CrossPropertyTL for Deterministic Analysis

This repository contains the code for performing cross-property deep transfer learning to predict materials properties. The DL code has been optimized to provide deterministic result. The input embeddings used for training are extracted using Hybrid-LLM-GNN framework . This repo provides the following functions:

• Train a ElemNet model on a given dataset using Tensorflow or Pytorch framework
• Predict material properties of new compounds with a pre-trained ElemNet model
• Deterministic analysis based on prediction results

Installation Requirements

The basic requirement for using the files are a Python 3.9 with the packages listed in requirements.txt

Source Files

Here is a brief description about the folder content:

• elemnet/dl_regressors_tf2_dtm.py: code for training ElemNet model using Tensorflow framework from scratch with deterministic behavior controlled.

• elemnet/dl_regressors_torch.py: code for training ElemNet model using PyTorch framework from scratch with deterministic behavior controlled.

• elemnet/dl_regressors_torch_predict.py: code for predict on unseen dataset using pretrained ElemNet model built with PyTorch framework with deterministic behavior controlled.

• CrossPropertyTL_cystal_run.ipynb: Example notebook demonstrating the full pipeline, including dataset download, environment setup, model training from scratch, and inference with pretrained models.
  Also available on Colab. The convenient way to do a quick-run on different hardware setups.

• data: 39 different datasets used for training ElemNet model. Transfer learning datasets can be downloaded from figshare
  
  

• determinstics_analysis.ipynb: Jupyter Notebook to perform analysis based on test set predictions.

Run ElemNet model

The code to run the ElemNet model is provided in the elemnet folder. In order to run the model you can pass a sample config file to the dl_regressors_tf2.py from inside of your elemnet directory:

python dl_regressors_torch.py --config_file elemnet/sample/example_alignn_matbert-base-cased_robo_prop_mbj_bandgap_local.config

The config file defines all the related hyperparameters associated with the model training and model testing such as loss_type, training_data_path, val_data_path, test_data_path, label, input_type, etc.

For model inference with pre-trained model, you need to set 'model_path' [e.g. model/sample_model]. After training, you will get the following files:

• The output log from the training will be saved in the log folder as log/sample-run_example_tf2.log file.

• The trained model will be saved in model folder.

Note: Your results should remain deterministic (bit-exact) on a single machine across repeated runs but may differ across machines due to GPU parallelism and hardware-specific variation in numerical execution.

Developer Team

The code was developed by Youjia Li from the CUCIS group at the Electrical and Computer Engineering Department at Northwestern University.

Acknowledgements

The open-source implementation of ElemNet here provided significant initial inspiration for the structure of this code-base.

Disclaimer

The research code shared in this repository is shared without any support or guarantee on its quality. However, please do raise an issue if you find anything wrong and I will try my best to address it.

email: [email protected]

Copyright (C) 2025, Northwestern University.

See COPYRIGHT notice in top-level directory.