cjpcool/metamaterial-MetaModulus

📊 Metamaterial MetaModulus Dataset

🧾 Dataset Summary

This dataset contains 3D metamaterial lattice structures for predicting mechanical modulus properties, including Young's modulus, Shear modulus, and Poisson's ratio. Each sample is preprocessed into a format compatible with PyTorch Geometric (PyG) for downstream machine learning tasks such as structure-property prediction, graph representation learning, and lattice optimization.

Check our paper on arxiv.org/abs/2505.20299 for more details.

---

📦 Dataset Usage

You can easily download and convert the dataset into torch_geometric.data.Data objects using the following steps.

Step 1: Load and Convert to PyG Format

from datasets import load_dataset
from torch_geometric.data import Data
import torch

dataset = load_dataset("cjpcool/metamaterial-MetaModulus", split="full")

pyg_data_list = [
    Data(
        frac_coords=torch.tensor(d["frac_coords"]),
        cart_coords=torch.tensor(d["cart_coords"]),
        node_feat=torch.tensor(d["node_feat"]),
        node_type=torch.tensor(d["node_type"]),
        edge_feat=torch.tensor(d["edge_feat"]),
        edge_index=torch.tensor(d["edge_index"]),
        lengths=torch.tensor(d["lengths"]).unsqueeze(0),
        angles=torch.tensor(d["angles"]).unsqueeze(0),
        vector=torch.tensor(d["vector"]).unsqueeze(0),
        y=torch.tensor(d["y"]).unsqueeze(0),
        young=torch.tensor(d["young"]).unsqueeze(0),
        shear=torch.tensor(d["shear"]).unsqueeze(0),
        poisson=torch.tensor(d["poisson"]).unsqueeze(0),
        num_nodes=d["num_nodes"],
        num_atoms=d["num_atoms"]
    )
    for d in dataset
]

Step 2: Create Train/Valid/Test Splits

from sklearn.utils import shuffle

def get_idx_split(data_size, train_size=8000, valid_size=2000, seed=42):
    ids = shuffle(range(data_size), random_state=seed)
    train_idx = torch.LongTensor(ids[:train_size])
    val_idx = torch.LongTensor(ids[train_size:train_size + valid_size])
    test_idx = torch.LongTensor(ids[train_size + valid_size:])
    return {'train': train_idx, 'valid': val_idx, 'test': test_idx}

split = get_idx_split(len(dataset), seed=42)
train_data = [pyg_data_list[i] for i in split["train"]]
valid_data = [pyg_data_list[i] for i in split["valid"]]
test_data = [pyg_data_list[i] for i in split["test"]]

Step 3: Create PyG Dataloaders

from torch_geometric.loader import DataLoader

train_loader = DataLoader(train_data, batch_size=32, shuffle=True)
valid_loader = DataLoader(valid_data, batch_size=32, shuffle=False)
test_loader  = DataLoader(test_data, batch_size=32, shuffle=False)

---

📚 Dataset Source

• Repository: github.com/cjpcool/Metamaterial-Benchmark
• Paper: MetamatBench: Integrating Heterogeneous Data, Computational Tools, and Visual Interface for Metamaterial Discovery, KDD 2025 (Datasets and Benchmarks Track)

---

📌 Citation

If you use this dataset in your research, please cite the following paper:

@inproceedings{metamatBench,
  author={Chen, Jianpeng and Zhan, Wangzhi and Wang, Haohui and Jia, Zian and Gan, Jingru and Zhang, Junkai and Qi, Jingyuan and Chen, Tingwei and Huang, Lifu and Chen, Muhao and Li, Ling and Wang, Wei and Zhou, Dawei},
  title     = {MetamatBench: Integrating Heterogeneous Data, Computational Tools, and Visual Interface for Metamaterial Discovery},
  booktitle = {Proceedings of the 31th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)},
  year      = {2025},
  publisher = {ACM},
  doi       = {10.1145/3711896.3737416},
}

🧪 Raw Data Source

We acknowledge the original data creators:

@article{2021lumpeExploring,
  title     = {Exploring the property space of periodic cellular structures based on crystal networks},
  author    = {Lumpe, Thomas S and Stankovic, Tino},
  journal   = {Proceedings of the National Academy of Sciences},
  volume    = {118},
  number    = {7},
  pages     = {e2003504118},
  year      = {2021},
  publisher = {National Acad Sciences}
}

---

📬 Contact

For questions, feedback, or contributions, please reach out to:
📧 cjpcool [at] outlook [dot] com