Validate Configuration¶
Source code:
bff/workflows/configs.pybff/workflows/validate.pybff/workflows/runsims.py
Purpose¶
bff validate reruns selected parameter samples, usually drawn from the
posterior learned by bff learn.
The campaign runtime keys intentionally match bff trainset as closely as
possible.
Minimal Example¶
trainset_dir: ./validation-trainset
parameters: ../05-learning/posterior-samples.yaml
specs: ../02-training-data/trainset/specs.yaml
systems:
- assets: ../01-prepare/colvars/ace-colvars/training/system-000
n_steps: 1000
- assets: ../01-prepare/colvars/ace-colvars/training/system-001
n_steps: 1000
gmx_cmd: gmx
job_scheduler: local
Top-Level Keys¶
trainset_dirOutput directory for the validation campaign.parametersYAML file containing explicit parameter samples.specsForce-field specification file used to reconstruct constrained parameters.systemsNon-empty list of prepared asset directories plus validation MD lengths.gmx_cmdGROMACS executable.job_schedulerEitherlocalorslurm.dispatchIftrue, launch jobs immediately after staging them.compressIftrue, compress finished simulation outputs.cleanupIftrue, remove temporary files after successful runs.storeWhich trajectory outputs to keep. Defaults to["xtc"].slurmOptional Slurm runtime configuration.
systems[] Keys¶
assetsDirectory created bybff prepare, for exampletraining/system-001.n_stepsProduction MD length for this validation run.
Parameter File Format¶
Validation consumes YAML only. The expected structure is a mapping from explicit parameter name to a list of sampled values:
charge C1: [-0.5, -0.4, -0.3]
charge O1 O2: [-0.7, -0.6, -0.5]
The implicit charge is reconstructed from specs.yaml, so it does not need to
appear in the file.