nvalchemi.training.distillation.evaluation.RDFComparison#

class nvalchemi.training.distillation.evaluation.RDFComparison(*, jensen_shannon, l1, max_deviation, num_bins, pair=None)[source]#

Scalar divergences between two radial distribution functions.

The comparison has the species resolution of its curves. Two total g(r) curves are compared without regard to species, so a student that swapped two sublattices can score well while its partial curves are wrong. pair records which resolution was measured.

Parameters:
  • jensen_shannon (float)

  • l1 (float)

  • max_deviation (float)

  • num_bins (int)

  • pair (tuple[int, int] | None)

jensen_shannon#

Base-2 Jensen-Shannon divergence between the two normalized pair-distance histograms, in [0, 1].

Type:

float

l1#

Integrated absolute difference of the two g(r) curves, in A.

Type:

float

max_deviation#

Largest absolute difference between the two curves in any bin.

Type:

float

num_bins#

Bins the comparison ran over.

Type:

int

pair#

Atomic numbers both curves were resolved to, or None for species-blind totals.

Type:

tuple[int, int] | None

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

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