tfdf.builder.AbstractBuilder

Generic model builder.

tfdf.builder.AbstractBuilder(
    path: str,
    objective: tfdf.py_tree.objective.AbstractObjective,
    model_format: Optional[tfdf.builder.ModelFormat],
    import_dataspec: Optional[data_spec_pb2.DataSpecification],
    input_model_signature_fn: Optional[tf_core.InputModelSignatureFn],
    file_prefix: Optional[str] = None,
    verbose: int = 1,
    advanced_arguments: Optional[tfdf.builder.AdvancedArguments] = None,
    keras_model_name: Optional[str] = None
)

dataspec Dataspec, possibly partially constructed.

Can be called before close for advanced model edition.

objective Objective of the model.

Methods

close

View source

close()

Finalize the builder work.

This method should be called last.

get_dictionary

View source

get_dictionary(
    col_name: str
) -> List[str]

Gets the dictionary of a categorical(-set) string feature.

model_type

View source

@abc.abstractmethod
model_type() -> str

Unique key describing the type of the model.

observe_feature

View source

observe_feature(
    feature: tfdf.inspector.SimpleColumnSpec,
    categorical_values: Optional[Union[List[str], List[int]]] = None
)

Register a feature and some of its possible value.

Generally, users don't need to call this function. An example of advanced exception is if a model does not refer to a specific possible categorical value, and if this value should be treated differently than out-of-vocabulary values.

Should be called at least once on each of the model input features. If called multiple times with categorical_values the set of possible values will be the union of the categorical_valuess.

Args
feature Definition of the feature.
categorical_values Set of observed values. Only for categorical-like features.

set_dictionary

View source

set_dictionary(
    col_name: str, dictionary: List[str]
) -> None

Sets the dictionary of a categorical or categorical-set column.

yggdrasil_model_path

View source

yggdrasil_model_path()

Gets the path to the destination yggdrasil model.