This repository contains code and supplementary material for the paper "Modeling multiple adverse pregnancy outcomes:\ Learning from diverse data sources" AIME (2024).

• Eliciting initial BN structure from LLM.pdf contains the prompt and the response from Google Gemini that was used to construct the initial BN.
• black_list.csv contains the list of temporally impossible edges. These edges are excluded from the local search used to refine the initial BN.
• structure_learning.py is a module that implements the BN refinement procedure.
• refine.py reads an initial BN structure and a data set from disk and calls the refinement procedure to obtain a refined BN.

• pgmpy==0.1.24
• scikit-learn==1.1.3