declearn.model.api.register_vector_type

Decorate a Vector subclass to make it buildable with Vector.build.

Decorating a Vector subclass with this has three effects:

• Add the class to registered type (in the "Vector" group). See declearn.utils.register_type for details.
• Make instances of that class JSON-serializable, embarking the wrapped data by using the pack and unpack methods of the class. See declearn.utils.add_json_support.
• Make the subclass buildable through Vector.build(coefs), based on the analysis of wrapped coefficients' type.

Parameters:

Name	Type	Description	Default
v_type	Type[Any]	Type of wrapped data that is to trigger using cls.	required
*types	Type[Any]	Additional v_type alternatives for wrapped data.	()
name	Optional[str]	Optional name under which to register the type, shared by register_type and add_json_support. If None, use cls.__name__.	None

Returns:

Name	Type	Description
register	func(cls) -	A closure that performs the registration operations. Hence register_vector_type is designed to be used as a class decorator.

Source code in declearn/model/api/_vector.py

def register_vector_type(
    v_type: Type[Any],
    *types: Type[Any],
    name: Optional[str] = None,
) -> Callable[[Type[Vector]], Type[Vector]]:
    """Decorate a Vector subclass to make it buildable with `Vector.build`.

    Decorating a Vector subclass with this has three effects:

    * Add the class to registered type (in the "Vector" group).
      See `declearn.utils.register_type` for details.
    * Make instances of that class JSON-serializable, embarking
      the wrapped data by using the `pack` and `unpack` methods
      of the class. See `declearn.utils.add_json_support`.
    * Make the subclass buildable through `Vector.build(coefs)`,
      based on the analysis of wrapped coefficients' type.

    Parameters
    ----------
    v_type: type
        Type of wrapped data that is to trigger using `cls`.
    *types: type
        Additional `v_type` alternatives for wrapped data.
    name: str or None, default=None
        Optional name under which to register the type, shared
        by `register_type` and `add_json_support`.
        If None, use `cls.__name__`.

    Returns
    -------
    register: func(cls) -> cls
        A closure that performs the registration operations.
        Hence `register_vector_type` is designed to be used
        as a class decorator.
    """
    v_types = (v_type, *types)

    # Set up a registration function.
    def register(cls: Type[Vector]) -> Type[Vector]:
        nonlocal name, v_types
        if name is None:
            name = cls.__name__
        # Register the Vector type. Note: this type-checks cls.
        register_type(cls, name=name, group="Vector")
        # Add support for JSON (de)serialization, relying on (un)pack.
        add_json_support(cls, cls.pack, cls.unpack, name=name)
        # Make the subclass buildable through `Vector.build(coefs)`.
        for v_type in v_types:
            VECTOR_TYPES[v_type] = cls
        return cls

    # Return the former, enabling decoration syntax for `register_vector_type`.
    return register