Plugin interfaces
kd.search holds the machinery behind the algorithm plugins; this page lists
the part a plugin author implements. SurrogateTrainer is the protocol an
algorithm whose descriptor names config_artifact_keys must satisfy, a
train_surrogate method and an artifacts mapping; a registry-level test
enforces the pairing for every registered algorithm.
The name is imported from the subpackage rather than the package root:
from kd.search import SurrogateTrainer.
Bases: Protocol
Plugin face for training the derivative surrogate as a standalone action.
property
artifacts: dict[str, dict[str, str | int]] | None
Identity of the injected surrogate keyed by its config key; None
when nothing was injected.
train_surrogate(
dataset: PDEDataset,
) -> tuple[Module, TrainingResult]
Train and return (module, training_result) without installing it.