mmml efield-evaluate¶
Evaluate external electric-field PhysNet.
Usage¶
mmml efield-evaluate --help
Options¶
usage: mmml efield-evaluate [-h] [--params PARAMS] [--config CONFIG]
[--data DATA] [--test-npz TEST_NPZ]
[--output-dir OUTPUT_DIR] [--batch-size BATCH_SIZE]
[--num-test NUM_TEST] [--model-config MODEL_CONFIG]
[--features FEATURES] [--max-degree MAX_DEGREE]
[--num-iterations NUM_ITERATIONS]
[--num-basis-functions NUM_BASIS_FUNCTIONS]
[--cutoff CUTOFF]
[--max-atomic-number MAX_ATOMIC_NUMBER]
[--electrostatics-damping-sigma ELECTROSTATICS_DAMPING_SIGMA]
[--save-output-npz] [--output-h5 PATH]
[--rot-augment]
[--rot-perturbation ROT_PERTURBATION]
Evaluate trained model
Input & configuration:
--config CONFIG Path to config JSON file (will be auto-detected from
params UUID if not provided)
--data DATA Path to dataset NPZ file
--model-config MODEL_CONFIG
Path to model config JSON (deprecated: use --config)
Scientific model:
--num-basis-functions NUM_BASIS_FUNCTIONS
Number of basis functions (default: 64)
--cutoff CUTOFF Cutoff radius (default: 10.0)
Execution:
--batch-size BATCH_SIZE
Batch size for evaluation
--electrostatics-damping-sigma ELECTROSTATICS_DAMPING_SIGMA
Override learned-charge Coulomb erf damping sigma; set 0
to disable
Output & artifacts:
--output-dir OUTPUT_DIR
Output directory for plots and metrics
--save-output-npz Save evaluation outputs (predictions, targets) to NPZ
file
--output-h5 PATH Write HDF5 trajectory for mmml gui
(R,Z,N,E,E_pred,F,F_pred,Dxyz,Dxyz_pred,Ef). Requires
h5py.
Diagnostics & safety:
-h, --help show this help message and exit
Other options:
--params PARAMS Path to parameters JSON file (can be params-UUID.json or
params.json)
--test-npz TEST_NPZ Test split NPZ (alias for --data; same as ef-train
--test-npz)
--num-test NUM_TEST Number of test samples to use (None = use all)
--features FEATURES Model features (will be inferred from params/config if
not provided)
--max-degree MAX_DEGREE
Max degree (default: 2)
--num-iterations NUM_ITERATIONS
Number of iterations (default: 2)
--max-atomic-number MAX_ATOMIC_NUMBER
Max atomic number (default: 55)
--rot-augment Apply random SO(3) rotation augmentation when building
batches (via prepare_batches)
--rot-perturbation ROT_PERTURBATION
Rotation perturbation strength in [0, 1] (used with
--rot-augment)