Deep Learning-Based Causal Discovery Algorithms for pgmpy

This proposal aims to implement three deep learning-based causal discovery algorithms in pgmpy: DiffAN, GraN-DAG, and CAREFL. Each algorithm leverages neural networks to go beyond classical score-based or constraint-based methods, enabling discovery on non-linear, non-Gaussian data. All three will be implemented inside pgmpy/causal discovery to isolate the soft dependencies (PyTorch, diffusers, nflows) from the rest of the library.