New Mapping Type

[crdoconnor]

New Mapping Type

[crdoconnor]

No, that looks pretty much exactly how I'd do it.

[bt2901]

I'm interested in something similar: some sort of Map() | Map(). The use case:

We have a sequence of building blocks (taken together, they specify a training process for a particular topic model). There is a small list of building blocks available (at the moment, only two: CubeCreator and RegularizerModifierCube), but each takes a lot of input parameters. Different blocks require different parameters, however.

For now, I'm handling the situation like this:

base_schema = Map({
    'stages': Seq(Any()),
...
})
SUPPORTED_CUBES = [CubeCreator, RegularizersModifierCube]

def build_schema_for_cubes():
    """
    Returns
    -------
    dict
        each element is str -> strictyaml.Map
        where key is name of cube,
        value is a schema used for validation and type-coercion
    """
    schemas = {}
    for class_of_object in SUPPORTED_CUBES:
        res = build_schema_from_signature(class_of_object)
        res = Map(res)

        specific_schema = Map({class_of_object.__name__: res})
        schemas[class_of_object.__name__] = specific_schema
    return schemas

parsed = load(yaml_string, base_schema)
schemas = build_schema_for_cubes()

for i, stage in enumerate(parsed['stages']):
    assert len(stage) == 1
    name = list(stage.data)[0]

    if name not in schemas:
        raise ValueError(f"Unsupported stage ID: {name} at line {stage.start_line}")
    local_schema = schemas[name]

    stage.revalidate(local_schema)

Is there a better way?