-
Notifications
You must be signed in to change notification settings - Fork 14
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Explain source of the obtained utility #69
Comments
Hi Nikolay, Thanks for reaching out! I think you can do something similar to what you’re trying to do in PyCID, but you’re right that making it an explicit feature would be helpful! I think the easiest thing will be for you to create a Google Colab notebook with your example above in it, and then share it with me (either on here or direct via my email I can then edit that colab notebook to suggest what one can currently do with PyCID. We can then perhaps chat over a call/liase over email about what new features would be most useful for your work – and yes, any additional features can then be merged into the codebase. Best, |
Sorry for the long response time. I did not expect you would respond that quick=) You may reach me at [email protected] or here as well. |
Hi @bbkjunior I assume this was resolved? I just had a look a look at your colab and I couldn't immediately see that you were pointing out any issues? It looked like a nice application though. Feel free to polish that colab slightly more and then create a pull request to add it to the collection of notebooks https://github.com/causalincentives/pycid/tree/master/notebooks. It's good to have a range of notebooks displaying a wide range of applications of this codebase :) |
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!
The text was updated successfully, but these errors were encountered: