Feature Definitions

FeatureDefinitions(perfdb)

Class used for handling feature definitions. Can be accessed via perfdb.features.definitions.

Parameters:

• perfdb

  (PerfDB) –

  Top level object carrying all functionality and the connection handler.

Source code in echo_postgres/feature_definitions.py

def __init__(self, perfdb: e_pg.PerfDB) -> None:
    """Class used for handling feature definitions. Can be accessed via `perfdb.features.definitions`.

    Parameters
    ----------
    perfdb : PerfDB
        Top level object carrying all functionality and the connection handler.
    """
    super().__init__(perfdb)

    from .feature_definition_attributes import FeatureDefinitionAttributes

    # * subclasses

    self.attributes = FeatureDefinitionAttributes(perfdb)

get(object_names=None, object_models=None, feature_names=None, data_source_types=None, attributes=None, filter_type='and', regex_feature_names=False, get_attributes=False, attribute_names=None, output_type='DataFrame')

Gets all feature definitions.

The most useful keys/columns returned are:

• display_name
• description
• object_model_name
• data_source_type_name
• name_in_data_source
• id_in_data_source

Parameters:

• object_names

  (list[str] | None, default: None ) –

  List of object names to filter the results. If set, will add the models of these objects to the object_models filter. By default None

• object_models

  (list[str] | None, default: None ) –

  List of object model names to filter the results. By default None

• feature_names

  (list[str] | None, default: None ) –

  List of feature names to filter the results. By default None

• data_source_types

  (list[str] | None, default: None ) –

  List of data source type names to filter the results. By default None

• attributes

  (dict[str, Any] | None, default: None ) –

  Dictionary with the attributes to filter the results. It must be in the format {attribute_name: attribute_value, ...} By default None

• filter_type

  (Literal['and', 'or'], default: 'and' ) –

  How to treat multiple filters. Can be one of ["and", "or"]. By default "and"

• regex_feature_names

  (bool, default: False ) –

  If True, will treat the feature_names as regex patterns.

  For example, if feature_names=['^active.', '^reactive.'], it will return all features that start with 'active' or 'reactive'.

  By default False

• get_attributes

  (bool, default: False ) –

  If True, will also get the attributes of the features. It's highly recommended to specify the attribute names to get to speed up the query.

  By default False

• attribute_names

  (list[str] | None, default: None ) –

  List of attribute names to get if get_attributes is True. In case get_attributes is False, this parameter is ignored. If None, all attributes are returned.

  By default None

• output_type

  (Literal['dict', 'DataFrame'], default: 'DataFrame' ) –

  Output type of the data. Can be one of ["dict", "DataFrame"] By default "DataFrame"

Returns:

• DataFrame –

  If output_type is "DataFrame", returns a DataFrame with the following format: index=MultiIndex[object_model_name, name], columns=[attribute, ...]

• dict[str, dict[str, dict[str, Any]]] –

  If output_type is "dict", returns a dictionary with the following format: {object_model_name: {feature_name: {attribute: value, ...}, ...}, ...}

Source code in echo_postgres/feature_definitions.py

@validate_call
def get(
    self,
    object_names: list[str] | None = None,
    object_models: list[str] | None = None,
    feature_names: list[str] | None = None,
    data_source_types: list[str] | None = None,
    attributes: dict[str, Any] | None = None,
    filter_type: Literal["and", "or"] = "and",
    regex_feature_names: bool = False,
    get_attributes: bool = False,
    attribute_names: list[str] | None = None,
    output_type: Literal["dict", "DataFrame"] = "DataFrame",
) -> DataFrame | dict[str, dict[str, dict[str, Any]]]:
    """Gets all feature definitions.

    The most useful keys/columns returned are:

    - display_name
    - description
    - object_model_name
    - data_source_type_name
    - name_in_data_source
    - id_in_data_source

    Parameters
    ----------
    object_names : list[str] | None, optional
        List of object names to filter the results. If set, will add the models of these objects to the object_models filter.
        By default None
    object_models : list[str] | None, optional
        List of object model names to filter the results.
        By default None
    feature_names : list[str] | None, optional
        List of feature names to filter the results.
        By default None
    data_source_types : list[str] | None, optional
        List of data source type names to filter the results.
        By default None
    attributes : dict[str, Any] | None, optional
        Dictionary with the attributes to filter the results. It must be in the format {attribute_name: attribute_value, ...}
        By default None
    filter_type : Literal["and", "or"], optional
        How to treat multiple filters. Can be one of ["and", "or"].
        By default "and"
    regex_feature_names : bool, optional
        If True, will treat the feature_names as regex patterns.

        For example, if feature_names=['^active.*', '^reactive.*'], it will return all features that start with 'active' or 'reactive'.

        By default False
    get_attributes : bool, optional
        If True, will also get the attributes of the features. It's highly recommended to specify the attribute names to get
        to speed up the query.

        By default False
    attribute_names : list[str] | None, optional
        List of attribute names to get if get_attributes is True. In case get_attributes is False, this parameter is ignored. If None, all attributes are returned.

        By default None
    output_type : Literal["dict", "DataFrame"], optional
        Output type of the data. Can be one of ["dict", "DataFrame"]
        By default "DataFrame"

    Returns
    -------
    DataFrame
        If output_type is "DataFrame", returns a DataFrame with the following format: index=MultiIndex[object_model_name, name], columns=[attribute, ...]
    dict[str, dict[str, dict[str, Any]]]
        If output_type is "dict", returns a dictionary with the following format: {object_model_name: {feature_name: {attribute: value, ...}, ...}, ...}
    """
    # checking inputs
    where = self._check_get_args(
        object_names=object_names,
        object_models=object_models,
        feature_names=feature_names,
        data_source_types=data_source_types,
        attributes=attributes,
        filter_type=filter_type,
        regex_feature_names=regex_feature_names,
    )

    # getting the feature definitions
    query = [
        sql.SQL("SELECT * FROM performance.v_features "),
    ]
    if where:
        query.append(where)
    query.append(sql.SQL(" ORDER BY object_model_name, name"))

    query = sql.Composed(query)

    with self._perfdb.conn.reconnect() as conn:
        df = conn.read_to_pandas(query)
    df = df.set_index(["object_model_name", "name"])

    # getting attributes
    if get_attributes:
        # names of the models
        got_models = df.index.get_level_values("object_model_name").unique().tolist()
        for model in got_models:
            # features of the model
            got_model_features = df.loc[IndexSlice[model, :], :].index.get_level_values(1).tolist()
            # getting the attributes
            attrs: DataFrame = self._perfdb.features.definitions.attributes.get(
                object_models=[model],
                feature_names=got_model_features,
                attribute_names=attribute_names,
                output_type="DataFrame",
                values_only=True,
            )
            # renaming indexes
            attrs.index = attrs.index.rename(["object_model_name", "name", "attribute_name"])
            # pivot the attributes
            attrs = attrs.reset_index(drop=False).pivot(
                index=["object_model_name", "name"],
                columns="attribute_name",
                values="attribute_value",
            )

            # adding attributes not in columns via merge
            new_cols = attrs.columns.difference(df.columns)
            df = df.merge(attrs[new_cols], left_index=True, right_index=True, how="left")

            # updating all columns
            df.update(attrs)

    if output_type == "dict":
        # getting rid of unnecessary columns
        df = df.drop(columns=["object_model_id"])
        # converting to dictionary
        result = df.to_dict(orient="index")
        # converting from format {(obj_model, feature_name): {attribute: value, ...}, ...} to {obj_model: {feature_name: {attribute: value, ...}, ...}, ...}
        final_result = {}
        for (obj_model, feature_name), feature_data in result.items():
            final_result.setdefault(obj_model, {})[feature_name] = feature_data
        return final_result

    return df

get_details(object_name)

Gets where (which table) each feature of an object is located in the database.

The most useful keys/columns returned are: - id - name - table_name - column_name

Parameters:

• object_name

  (str) –

  Name of the desired object.

Returns:

• DataFrame –

  DataFrame where index is the name of the features.

Source code in echo_postgres/feature_definitions.py

@validate_call
def get_details(self, object_name: str) -> DataFrame:
    """Gets where (which table) each feature of an object is located in the database.

    The most useful keys/columns returned are:
    - id
    - name
    - table_name
    - column_name

    Parameters
    ----------
    object_name : str
        Name of the desired object.

    Returns
    -------
    DataFrame
        DataFrame where index is the name of the features.
    """
    query = sql.SQL("SELECT * FROM performance.fn_get_object_features_and_tables({object_name});").format(
        object_name=sql.Literal(object_name),
    )
    with self._perfdb.conn.reconnect() as conn:
        df = conn.read_to_pandas(query)

    df = df.set_index("name")

    return df

get_ids(object_names=None, object_models=None, feature_names=None, data_source_types=None, attributes=None, filter_type='and', regex_feature_names=False)

Gets all feature definitions and their respective ids.

Parameters:

• object_names

  (list[str] | None, default: None ) –

  List of object names to filter the results. If set, will add the models of these objects to the object_models filter. By default None

• object_models

  (list[str] | None, default: None ) –

  List of object model names to filter the results. By default None

• feature_names

  (list[str] | None, default: None ) –

  List of feature names to filter the results. By default None

• data_source_types

  (list[str] | None, default: None ) –

  List of data source type names to filter the results. By default None

• attributes

  (dict[str, Any] | None, default: None ) –

  Dictionary with the attributes to filter the results. It must be in the format {attribute_name: attribute_value, ...} By default None

• filter_type

  (Literal['and', 'or'], default: 'and' ) –

  How to treat multiple filters. Can be one of ["and", "or"]. By default "and"

• regex_feature_names

  (bool, default: False ) –

  If True, will treat the feature_names as regex patterns.

  For example, if feature_names=['^active.', '^reactive.'], it will return all features that start with 'active' or 'reactive'.

  By default False

Returns:

• dict[str, dict[str, int]] –

  Dictionary with the following format: {object_model_name: {feature_name: feature_id, ...}, ...}

Source code in echo_postgres/feature_definitions.py

@validate_call
def get_ids(
    self,
    object_names: list[str] | None = None,
    object_models: list[str] | None = None,
    feature_names: list[str] | None = None,
    data_source_types: list[str] | None = None,
    attributes: dict[str, Any] | None = None,
    filter_type: Literal["and", "or"] = "and",
    regex_feature_names: bool = False,
) -> dict[str, dict[str, int]]:
    """Gets all feature definitions and their respective ids.

    Parameters
    ----------
    object_names : list[str] | None, optional
        List of object names to filter the results. If set, will add the models of these objects to the object_models filter.
        By default None
    object_models : list[str] | None, optional
        List of object model names to filter the results.
        By default None
    feature_names : list[str] | None, optional
        List of feature names to filter the results.
        By default None
    data_source_types : list[str] | None, optional
        List of data source type names to filter the results.
        By default None
    attributes : dict[str, Any] | None, optional
        Dictionary with the attributes to filter the results. It must be in the format {attribute_name: attribute_value, ...}
        By default None
    filter_type : Literal["and", "or"], optional
        How to treat multiple filters. Can be one of ["and", "or"].
        By default "and"
    regex_feature_names : bool, optional
        If True, will treat the feature_names as regex patterns.

        For example, if feature_names=['^active.*', '^reactive.*'], it will return all features that start with 'active' or 'reactive'.

        By default False

    Returns
    -------
    dict[str, dict[str, int]]
        Dictionary with the following format:
        {object_model_name: {feature_name: feature_id, ...}, ...}
    """
    # checking inputs
    where = self._check_get_args(
        object_names=object_names,
        object_models=object_models,
        feature_names=feature_names,
        data_source_types=data_source_types,
        attributes=attributes,
        filter_type=filter_type,
        regex_feature_names=regex_feature_names,
    )

    # getting the feature definitions
    query = [
        sql.SQL("SELECT object_model_name, name, id FROM performance.v_features "),
    ]
    if where:
        query.append(where)
    query.append(sql.SQL(" ORDER BY object_model_name, name"))

    query = sql.Composed(query)

    with self._perfdb.conn.reconnect() as conn:
        df = conn.read_to_pandas(query)
    df = df.set_index(["object_model_name", "name"])

    # converting to dictionary
    result = df["id"].to_dict()
    # converting from format {(obj_model, feature_name): feature_id, ...} to {obj_model: {feature_name: feature_id, ...}, ...}
    final_result = {}
    for (obj_model, feature_name), feature_id in result.items():
        final_result.setdefault(obj_model, {})[feature_name] = feature_id

    return final_result