### Author Topic: Continous Distributions  (Read 10505 times)

#### Dave

• Newbie
• Posts: 1
##### Continous Distributions
« on: June 23, 2011, 18:36:10 »
I am using Hugin Lite to investigate the capabilities.

I am Struggling to understand How to integrate a continous chnage node into a BBN. The help and tutorial are very sparce with information.

I uderstand from the help that a Continous node cannote be the parent of a discrete cahnce node, consequtnly i am at a loss to usderstrnd how to use it.

In my naivety I am Trying to implement the following

a Discrete chance node N_Trials - 20 equally likely States 0-19

a Continuous Chance Node P_Success Mean 0.5, Var, 0.01

and a Distrete Chance Node N_Succes for which I i want the table for N_Success to be the Expression Binomial(N_trials,P_Success)

am i totally on the wrong Track here?

#### Frank Jensen

• HUGIN Expert
• Hero Member
• Posts: 575
##### Re: Continous Distributions
« Reply #1 on: June 27, 2011, 20:08:37 »
The continuous chance nodes in Hugin must have a socalled Conditional Linear Gaussian distribution.  This means that, for each configuration of the discrete parents, the child must have a normal distribution with the mean being a linear finction of the continuous parents and the variance being a constant.  These restrictions permit inference to be performed exactly (as opposed to approximately).

If these restrictions are not satisfied, then Hugin also offers socalled "interval" nodes.  These nodes represent continuous ranges of real numbers divided into intervals.  The "table generator" facility in Hugin can be used to specify distributions for these nodes (Hugin has many built-in standard statistical distributions).

In your example, you can specify "P_Success" as an interval node: each state of the node represents a subinterval of the interval [0,1].  Then you can specify a distribution for the node (for example, a beta or a truncated normal distribution).