How txie-93/cdvae works
CDVAE learns periodic crystal structures from curated datasets, then supports reconstruction, latent-space generation, and property-guided inverse design. The documented workflow is configured training, followed by model-backed evaluation and metric computation. The graph covers data preparation and the principal model components; evaluation links are based on the README and source index. Some internal model wiring and unsampled evaluation behavior are not shown, so those connections are omitted.
Main entry points: eval_utils.py Evaluation utilities · datamodule.py Data module · run.py Training entry · evaluate.py Task evaluation
Components 17
User workflow
- Training entry
cdvae/run.py - Run configuration
conf
Crystal data
- Crystal datasets
data - Data module
cdvae/pl_data/datamodule.py - Crystal datasets
cdvae/pl_data/dataset.py - Crystal preprocessing
cdvae/common/data_utils.py
Generative model
- CDVAE model
cdvae/pl_modules/model.py - Graph encoders
cdvae/pl_modules/gnn.py - Crystal decoder
cdvae/pl_modules/decoder.py - GemNet network
cdvae/pl_modules/gemnet/gemnet.py - Message-passing layers
cdvae/pl_modules/gemnet/layers - Element embeddings
cdvae/pl_modules/embeddings
Evaluation
- Task evaluation
scripts/evaluate.py - Evaluation utilities
scripts/eval_utils.py - Metric computation
scripts/compute_metrics.py - Evaluation artifacts
Other components
- Researcher
Connections 20
- Researcher → Training entry: starts training
README.md - Training entry → Run configuration: uses configuration
README.md - Training entry → Data module: trains with data
- Crystal datasets → Data module: provides splits
README.md - Data module → Crystal datasets: instantiates
cdvae/pl_data/datamodule.py - Crystal datasets → Crystal preprocessing: preprocesses crystals
cdvae/pl_data/dataset.py - CDVAE model → Crystal preprocessing: uses crystal utilities
cdvae/pl_modules/model.py - CDVAE model → Element embeddings: uses embeddings
cdvae/pl_modules/model.py - Crystal decoder → GemNet network: uses network
cdvae/pl_modules/decoder.py - GemNet network → Message-passing layers: uses layers
cdvae/pl_modules/gemnet/gemnet.py - Message-passing layers → Element embeddings: embeds atoms
cdvae/pl_modules/gemnet/layers/embedding_block.py - Researcher → Task evaluation: selects tasks
README.md - Task evaluation → Evaluation utilities: loads evaluation helpers
scripts/evaluate.py - Evaluation utilities → Crystal datasets: builds tensor dataset
scripts/eval_utils.py - Evaluation utilities → Data module: loads data module
scripts/eval_utils.py - Evaluation utilities → CDVAE model: loads checkpoint
scripts/eval_utils.py - Task evaluation → Evaluation artifacts: writes task outputs
README.md - Researcher → Metric computation: requests metrics
README.md - Metric computation → Evaluation utilities: uses evaluation utilities
scripts/compute_metrics.py - Metric computation → Evaluation artifacts: writes metrics
README.md