Graph Learning for Chemical Process Flowsheets

GNN-based experimentation subpackage for analyzing and predicting properties of chemical process flowsheets using Graph Attention Networks (GATs).

See GNN_PROJECT_README.md for full documentation.

Quick Start

cd graph_learning
pip install -r requirements.txt
python quick_demo.py

Entry Points

Script	Description
quick_demo.py	Minimal end-to-end demo
main_pipeline.py	Full training pipeline (configurable via config.yaml)
demo_graph_generation.py	Graph generation with VAE, link prediction, and node type prediction
example_workflow.ipynb	Interactive tutorial notebook
graph_generation_deep_dive.ipynb	Detailed graph generation walkthrough

Package Structure

src/
├── data/           # Data loading, feature extraction, graph building
├── models/         # GNN architectures (GAT, GraphVAE, link prediction)
├── training/       # Training loops, generation trainers, utilities
├── evaluation/     # Metrics, graph metrics, weighted scoring
├── inference/      # Prediction, Gemini LLM explanations
├── agent/          # Flowsheet design agent, dialogue management
└── deployment/     # Production generator

Configuration

Edit config.yaml to customize model architecture, training parameters, and scoring. Data is loaded from ../exported_flowsheets/ (the parent directory's shared data).