PolyIDTM provides a framework for building, training, and predicting polymer properities using graph neural networks. The codes leverages nfp, for building tensorflow-based message-passing neural networ, and m2p, for building polymer structures. The notebooks have been provided that demonstrate how to: (1) build polymer structures from a polymer database and split into a training/validation and test set, (2) train a message passing neural network from using the trainining/validation set, and (3) evaluate the trained network on the test set. These three notebooks follow the methodology used in the forthcoming publication.
- Building polymer structures:
examples/1_generate_polymer_structures.ipynb
- Training a message passing neural network:
examples/2_generate_and_train_models.ipynb
- Predicting and evaluating a trained network:
examples/3_evaluate_model_performance_and_DoV.ipynb
Additional notebooks have been provided to provide more examples and capabilities of the PolyID code base.
- Checking domain of validity:
examples/example_determine_domain-of-validity.ipynb
- Generating hierarchical fingerprints for performance comparison:
examples/example_hierarchical_fingerprints.ipynb
- Predicting with the trained model:
examples/example_predict_with_trained_models.ipynb
Details for the methods are forthcoming in an upcoming manuscript.