Probabilistic Graphical Models, Fall 2024

[cwyoo]

Class projects correspondences in Probabilistic Graphical Models, Fall 2024. Please post your weekly proposed outcome here every Friday.

[samanthagonzales]

Weekly outcomes 25 October 2024:

- generated simulation datasets with interventions at each variable and for each state to compare inference results with. Datasets include close9 and sparse9 variables, and are attached here for reference.
- code associated with the class project is on GitHub: https://github.com/scasgonzal/causal-di ... tervention (I might need to move it to SMLG for it to be viewable)

Next weeks goals:
- Work on implementing inference based on Gibbs sampling on each nodes Markov blanket.

[Sharika]

Here is my updated proposal for the project.

[samanthagonzales]

Weekly outcomes 15 November 2024:

Working on implementing Gibbs sampling with application to EE-ILVS - I made some significant progress in the sampling code, and should be able to complete it by the weekend. I've posted a draft of the code here for feedback/review (note its using the Sprinkler network as a test case because the code is for the weekly discussion).
Received updated ideas on the EE-ILVS methodology that I can focus on.

When I initially simulated the experimental datasets, I created a file for each state being intervened - For example, mAnaphylaxis[0] and mAnaphylaxis[1] are two separate datasets with 100 cases each - so I will have to concatenate them to generate a single dataset with n cases, with n/2 cases mAnaphylaxis[0] and n/2 cases mAnaphylaxis[1]. This shouldn't be too difficult to modify.

I plan to do the following this weekend:
- Update simulation datasets to contain the number of intervention cases proportional to the number of states of the intervened variable
- Update proposal to reflect updates to methodology improvement, simulated datasets
- Complete Gibbs sampling implementation for generating data and performing inference with Markov blanket

And the following next week:
- Begin implementing ILVS on one specific dataset for testing