nvalchemi.training.distillation.evaluation.ExtensivityMetrics#

class nvalchemi.training.distillation.evaluation.ExtensivityMetrics(*, repeats, num_graphs, max_error_per_atom, mean_error_per_atom, max_relative_error)[source]#

Energy-scaling error of a model across replicated cells.

Extensivity is the property that a structure’s energy grows in proportion to its size: a k-fold supercell has k times the energy of the cell it replicates. This record measures how far a model departs from it.

Parameters:
  • repeats (tuple[int, int, int])

  • num_graphs (int)

  • max_error_per_atom (float)

  • mean_error_per_atom (float)

  • max_relative_error (float)

repeats#

Replication factors applied along each lattice vector.

Type:

tuple[int, int, int]

num_graphs#

Number of structures checked.

Type:

int

max_error_per_atom, mean_error_per_atom

Largest and mean absolute deviation of the supercell energy from the copy count times the input cell’s energy, divided by the supercell’s atom count.

Type:

float

max_relative_error#

Largest ratio of the same deviation to the magnitude of the expected supercell energy.

Type:

float

model_config = {'extra': 'forbid', 'frozen': True}#

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].