Author Topic: Doubts about configuring a bayesian network  (Read 12024 times)

Offline eb20

  • Newbie
  • *
  • Posts: 9
    • View Profile
Doubts about configuring a bayesian network
« on: July 26, 2011, 18:26:56 »
Hi,

I'm trying to configure the bayesian network I have attached:

For B node: I have mu and variance for (B | A1) and other mu and variance for (B | A2)
For C node: I have mu and variance for each value of A too (C | A1 B) and (C | A2 B)

In the tables for C:
- I suppose that I have to set the variance value in variance (this is obvious) and the mu value in B. Is this right?
- There is another state called intercept and I don't know what that means. What value do I set in this state?

Thank you in advance.

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2295
    • View Profile
Re: Doubts about configuring a bayesian network
« Reply #1 on: July 27, 2011, 13:47:04 »
In a Bayesian network with both discrete and continuous nodes, each continuous node is assumed to follow a conditional linear Gaussian (CLG) distribution.

If a continuous node has only discrete parents, then you specify a mean and variance of the child for each configuration of the parents.

If a continuous node has both discrete and continuous parents, then for each discrete configuration you specify a variance for the child and a set of parameters that encode the mean of the child as a linear function of the continuous parents. Hence, for the mean you specify a constant (intercept) and a (regression) coefficient for each continuous parent.

In your example node C is assumed to have a mean value that is a linear function of continuous parent B conditional on the states of A such that mu_C = Intercept + coefficient_B * B

Hope this helps.
HUGIN EXPERT A/S

Offline eb20

  • Newbie
  • *
  • Posts: 9
    • View Profile
Re: Doubts about configuring a bayesian network
« Reply #2 on: July 27, 2011, 22:45:24 »
Got it. Thank you very much for your help!