Feature Generation Configuration#
- pydantic model dpeva.config.FeatureConfig[源代码]#
Configuration for Feature Generation Workflow.
- field batch_size: int = 1000#
- field data_path: Path [Required]#
Path to dataset.
- field embedding_dtype: Literal['fp32', 'fp64', 'native'] = 'fp32'#
Output dtype used by DeepMD dp embed.
- field feature_exporter: Literal['eval_desc', 'embed'] = 'eval_desc'#
Feature export implementation for CLI mode. eval_desc preserves legacy .npy descriptor output; embed writes DeepMD HDF5 embeddings.
- field feature_kind: Literal['descriptor', 'fitting_last_layer'] = 'descriptor'#
Feature type to export. fitting_last_layer requires Python mode unless feature_exporter='embed' is used.
- field mode: Literal['cli', 'python'] = 'cli'#
- field model_head: str | None = None#
Model head name (optional for single-task models).
- field model_path: Path [Required]#
Path to model file.
- field output_mode: Literal['atomic', 'structural'] = 'atomic'#
Descriptor output format. Options: atomic (per-atom features) or structural (global features).
- field savedir: Path | None = None#