Source code for autogluon.features.generators.dummy

import logging

from pandas import DataFrame

from autogluon.common.features.feature_metadata import FeatureMetadata

from .abstract import AbstractFeatureGenerator

logger = logging.getLogger(__name__)




class DummyFeatureGenerator(AbstractFeatureGenerator):
    """
    Ignores all input features and returns a single int feature with all 0 values.
    Useful for testing purposes or to avoid crashes if no features were given.
    """

    def __init__(self, features_in="empty", feature_metadata_in="empty", **kwargs):
        if features_in == "empty":
            features_in = []
        if feature_metadata_in == "empty":
            feature_metadata_in = FeatureMetadata(type_map_raw={})
        super().__init__(features_in=features_in, feature_metadata_in=feature_metadata_in, **kwargs)

    def _fit_transform(self, X: DataFrame, **kwargs) -> (DataFrame, dict):
        X_out = self._transform(X)
        return X_out, dict()

    def _transform(self, X: DataFrame) -> DataFrame:
        return self._generate_features_dummy(X)

    @staticmethod
    def get_default_infer_features_in_args() -> dict:
        return dict(valid_raw_types=[])

    @staticmethod
    def _generate_features_dummy(X: DataFrame):
        X_out = DataFrame(index=X.index)
        X_out["__dummy__"] = 0
        return X_out

    def is_valid_metadata_in(self, feature_metadata_in: FeatureMetadata):
        return True