problems building networks based on a database

[jam]

I have tried to build a Bayesian network from a database, apparently genie takes the data and unites them at random without being able to make some sort of constraint for the construction that one wishes, and tried with all methods from the naive to pc, which is required prior knowledge and apparently to prevent certain variables join, they appear like a bow I would like to know how I want to make and not the genie who choose default.

[shooltz[BayesFusion]]

I have tried to build a Bayesian network from a database, apparently genie takes the data and unites them at random

GeNIe certainly does not 'unite data at random'. All available data is read and used by learning algorithms.

The naive algorithm doesn't allow any background knowledge, but you can specify forced/forbidden arcs and tiers for GreedyThickThinning or PC (click the 'Background Knowledge' button when specifying the parameters for learning algorithm). Note that we'll be releasing new implementation of PC learning soon.

[jam]

The method used and the naive but not as prohibit between nodes with this method, and the method pc at the time to ban these conditions appear as a union between nodes, or other wing genie takes random and it is here where I have problems that No other option is to take pra build the network that meets my needs I would like to know that I have other options in Genie.

[shooltz[BayesFusion]]

I'm not sure what exactly is your problem. Did you used automatic translator? If so, please ask somebody who knows English to translate for you.

[jam]

I want to know that other options exist apart from the background knowledge and parameter naive to build my network without the Genie randomization

[shooltz[BayesFusion]]

I want to know that other options exist apart from the background knowledge and parameter naive to build my network without the Genie randomization

What do you mean by 'parameter naive'?

[jam]

Simple is a method that does not require basic knowledge, the interesting thing the network is totally contrary to the desire to validate.

On the oder hand, , in the help of genie appears one thing, in my software appears otherwise, I would like to know why. Additions a document

[shooltz[BayesFusion]]

On the oder hand, , in the help of genie appears one thing, in my software appears otherwise, I would like to know why. Additions a document

The screenshot in help file shows old dialog, which had all parameters for all structure learning algorithms visible. They were just enabled/disabled depending on the selection of algorithm. The 'Class Variable' combo box would be enabled only when Naive Bayes algorithm was selected. It would be disabled when the choice of algorithm was PC (this is shown in the screenshot).

Current dialog displays only controls relevant to selected algorithm. For PC, there will be no 'Class Variable' - it'll only show up if you select Naive Bayes algorithm.

Simple is a method that does not require basic knowledge, the interesting thing the network is totally contrary to the desire to validate.

I don't quite understand what you're trying to say here.

[jam]

The parameter naive doing everything contrary to what we wish to undertake, and i do not understand why.

Thankyou

ATT JAm

[shooltz[BayesFusion]]

The parameter naive doing everything contrary to what we wish to undertake, and i do not understand why.

Can you post your data/learned network and explain what's the problem with the net?

[jam]

Ok no problem.

Sending the table with varables and the different possibilities that genie offers for the construction of the network.

The idea is that the nodes called Female_per, Total_7_pe, Women_per, Shan_Ethni; should be directed to the node called Off_Alw_Fo without nodes that name at the beginning were not crossing each other.

Thanks. JAM

[shooltz[BayesFusion]]

The naive Bayes algorithm can't be applied in your case - it creates simple model with arcs directed from 'class node' into all other nodes. Since you're interested in structure where arcs go into specified node (Off_Alw_Fo), the PC algorithm should work. In fact, you've obtained the model which is close to your expected structure.

I guess you're concerned with two arcs:
Total_7_pe->Female_per
Shan_Ethni->Women_per

Can you simply delete them after learning is complete? Currently GeNIe doesn't allow for bidirectional forbidden arcs.

Your other option would be creating the structure manually (since you seem to know what it should like), then learn the parameters with EM algorithm. You'll need to discretize your data file before running EM.

[jam]

I can give more information about the EM algorithm.

Thanks again for your help has been invaluable

[shooltz[BayesFusion]]

Check this and/or GeNIe's help file:

http://genie.sis.pitt.edu/wiki/GeNIe_Tu ... Parameters

[jam]

Could give me a link to more talk of EM algorithm, or any pdf.

Thanks.

JAM