Add dataset card

README.md

@@ -0,0 +1,54 @@
+---
+license: cc-by-4.0
+pretty_name: CHILI-3K (CHILI nanomaterials, unofficial mirror)
+tags:
+  - chemistry
+  - materials-science
+  - nanomaterials
+  - graph-machine-learning
+size_categories:
+  - 1K<n<10K
+---
+
+# CHILI-3K — unofficial Hugging Face mirror
+
+Raw archive of **CHILI-3K** from the CHILI dataset, mirrored to HF for convenient
+cloud access. This repo contains the original `.h5` files packaged as `CHILI-3K.zip`
+(one HDF5 file per crystal-type / composition; each holds all 5 nanoparticle sizes).
+
+## Attribution (please cite the original authors)
+
+> Friis-Jensen, U., Selvan, R., et al. *CHILI: Chemically-Informed Large-scale
+> Inorganic Nanomaterials Dataset for Advancing Graph Machine Learning.* KDD '24.
+
+- Paper: https://arxiv.org/abs/2402.13221
+- Code:  https://github.com/UlrikFriisJensen/CHILI  (Apache-2.0)
+- Data DOI: https://doi.org/10.17894/ucph.e37b6615-8635-49cf-819d-eae60e781a96
+
+## Licensing
+
+- **Data** (the `.h5` archive in this repo): **CC BY 4.0** — the license of the
+  original CHILI dataset/paper. You may share and adapt it *with attribution* to
+  the authors above. This is the repo's primary `license:` tag.
+- **Original CHILI code** (dataset class, generation scripts): **Apache-2.0**,
+  see https://github.com/UlrikFriisJensen/CHILI.
+- **Any loader/wrapper code added to this mirror**: **MIT** (covers only the
+  helper code, not the data).
+
+This mirror is **unofficial**; all credit to the original authors. Attribution is
+preserved as required by CC BY 4.0.
+
+## Contents
+
+- `CHILI-3K.zip` — original raw HDF5 archive from the DTU/ERDA repository.
+
+## Graph-level schema (`y` dict, per the paper's Table 2)
+
+Node `x = [atomic_number, atomic_radius, atomic_weight, electron_affinity]`;
+`edge_attr = [distance (Å)]`; `pos_abs` / `pos_frac` atomic coordinates.
+`y` holds crystal_type, space_group_*, crystal_system(_number), cell_params[6],
+the full unit-cell subgraph, and simulated scattering signals
+(nd, xrd, nPDF, xPDF, sans, saxs) — usable as targets **or** conditioning inputs.
+
+Load the original way with the authors' `CHILI` PyG dataset class, or unzip and
+read the `.h5` files directly with `h5py`.