noether.core.schemas.lib¶

Attributes¶

Annotated-metadata marker for polymorphic registry config fields.

`resolve_config_class`(kind, base_cls)	Resolve a config class from a dotted kind string.
`ConfiguredBy`(config_class)	Decorator to mark a class as being configured by a specific config class.

noether.core.schemas.lib.resolve_config_class[T](kind, base_cls)¶

Resolve a config class from a dotted kind string.

Resolution order:

If the imported class is already a base_cls subclass, return it directly.
Check for a _config_class attribute (set by @ConfiguredBy).
Inspect __init__ type hints for the first parameter that is a base_cls subclass.

Parameters:

kind (str) – fully qualified dotted path (e.g., "noether.training.trainers.WeightedLossTrainer").
base_cls (type[T]) – the base config class to resolve against.

Raises:

ValueError – if the config class cannot be determined.

Return type:

type[T]

class noether.core.schemas.lib.Discriminated(registry_cls)¶

Annotated-metadata marker for polymorphic registry config fields.

Usage: field: Annotated[Any, Discriminated(MyComponent)]

Contributes two behaviors to the annotated field:

Validation — dicts are instantiated to the concrete config class resolved from their registry key (kind/type field), like a BeforeValidator around _discriminated_validator().
Serialization — the value is serialized by its runtime class (equivalent to pydantic.SerializeAsAny), so subclass-only fields survive model_dump even when the field is annotated with the base class. Unlike a global model_dump(serialize_as_any=True), this duck-types only the model boundary and still respects field-level serializers nested inside the concrete config (e.g. TorchTensor).

noether.core.schemas.lib.ConfiguredBy(config_class)¶

Decorator to mark a class as being configured by a specific config class. Usage:

@ConfiguredBy(MyConfig) class MyClass:

…