遇见数据集

SymmRelax processed binary-crystal graph dataset, trained checkpoints and source data

收藏
Zenodo2026-06-06 更新2026-06-05 收录
官方服务:

资源简介:

This record contains the processed dataset, trained model checkpoints and source data associated with the manuscript “Symmetry Preserving Pre-Relaxation Framework for High-Throughput and Reliable Crystal Screening”. The processed dataset is provided as processed_sqlite.tar.xz. After extraction, it contains the SQLite graph dataset, crystallographic metadata, preprocessing summary and fixed train/validation/test split files used for SymmRelax training, inference and evaluation, including data.sqlite, meta.csv, skip_summary.csv, split_train.npy, split_val.npy and split_test.npy. The record also includes two trained checkpoints. ckpt_symmrelax.pt is the trained SymmRelax geometry checkpoint used for symmetry-preserving pre-relaxation. model_best.pth.tar is the trained CGCNN-EAF formation-energy predictor trained on DFT-relaxed structures from the full 524,049-structure benchmark using the original 8:1:1 split. This checkpoint corresponds to the best-performing DFT-relaxed-structure input model reported in Extended Data Fig. 2 and was used as the frozen DFT-trained evaluator for the transfer analysis of initial and SymmRelax-predicted structures in Extended Data Fig. 3. SourceData.zip contains the source data underlying the figures, Table 1, Extended Data and selected Supplementary Figures and Tables. It includes source data for Fig. 1, Fig. 3, Fig. 4, Fig. 5, Extended Data Fig. 1, Extended Data Fig. 2, Extended Data Fig. 3, Extended Data Tables 2–3, Supplementary Fig. 1, Supplementary Table 1, Supplementary Table 9 and Supplementary Table 10. Shared source-data folders are used where the same files support multiple display items, including Table 1/Fig. 3, Fig. 5/Supplementary Table 9 and Extended Data Fig. 1/Extended Data Table 2. Fig. 2 is a conceptual schematic of the SymmRelax architecture and does not have associated numerical source data. Supplementary Tables 4–8 summarize model, training, baseline, downstream-prediction and evaluation settings; these are reproducibility metadata and are also represented in the configuration files and scripts available in the SymmRelax code repository. The original crystal entries were collected from publicly available materials databases, including Alexandria, the Materials Project, OQMD and AFLOW, as described in the manuscript. Users should comply with the terms of use of the original databases when using or reconstructing raw crystal structures. The source code, preprocessing workflow, training scripts and inference scripts are available at:https://github.com/zrfcms-buaa/SymmRelax

提供机构:
Zenodo
创建时间:
2026-05-31
二维码
社区交流群
二维码
科研交流群
商业服务