ase_uhal.committee_calculators.MACEHALCalculator

class ase_uhal.committee_calculators.MACEHALCalculator(mace_calculator, committee_size, prior_weight, num_layers=-1, invariants_only=True, batch_size=None, **kwargs)

__init__(mace_calculator, committee_size, prior_weight, num_layers=-1, invariants_only=True, batch_size=None, **kwargs)

Parameters:

• mace_calculator (mace.calculators.MACECalculator object) – MACE architecture to use to define a MACE descriptor

• committee_size (int) – Number of members in the linear committee

• prior_weight (float) – Weight corresponding to the prior matrix in the linear system

• num_layers (int (default: -1)) – Number of layers in the MACE model to keep for descriptor evaluation Default of -1 uses all but the readout layer (equivalent to MACECalculator.get_descriptors() default)

• invariants_only (bool) – Whether to only keep the invariants partition of the descriptor vector, see MACECalculator.get_descriptors for more details

• batch_size (int) – Batch size to use for descriptor gradient evaluation. Lower batch size reduces overhead. If batch_size > len(atoms), then len(atoms) is used as the batch size instead

• **kwargs (Keyword Args) – Extra keywork arguments fed to TorchCommitteeCalculator

resample_committee(committee_size=None)

Resample the committee, based on the states of self.likelihood and self.sqrt_prior Populates self.committee_weights based on the newly sampled committee

Parameters:

committee_size (int, optional) – New size of the committee, if supplied. By default, a committee of size self.n_comm is drawn

select_structure(atoms)

sync()

get_descriptor_energy(atoms=None)

Get “descriptor energy”, which is the average of descriptor vectors in the structure

Returns an array of length self.n_desc

get_descriptor_forces(atoms=None)

Get “descriptor forces”, which are the derivatives w.r.t atomic positions of the total descriptor energy

Returns an array of shape (self.n_desc, Nats, 3)

get_descriptor_stress(atoms=None)

Get “descriptor stresses”, which are the stress analogues to the total descriptor energy

Returns an array of shape (self.n_desc, 3, 3)

get_committee_energies(atoms=None)

Get energy predictions by each member of the committee

Returns an array of length self.n_comm

get_committee_forces(atoms=None)

Get force predictions by each member of the committee

Returns an array of shape (self.n_comm, Nats, 3)

get_committee_stresses(atoms=None)

Get stress predictions by each member of the committee

Returns an array of shape (self.n_comm, 3, 3)

get_bias_energy(atoms=None)

get_bias_forces(atoms=None)

get_bias_stress(atoms=None)

Methods

__init__(mace_calculator, committee_size, ...)
band_structure()	Create band-structure object for plotting.
calculate(atoms, properties, system_changes)	Calculation for descriptor properties, committee properties, normal properties, and HAL properties
calculate_numerical_forces(atoms[, d])	Calculate numerical forces using finite difference.
calculate_numerical_stress(atoms[, d, voigt])	Calculate numerical stress using finite difference.
calculate_properties(atoms, properties)	This method is experimental; currently for internal use.
calculation_required(atoms, properties)
check_state(atoms[, tol])	Check for any system changes since last calculation.
export_properties()
get_atoms()
get_bias_energy([atoms])
get_bias_forces([atoms])
get_bias_stress([atoms])
get_charges([atoms])
get_committee_energies([atoms])	Get energy predictions by each member of the committee
get_committee_forces([atoms])	Get force predictions by each member of the committee
get_committee_stresses([atoms])	Get stress predictions by each member of the committee
get_default_parameters()
get_descriptor_energy([atoms])	Get "descriptor energy", which is the average of descriptor vectors in the structure
get_descriptor_forces([atoms])	Get "descriptor forces", which are the derivatives w.r.t atomic positions of the total descriptor energy
get_descriptor_stress([atoms])	Get "descriptor stresses", which are the stress analogues to the total descriptor energy
get_dipole_moment([atoms])
get_forces([atoms])
get_magnetic_moment([atoms])
get_magnetic_moments([atoms])	Calculate magnetic moments projected onto atoms.
get_potential_energies([atoms])
get_potential_energy([atoms, force_consistent])
get_property(name[, atoms, allow_calculation])	Overload of Calculator.get_property, converts from torch tensors to numpy arrays Allows for torch tensors to be stored in self.results between calls to self.calculate
get_stress([atoms])
get_stresses([atoms])	the calculator should return intensive stresses, i.e., such that stresses.sum(axis=0) == stress
read(label)	Read atoms, parameters and calculated properties from output file.
read_atoms(restart, **kwargs)
resample_committee([committee_size])	Resample the committee, based on the states of self.likelihood and self.sqrt_prior Populates self.committee_weights based on the newly sampled committee
reset()	Clear all information from old calculation.
select_structure(atoms)
set(**kwargs)	Set parameters like set(key1=value1, key2=value2, ...).
set_label(label)	Set label and convert label to directory and prefix.
sync()
todict([skip_default])	Obtain a dictionary of parameter information

Attributes

default_parameters	Default parameters
directory
discard_results_on_any_change	Whether we purge the results following any change in the set() method.
energy_weight
forces_weight
ignored_changes	Properties of Atoms which we ignore for the purposes of cache
implemented_properties	Properties calculator can handle (energy, forces, ...)
label
name
stress_weight
torch_device