XRDiff

An XRD-conditioned crystal structure generator, built on top of DiffCSP.

Setup

python==3.8.13
torch==1.9.0
torch-geometric==1.7.2
pytorch_lightning==1.3.8
pymatgen==2023.8.10

Rename .env.template to .env and set:

PROJECT_ROOT: the absolute path of this repo
HYDRA_JOBS: the absolute path to save hydra outputs
WABDB_DIR: the absolute path to save wandb outputs

Training

<dataset> is alex_mp_20_poly or oqmd. Common options for all XRDiff models:

COMMON="data=<dataset> data.xrd_bins=0.1 data.for_exp=true model.decoder.xrd_cond=true model.decoder.xrd_cond_type=film model.decoder.film_mode=mp"

# full-spectrum conditioning
python diffcsp/run.py $COMMON data.xrd_type=full expname=xrdiff_film

# peak-list conditioning
python diffcsp/run.py $COMMON data.xrd_type=peaks expname=xrdiff_film_peaks

# full spectrum + RRUFF-calibrated augmentation
python xrd_scripts/augment.py --pt_path data/<dataset>/train_xrd_0.1_for_exp.pt   # static: peak shapes / intensities
python diffcsp/run.py $COMMON data.xrd_type=full data.xrd_augmented=true model.xrd_noise_augment=true expname=xrdiff_film_aug

The augmentation constants are fitted on the RRUFF calibration set by scripts/rruff_eval/calibrate.py and hard-coded in xrd_scripts/augment.py.

Data

• Alex MP 20 Poly
• OQMD Poly
• RRUFF

Evaluation

Test sets (Alex-MP-20-Poly, OQMD-Poly): see scripts/evaluation/README.md.

python scripts/evaluation/evaluate.py --model_path=<run_dir> --eval_split=test --num_evals=3
python scripts/evaluation/compute_metrics.py --root_path=<run_dir> --eval_split=test --label=n_evals_3 --multi_eval

RRUFF-148 (20 candidates per sample, ranked by Rietveld refinement against the measured pattern; see scripts/rietveld_ranking/README.md):

bash scripts/rruff_eval/rank20.sh <run_dir> rruff_pv <name>              # full-spectrum models
bash scripts/rruff_eval/rank20.sh <run_dir> rruff_pv_peaks <name>   # peaks models

The rruff_pv_peaks peak lists are built by scripts/rruff_eval/build_rruff_pv_peaks.py.

Acknowledgments

Built on DiffCSP.