weighting the probabilities

The front end.
Post Reply
Sadir
Posts: 4
Joined: Tue Oct 28, 2014 9:57 pm

weighting the probabilities

Post by Sadir »

Hello,
I have a BN consisted from two branches, the target node of the first branch contains a probabilities of ordered states (zero, low, med, high, V high) which represent levels of impact of the network input variables. Based on those probabilities, I want to make an estimation for the execution time of the network.
My idea her is to propose a weight for each probability of the target node, I mean add (0 for Zero), (1 for low),(2 for medium)...and so on; then multiplying each outcome probability with the corresponding weight and finally make summation of the results.

The resulted value will be input for the next node, that contains equation.
The question her, is it possible to implement this estimation in the aforementioned way?

I appreciate your help in advance
Regards
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: weighting the probabilities

Post by marek [BayesFusion] »

Sadir,

I need to know a little more about your model. What do you mean by "branches"? A Bayesian network is a graph, not a tree, so you will need to clarify this. It seems that you want to model the problem by means of a formula. Have you tried continuous/equation nodes? There is also another way to model a simple formula, along the lines that you propose: MAU nodes, which also allow for formulas in the context of a Bayesian network. Generally, however, I have a hard time understanding the details of your model. Would it be too much to ask you to share the model with us?
Cheers,

Marek
Sadir
Posts: 4
Joined: Tue Oct 28, 2014 9:57 pm

Re: weighting the probabilities

Post by Sadir »

Thanks Marek for your reply

I used a "tree" vocabulary just for clarification. Now I hope to be lucky to describe my question in different way.
suppose I have a simple Bayesian network consisted of three nodes linked together; node (A=root), (B=parent), and (C=child).
All variables of root level (A) will converge all together in node(B) (which represent a set of ordered states (zero, low, med, high, V high)); and node (B) will be linked to the child node (C)(which is equation node).
My question is: I want to convert the probabilities of node (B) to time by multiplying each level of probability (zero, low, med, high, V high) with a weights (0,1,2,3,4) in order to get an estimation for the required time, and use it in node (C) for final calculation.

regards
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: weighting the probabilities

Post by marek [BayesFusion] »

Again, I have a hard time understanding what you want to do. Do you mean the following structure? A->B->C?

If I understand correctly, you have no problems with the CPT in B but are interested in generating the probability distribution in C in such a way that it reflects time. First question: Have you considered a dynamic Bayesian network (DBN)? What is the formula that you would like to use to generate the CPT in C?
Cheers,

Marek
Sadir
Posts: 4
Joined: Tue Oct 28, 2014 9:57 pm

Re: weighting the probabilities

Post by Sadir »

Hi,

My question concentrated on the output of node(B) only.
In the network structure (A->B->C), if the states of node (B) are represented on distributed probabilities on different levels ( Zero, ...., Very High);
my question will be: can I use weights (0, ..., 4) with the states of node (B)?
I want to convert the resulted probabilities of node (B) to a (value) representing the estimated time as explained in the following formula:
((weight_1 * Stat_1) + (weight_2 * Stat_2) + ..... + (weight_5 * Stat_5) = value).

Regards
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: weighting the probabilities

Post by marek [BayesFusion] »

It seems that what you are describing is a utility node (node C). If you enter the weights instead of utilities, you will get exactly the formula that you want, i.e., the probabilities of different states weighted by the utilities. Utility nodes calculate the expected utilities. I hope this helps.
Cheers,

Marek
Sadir
Posts: 4
Joined: Tue Oct 28, 2014 9:57 pm

Re: weighting the probabilities

Post by Sadir »

Hi marek,

I highly appreciate your replies to me; also might I found the first step for answering my question in the literature "Generating Conditional Probabilities for Bayesian Networks: Easing and Knowledge Acquisition Problem".

Thank's and Kind regards
marek [BayesFusion]
Site Admin
Posts: 430
Joined: Tue Dec 11, 2007 4:24 pm

Re: weighting the probabilities

Post by marek [BayesFusion] »

I'm not sure whether you are interested in reasoning or model building. If you want to use your formula for building a model, you will have to use your own program and link it with SMILE. The solution that I gave you applies to inference. I hope this helps.
Cheers,

Marek
Post Reply