UnconvBench

Benchmark for property prediction of unconventional crystal materials. For a comprehensive evaluation, we collected various datasets with 2D materials, MOFs, defected crystals and curated sets of bulk crystals.

Get started

Installation:

git clone https://github.com/howardwang1997/unconvbench
cd unconvbench
python setup.py install

Required packages can be found in requirements.txt

Datasets

Please find the datasets required on Figshare: https://figshare.com/articles/dataset/UnconvBench/26657428 and download them to directory datasets.

Run the benchmark

For easier usage, the workflow of this benchmark mostly complies with another benchmark for crystals named matbench. The usage of this benchmark and the API is similar with matbench. Example script to run this benchmark is shown in the file example_script.py.

Submit the result

Please include your model, the YourModel_result.json.gz in a subdirectory under benchmarks. You are encouraged to include a json file to introduce your model and the hyperparameters. The benchmark results will be shown in benchmarks/README.md.