Code Target API

Dense taxonomy distances and unary partners provide code-to-code targets and keep masks in codebook order. The anchor and its unary partner are removed from each target/candidate row.

The fixed inputs of Req 11's code–code and radial terms (spec 4.1(ii), (iii)), read once.

For every code, in codebook order: its level λ, the number of its digits; the tree metric D* to every code (Req 7); and its unary partner (Req 9), for a five-digit code and its only six-digit child each other. The code–code term's J_a is every code but the anchor and its partner, so a unary pair is neither a positive nor a negative.

The pair facts give D* and the unary flag. No exclusion column is read, so no exclusion pair can act as a code–code negative (Req 8(c)).

CodeTargets dataclass

Each code's level, D* to every code and unary partner, in codebook order.

Built by from_bundle. The arrays are fresh and writable, so torch.from_numpy can share them; nothing here writes to them.

Attributes:

Name Type Description
codes Tuple[str, ...]

The codes, in codebook order.
levels ndarray

λ, each code's number of digits, (N,) int64.
structural_distance ndarray

D*, (N, N) float32, symmetric and zero on the diagonal only.
unary_partner ndarray

Each code's unary partner's id, or -1 for none, (N,) int64.

__eq__(other)

Equal codes, and arrays of equal dtype and values.

from_bundle(bundle) classmethod

Read the codebook and the pair facts' PAIR_FACT_COLUMNS from a validated bundle.

The loader has checked that the code ids run from zero in codebook order, that the pair facts hold every pair of distinct codes once with D* as their distance, and that the unary flag marks exactly each five-digit code and its only six-digit child.

keep(ids)

J_a for each anchor: every code but the anchor itself and its unary partner.

Parameters:

Name Type Description Default
ids ndarray

The anchors' code ids, (A,), an integer array (a CPU tensor converts).

 required

Returns:

Type Description
ndarray

A fresh bool mask over the codes, (A, N).

Raises:

Type Description
ValueError

If ids is not a one-dimensional integer array of code ids.