Policies and assessment¶

Goal. Define explicit substitution and linear-transfer policies, inspect their provenance, and summarize their behavior over several elements.

Prerequisites. Familiarity with direct get_* and provenance-carrying lookup_* calls from the scalar-data quickstart is helpful. All values in this notebook are scalar radii in angstrom.

Setup¶

Import the public namespace used for policy, dataset, transfer, and assessment objects.

In [1]:

import atomref as ar

import atomref as ar

Ordered substitution¶

This policy prefers Cordero covalent radii and explicitly substitutes the legacy CSD covalent table when the preferred table lacks an element. The result reports that substitution and the dataset that supplied it.

In [2]:

covalent_policy = ar.RadiiPolicy(
    kind='covalent',
    base_set='cordero2008',
    transfers=(
        ar.SubstitutionTransfer(
            source=ar.DatasetRef('covalent_radius', 'csd_legacy_cov')
        ),
    ),
)
lookup = ar.lookup_covalent_radius('Bk', policy=covalent_policy)
print(lookup.source)
print(f"{lookup.value:.12f}")
print(lookup.resolved_from)

covalent_policy = ar.RadiiPolicy( kind='covalent', base_set='cordero2008', transfers=( ar.SubstitutionTransfer( source=ar.DatasetRef('covalent_radius', 'csd_legacy_cov') ), ), ) lookup = ar.lookup_covalent_radius('Bk', policy=covalent_policy) print(lookup.source) print(f"{lookup.value:.12f}") print(lookup.resolved_from)

transfer_substitution
1.540000000000
(DatasetRef(quantity='covalent_radius', set_id='csd_legacy_cov'),)

Linear transfer from support data¶

For a missing van der Waals value, a fitted relation uses the packaged Rahm atomic-radius support set. The fit record exposes its coefficients and training-point count rather than hiding the inference.

In [3]:

vdw_policy = ar.RadiiPolicy(
    kind='van_der_waals',
    base_set='alvarez2013',
    transfers=(
        ar.LinearTransfer(
            predictors=(ar.DatasetRef('atomic_radius', 'rahm2016'),)
        ),
    ),
)
lookup = ar.lookup_vdw_radius('Pm', policy=vdw_policy)
print(f"{lookup.value:.12f}")
print(lookup.source)
print(
    f"slope={lookup.fit.coefficients[0]:.12f} intercept={lookup.fit.intercept:.12f} n={lookup.fit.n_points}"
)

vdw_policy = ar.RadiiPolicy( kind='van_der_waals', base_set='alvarez2013', transfers=( ar.LinearTransfer( predictors=(ar.DatasetRef('atomic_radius', 'rahm2016'),) ), ), ) lookup = ar.lookup_vdw_radius('Pm', policy=vdw_policy) print(f"{lookup.value:.12f}") print(lookup.source) print( f"slope={lookup.fit.coefficients[0]:.12f} intercept={lookup.fit.intercept:.12f} n={lookup.fit.n_points}" )

2.897226539515
transfer_linear
slope=1.135336645553 intercept=-0.315776167399 n=90

Assess behavior over a selection¶

Assessment counts direct, transferred, and missing outcomes and can retain a per-element result for auditing the policy before it enters an algorithm.

In [4]:

assessment = ar.assess_radii_policy(
    ['C', 'Xe', 'Pm', 'Bk'],
    policy=vdw_policy,
    detail=True,
)
print(assessment.n_base, assessment.n_transfer_linear, assessment.n_missing)
for row in assessment.per_element:
    value = 'None' if row.lookup.value is None else f"{row.lookup.value:.12f}"
    print(row.symbol, row.lookup.source, value)

assessment = ar.assess_radii_policy( ['C', 'Xe', 'Pm', 'Bk'], policy=vdw_policy, detail=True, ) print(assessment.n_base, assessment.n_transfer_linear, assessment.n_missing) for row in assessment.per_element: value = 'None' if row.lookup.value is None else f"{row.lookup.value:.12f}" print(row.symbol, row.lookup.source, value)

3 1 0
C base 1.770000000000
Xe base 2.060000000000
Pm transfer_linear 2.897226539515
Bk base 3.400000000000

What this demonstrated and limitations¶

Policies make fallback order and inferred values explicit, while assessment shows how that choice behaves over a requested element set. Transfer quality still depends on scientifically appropriate target and support datasets. These scalar policies do not apply to proatomic radial profiles.