Hey! I am making my Ph.D. in NLG to explain Bayesian Networks inference.
I was playing with your code to solve some almost real-life complex questions with the power of Decision Networks.
I was able to do that but some parts of the results did not seem very intuitive to me, so I tried to think about how to make it explainable.
I simplified the Network just to two evidence nodes, one decision, and one Utility node.
I used your code (CID class) and my simplified version of code suitable for only this kind of simple conditions, verified that I get the same utility, and then I had an idea of how to get the entity that would make the calculated utility more understandable.
That is what I did
To make it clear, these two tables correspond to possible actions that can be selected.
Each column corresponds to possible scenarios that can happen, and then for each scenario, we have the corresponding utility, the probability, and the utility weighted wrt the probability. When we have this we can, for example, sort the list of weighted utilities and see which elements corresponding to which scenario affect most of all the expected utility.
So my question or probably a feature request is whether I can get something like that from your code.
I tried to dive into your code to understand whether I can get something similar from it. Probably I need to spend more time on it but for now, I can see that the point that looks closer to what I want is StochasticFunctionCPD which stores every combination of evidence nodes and utility value corresponding to it. If you think is feasible within your code and give me an incentive from where to start I can consider implementing the feature and making a pull request. If this is already doable in your code I will also be happy=)
Thank you in advance!
Hey! I am making my Ph.D. in NLG to explain Bayesian Networks inference.
I was playing with your code to solve some almost real-life complex questions with the power of Decision Networks.
I was able to do that but some parts of the results did not seem very intuitive to me, so I tried to think about how to make it explainable.
I simplified the Network just to two evidence nodes, one decision, and one Utility node.
I used your code (CID class) and my simplified version of code suitable for only this kind of simple conditions, verified that I get the same utility, and then I had an idea of how to get the entity that would make the calculated utility more understandable.
That is what I did
To make it clear, these two tables correspond to possible actions that can be selected.
Each column corresponds to possible scenarios that can happen, and then for each scenario, we have the corresponding utility, the probability, and the utility weighted wrt the probability. When we have this we can, for example, sort the list of weighted utilities and see which elements corresponding to which scenario affect most of all the expected utility.
So my question or probably a feature request is whether I can get something like that from your code.
I tried to dive into your code to understand whether I can get something similar from it. Probably I need to spend more time on it but for now, I can see that the point that looks closer to what I want is StochasticFunctionCPD which stores every combination of evidence nodes and utility value corresponding to it. If you think is feasible within your code and give me an incentive from where to start I can consider implementing the feature and making a pull request. If this is already doable in your code I will also be happy=)
Thank you in advance!