Command-Line Interface¶
The CLI entry point is implemented in
bff/cli.py.
Public Commands¶
bff prepare CONFIG.yamlStage equilibrated systems, reference inputs, and training assets.bff reference CONFIG.yamlRun staged CP2K reference calculations from prepared assets locally or through Slurm.bff trainset CONFIG.yamlRun a sampled GROMACS campaign from prepared assets.bff qoi CONFIG.yamlCompute quantities of interest from trainset and reference trajectories.bff train CONFIG.yamlFit surrogate models from analyzed QoI datasets.bff learn CONFIG.yamlRun posterior inference from selected trained surrogate models.bff validate CONFIG.yamlRerun selected parameter samples for validation.bff examplesCopy or download the repository example tree.bff versionPrint the installed package version.
bff md exists as an internal low-level command used by scheduled campaign
jobs and is intentionally hidden from normal workflow navigation.
Config Philosophy¶
Each top-level workflow uses one focused config file:
- prepare config: how to stage systems and reusable assets
- reference config: how to execute staged CP2K reference assets
- trainset config: how to turn prepared assets into a sampled trainset
- QoI config: how to compute observables from trajectories
- train config: how to fit surrogates
- learn config: which models to use for posterior inference
- validate config: how to rerun chosen parameter samples
Detailed key-by-key documentation is in the configuration reference.
Shell Completion¶
When bff runs inside an activated conda environment, it writes a small
completion hook for bash and zsh into that environment. The hook is deliberately
minimal: bff <TAB> shows the public workflows and completes config-file
arguments for commands that expect a path.
After the first bff run, reactivate the environment once:
conda deactivate
conda activate bfflearn
After that, bff <TAB> should offer:
preparereferencetrainsetqoitrainlearnvalidateexamples