.

maybe down this link

http://www.kbs.twi.tudelft.nl/docs/syllabi/Dynamic%20Basian%20Network.pdf

.

Dean Fabrice, thank you a lot for the document.

Ines, I'm afraid GeNIe/SMILE cannot do this quite yet but please do stay tuned.  We are about to release modeling capability for mixtures of discrete and continuous nodes in static networks.  DBNs with this mixture will follow.  At the moment, you should discretize your continuous DBN nodes.  I assume you know how to use autodiscretization to help you in deriving CPTs for these nodes?

Fabrice's link is to a MSc thesis done in my lab that outlines GeNIe's implementation of a GUI for DBNs.  It is quite different from what other implementations offer.  GeNIe's GUI makes creating higher order dynamic models natural and easy.

All my greetings Professor Mark. I looked for such response. Thank you. Ines

Marek Jozef Druzdzel Does it now? If so, where is the documentation?

Thank you Professor Rafael,

No not now. It was 2 years ago.

Now every thing is well. But if you have relevant materials you can provide me with it.

Professor Marek Jozef Druzdzel ,

I would like to know, if it is possible, if now GeNIe has the capability for mixtures of discrete and continuous nodes in static networks.

I have been trying to model a bayesian network from a database (with discrete and continuos data) and the algorithms implemented in GENIE ( Bayesian Search, PC, Greedy Thick Thinning, Tree Augumented Naïve Bayes, Augmented Naive Bayes and Naive Bayes) don't allow to model the network from continuos data.

It is true that PC algorithm constructs the network from continuos data, but it doesn't works with discrete and continuos data simultaneously.

If you could respond to my question, or indicate me some documentation in order to investigate about the theme, I will be grateful

Sorry for the inconvenience,

Thank you in advance

Hi Lidia,

GeNIe can model now truly hybrid networks, i.e., networks that contain mixtures of discrete and continuous variables. It uses stochastic forward sampling for inference and, in case evidence is entered in a node with parents, performs discretization and turns the network into a discrete Bayesian network, just for the purpose of inference.

Your question, however, is about structure learning from data that are a mixture of discrete and continuous variables. For that, you need to discretize the continuous variables so that all variables are discrete. The only algorithm that can deal with continuous variables is the PC algorithm but in this case, all variables have to be continuous and follow the multi-variate normal distribution. This limitation is theoretical rather than a limitation in implementation -- there are no reliable general purpose independence tests for other cases.

I hope this helps,

Marek

Thank you very much Professor Marek Jozef Druzdzel .

The response has been helpful for the project