mason800
Hello everyone,

Does anyone know how to build a Dynamic Bayesian Network (DBN) Model? I understand the concept of DBN but I don't know how to create a DBN model. The tutorial provided by BayesiaLab that can be found from link below does not help with creating model. 

https://library.bayesia.com/display/BlabC/Dynamic+Bayesian+Networks

Any help is appreciated.

Thanks! 
Quote 0 0
Dan
Hi,

As described in the link you mentioned, there are two types of DBNs:
  • Unfolded: you have to create one "Bayesian Network fragment" per time slice
  • Compact: you have to create one "Bayesian Network fragment" for at least 2 consecutive time slices (1st Order Markovian assumption). Set the links that connect the Temporal Clones as Temporal Relations (right click on the link)
The unfolded version is used in the exactly as usual Bayesian Networks. Inference can be done in all directions. However, you need to know exactly the horizon you are interested in when you create your network.

The compact version is a special network. When you switch to Validation mode, you have new tools in the toolbar to make inference by increasing the Temporal Index. There is no limitation in the horizon (i.e. you can simulate as many time steps as you like) , but you can obviously only see the inference results on the current time slices.

Hope this helps!
Quote 0 0