On Fri, 25 Aug 2000, Ruan Tong wrote:

> 
> Hi,Everybody. 
>     Thanks in advance if anybody would like to help me.
>      
>     1. Is there some bayesian software which can help to construct 
> dynamic bayesian network? and is there any tutorial or guidance on it?
>  
>     2. When constructing DBN, is there some method to deal with the 
> change of variable's domain in different time-slice? For example variable 
> v1 has three possible value in time slice t1, t2, t3, and in time-slice 
> t15, it has 5 possible value, and theoretically, variable v1 can have 
> 10000 possible values.
> A real world application of such a suituation is user's interests of some 
> kinds of information. There thousands categories of information, but in 
> some particular periods, user may only focus on very limited domain. 
>  
>     
>     Thanks in advance if anybody could help me.
>  
> 
> 
> 
  For the 2nd question, there is a method to dynamically repartition the
range of a continuous variable into an 'optimal' number of partitions,
using the Minimum Descriptive Length metric, during BBN creation. This
method can also be used to create static partitions before BBN creation.
The method is described in:
  Clarke,E., and Barton,B., (2000), Entropy and MDL Discretization of
Continuous Variables for Bayesian Belief Networks, International Journal
of Intelligent Systems, 15, 1, 61-92.
  The method is designed to decrease the number of partitions, or variable
values, for a possible better metric score for BBN structure creation.
This may help with dynamic repartitioning. 
  Good luck.
  Ellis
_____________________________________________________________________
  Ellis Clarke, Ph.D.; CSEE, University of Maryland Baltimore County; 
  [EMAIL PROTECTED]
_____________________________________________________________________

Reply via email to