5) -> c1 (discrete, 15)
5: ur (discrete, 5) -> c1 (discrete, 15)
6: top (discrete, 7) -> c1 (discrete, 15)
7: b (discrete, 5) -> c1 (discrete, 15)
8: os (discrete, 4) -> c1 (discrete, 15)

initializing the distributions...
creating interference Engine
begin learning cases
Traceback (most recent call last):
File "bayesian_network_test.py", line 46, in ?
ie.LearnMLParams(cases)
File "/opt/local/Library/Frameworks/Python.framework/Versions/2.4/
lib/python2.4/site-packages/OpenBayes/inference.py", line 63, in
LearnMLParams
v.distribution.incrCounts(case)
File "/opt/local/Library/Frameworks/Python.framework/Versions/2.4/
lib/python2.4/site-packages/OpenBayes/distributions.py", line 227, in
incrCounts
self.counts[index] += 1
File "/opt/local/Library/Frameworks/Python.framework/Versions/2.4/
lib/python2.4/site-packages/OpenBayes/table.py", line 150, in
__getitem__
return self.cpt[numIndex]
IndexError: Index out of range

Any ideas what is going wrong? Right now I am just trying to learn
from a dummy test cases with only 2 samples.

You can see above what my Network looks like, the distributions
initialize properly, the IE used is MCMC, and here is the content of
my 2 samples cases:

cases = [ {'dow': 1, 'hod':13, 'r':3, 'cr':4, 'dr':3, 'ur':2, 'top':
2, 'b':1, 'os':1, 'c1':2},
{'dow': 5, 'hod':5, 'r':2, 'cr':1, 'dr':5, 'ur':2, 'top':1, 'b':4,
'os':2, 'c1':5} ]

Thank you in advance for your guidance.

Cheers!

Sébastien Arnaud

i have run into errors in that line of code before. I might have some ideas about where it is going wrong. However, my computer died on me a week ago, and i am just starting to get it back (lets hope), but i definitely don't have any kind of environment up yet. As soon as i get everything setup i will look at it and let you know.

-Elliot

I'm trying to localize your problem. Could you please send me a python script file that reproduces the error?

Please try to isolate the problem. By this I understand that the BN you will sent me must be the simplest possible that still reproduces the error.

Kosta

This bug has now been fixed!

Actually there were 2 distinct bugs: 1 in the JoinTree and the other one in the MCMCEngine.

Small description of the bugs :
JoinTree: After Learning the parameters of the network the clusters where not reinitialized to take into account the changes made by the learning engine. EasyFix

MCMCEngine: In order to perform forward sampling to acquire the samples of a BN we first need to make a topological sort of the network (starting from the nodes without parents and ending to the nodes without children). There was a small bug in the topological sort that lead to errors in certain configurations. Succesfully fixed by first finding a node without parents and then perform a classical topological starting from that node instead of starting at a random node (as was done before)

Thank you sebastien for helping us tracking down these.

Kosta

Thank you Kosta for getting those corrected. It is working great now!

Sorry if I haven't been posting on the forums my messages back to you. Email is just too easy sometimes

Sébastien