Evaluating Local-Source Models
Use scripts/eval_local_charges.py to label extxyz files with predictions from
local-source models.
Command
Typical usage:
python scripts/eval_local_charges.py \
--model split_charge_example_stage1.model \
--configs input.xyz \
--output labelled.xyz \
--batch_size 16 \
--device cuda \
--pbc_handling slab \
--external_field_key homogeneous_field \
--fermi_level_key VBM
formal charges per atom
As described in training, one can train a split charge model where the oxidation state is an input instead of being fixed per species. If you have trained such a model, you also need to specify the oxidation state per atom when evaluating the model. You can do this with --formal_charges_key.
Outputs
With the default MACE_ prefix, the script writes outputs such as:
MACE_energyMACE_forcesMACE_density_coefficientsMACE_dipoleoptional
MACE_polarizabilityoptional model contribution keys