FeaturizationConfig Class

Inheritance

builtins.object

FeaturizationConfig

Constructor

FeaturizationConfig(blocked_transformers: List[str] | None = None, column_purposes: Dict[str, str] | None = None, transformer_params: Dict[str, List[Tuple[List[str], Dict[str, Any]]]] | None = None, drop_columns: List[str] | None = None, dataset_language: str | None = None, prediction_transform_type: str | None = None)

Parameters

Remarks

Featurization customization has methods that allow you to:

• Add or remove column purpose. With the add_column_purpose and remove_column_purpose methods you can override the feature type for specified columns, for example, when the feature type of column does not correctly reflect its purpose. The add method supports adding all the feature types given in the FULL_SET attribute of the FeatureType class.

• Add or remove transformer parameters. With the add_transformer_params and remove_transformer_params methods you can change the parameters of customizable transformers like Imputer, HashOneHotEncoder, and TfIdf. Customizable transformers are listed in the SupportedTransformers class CUSTOMIZABLE_TRANSFORMERS attribute. Use the get_transformer_params to lookup customization parameters.

• Block transformers. Block transformers to be used for the featurization process with the add_blocked_transformers method. The transformers must be one of the transformers listed in the SupportedTransformers class BLOCKED_TRANSFORMERS attribute.

• Add a drop column to ignore for featurization and training with the add_drop_columns method. For example, you can drop a column that doesn't contain useful information.

• Add or remove prediction transform type. With add_prediction_transform_type and

remove_prediction_transform_type methods you can override the existing target column type. Prediction transform types are listed in the PredictionTransformTypes attribute.

The following code example shows how to customize featurization in automated ML for forecasting. In the example code, dropping a column and adding transform parameters are shown.

   featurization_config = FeaturizationConfig()
   # Force the CPWVOL5 feature to be numeric type.
   featurization_config.add_column_purpose("CPWVOL5", "Numeric")
   # Fill missing values in the target column, Quantity, with zeros.
   featurization_config.add_transformer_params(
       "Imputer", ["Quantity"], {"strategy": "constant", "fill_value": 0}
   )
   # Fill missing values in the INCOME column with median value.
   featurization_config.add_transformer_params(
       "Imputer", ["INCOME"], {"strategy": "median"}
   )
   # Fill missing values in the Price column with forward fill (last value carried forward).
   featurization_config.add_transformer_params("Imputer", ["Price"], {"strategy": "ffill"})

Full sample is available from https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/automated-machine-learning/forecasting-orange-juice-sales/auto-ml-forecasting-orange-juice-sales.ipynb

The next example shows customizing featurization in a regression problem using the Hardware Performance Dataset. In the example code, a blocked transformer is defined, column purposes are added, and transformer parameters are added.

   featurization_config = FeaturizationConfig()
   featurization_config.blocked_transformers = ["LabelEncoder"]
   # featurization_config.drop_columns = ['MMIN']
   featurization_config.add_column_purpose("MYCT", "Numeric")
   featurization_config.add_column_purpose("VendorName", "CategoricalHash")
   # default strategy mean, add transformer param for for 3 columns
   featurization_config.add_transformer_params("Imputer", ["CACH"], {"strategy": "median"})
   featurization_config.add_transformer_params(
       "Imputer", ["CHMIN"], {"strategy": "median"}
   )
   featurization_config.add_transformer_params(
       "Imputer", ["PRP"], {"strategy": "most_frequent"}
   )
   # featurization_config.add_transformer_params('HashOneHotEncoder', [], {"number_of_bits": 3})

Full sample is available from https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/automated-machine-learning/regression-explanation-featurization/auto-ml-regression-explanation-featurization.ipynb

The FeaturizationConfig defined in the code example above can then used in the configuration of an automated ML experiment as shown in the next code example.

   automl_settings = {
       "enable_early_stopping": True,
       "experiment_timeout_hours": 0.25,
       "max_concurrent_iterations": 4,
       "max_cores_per_iteration": -1,
       "n_cross_validations": 5,
       "primary_metric": "normalized_root_mean_squared_error",
       "verbosity": logging.INFO,
   }

   automl_config = AutoMLConfig(
       task="regression",
       debug_log="automl_errors.log",
       compute_target=compute_target,
       featurization=featurization_config,
       training_data=train_data,
       label_column_name=label,
       **automl_settings,
   )

Full sample is available from https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/automated-machine-learning/regression-explanation-featurization/auto-ml-regression-explanation-featurization.ipynb

Methods

add_blocked_transformers

add_blocked_transformers(transformers: str | List[str]) -> None

add_column_purpose

add_column_purpose(column_name: str, feature_type: str) -> None

add_drop_columns

add_drop_columns(drop_columns: str | List[str]) -> None

add_prediction_transform_type

add_prediction_transform_type(prediction_transform_type: str) -> None

add_transformer_params

add_transformer_params(transformer: str, cols: List[str], params: Dict[str, Any]) -> None

   featurization_config = FeaturizationConfig()
   # Force the CPWVOL5 feature to be numeric type.
   featurization_config.add_column_purpose("CPWVOL5", "Numeric")
   # Fill missing values in the target column, Quantity, with zeros.
   featurization_config.add_transformer_params(
       "Imputer", ["Quantity"], {"strategy": "constant", "fill_value": 0}
   )
   # Fill missing values in the INCOME column with median value.
   featurization_config.add_transformer_params(
       "Imputer", ["INCOME"], {"strategy": "median"}
   )
   # Fill missing values in the Price column with forward fill (last value carried forward).
   featurization_config.add_transformer_params("Imputer", ["Price"], {"strategy": "ffill"})

get_transformer_params

get_transformer_params(transformer: str, cols: List[str]) -> Dict[str, Any]

remove_column_purpose

remove_column_purpose(column_name: str) -> None

remove_prediction_transform_type

remove_prediction_transform_type() -> None

remove_transformer_params

remove_transformer_params(transformer: str, cols: List[str] | None = None) -> None

Attributes

blocked_transformers

column_purposes

dataset_language

drop_columns

prediction_transform_type

transformer_params