feature_types

Feature types module.

Classes:

Name	Description
NumericalFeature	Numerical feature.
CategoricalFeature	Categorical feature.
ImageFeature	Image feature.
MultivariateNumericalFeature	MultivariateNumericalFeature feature.
UnivariateTimeSeriesFeature	UnivariateTimeSerie feature.
MultivariateTimeSeriesFeature	MultivariateTimeSeries feature.
BoundingBoxesFeature	Feature that represents a list of bounding boxes, detection in a single image for instance.

Functions:

Name	Description
feature_type_from_model	Convert json features into the corresponding Feature.

NumericalFeature

Numerical feature.

It represents quantifiable values that can be measured and ordered. Its values may be continuous (e.g., real numbers) or discrete (e.g., integers). For instance, "age" or "price" could be set as numerical features.

Parameters:

Name	Type	Description	Default
name	Literal['NUMERICAL']		'NUMERICAL'

Methods:

Name	Description
to_model	Convert to NumericalFeatureType instance.

Attributes:

Name	Type	Description
name	Literal['NUMERICAL']

name: Literal['NUMERICAL'] = 'NUMERICAL'

to_model() -> NumericalFeatureType

Convert to NumericalFeatureType instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> NumericalFeatureType:
    """Convert to NumericalFeatureType instance."""
    return NumericalFeatureType(name=self.name)

CategoricalFeature

Categorical feature.

It represents data (bool, int or str only) that can be divided into distinct groups or categories.

It may be nominal or ordinal. For instance, "Education level" or "gender" could be set as categorical features.

Parameters:

Name	Type	Description	Default
categories	list[str] | list[int] | None	When building an AnalyzedSchema: If the preprocessor is one of label_encoder, onehot_encoder, then categories are automatically inferred from the preprocessor when fitted, and you should not provide any categories. Else, categories must be provided. When building a FittedSchema, categories must be provided. The categories' order must match the preprocessor's processing order. When using the dataset analyze and fit methods, no categories should be provided.	None
name	Literal['CATEGORICAL']		'CATEGORICAL'

Methods:

Name	Description
to_model	Convert to CategoricalFeatureType instance.
__attrs_post_init__	Return a fake category to be overwritten by the fitted preprocessor inferred categories.

Attributes:

Name	Type	Description
name	Literal['CATEGORICAL']
categories	list[str] | list[int] | None

name: Literal['CATEGORICAL'] = 'CATEGORICAL'

categories: list[str] | list[int] | None = None

to_model() -> CategoricalFeatureType

Convert to CategoricalFeatureType instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> CategoricalFeatureType:
    """Convert to CategoricalFeatureType instance."""
    return CategoricalFeatureType(
        name=self.name,
        categories=[] if self.categories is None or len(self.categories) == 0 else self.categories,
    )

__attrs_post_init__() -> None

Return a fake category to be overwritten by the fitted preprocessor inferred categories.

Source code in src/xpdeep/dataset/feature/feature_types.py

def __attrs_post_init__(self) -> None:
    """Return a fake category to be overwritten by the fitted preprocessor inferred categories."""
    if self.categories is None:
        self.categories = ["fake_categories"]

ImageFeature

Image feature.

It represents image feature objects. The corresponding data should use the channel-last format, i.e. batch_size x H x W x num_channels.

Parameters:

Name	Type	Description	Default
name	Literal['IMAGE']		'IMAGE'

Methods:

Name	Description
to_model	Convert to ImageFeatureType instance.

Attributes:

Name	Type	Description
name	Literal['IMAGE']

name: Literal['IMAGE'] = 'IMAGE'

to_model() -> ImageFeatureType

Convert to ImageFeatureType instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> ImageFeatureType:
    """Convert to ImageFeatureType instance."""
    return ImageFeatureType(name=self.name)

MultivariateNumericalFeature

MultivariateNumericalFeature feature.

It represents numerical data points divided into several channels. Oppositely as time serie features, there is no time relationship between points here.

Parameters:

Name	Type	Description	Default
channel_names	List[str]	List of channel names used for visualization.	required
name	Literal['MULTIVARIATE_NUMERICAL']		'MULTIVARIATE_NUMERICAL'

Methods:

Name	Description
to_model	Convert to VectorFeatureTypeInput instance.

Attributes:

Name	Type	Description
channel_names	list[str]
name	Literal['MULTIVARIATE_NUMERICAL']

channel_names: list[str]

name: Literal['MULTIVARIATE_NUMERICAL'] = 'MULTIVARIATE_NUMERICAL'

to_model() -> VectorFeatureTypeInput

Convert to VectorFeatureTypeInput instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> VectorFeatureTypeInput:
    """Convert to VectorFeatureTypeInput instance."""
    return VectorFeatureTypeInput(
        name="VECTOR", items=NumericalFeatureType(name="NUMERICAL"), channel_names=self.channel_names
    )

UnivariateTimeSeriesFeature

UnivariateTimeSerie feature.

It represents a time serie with a single channel, synchronized (no dynamic time warping required) or not.

The DTW will be applied automatically if this feature is required.

Parameters:

Name	Type	Description	Default
asynchronous	int	Whether the time serie is asynchronous (dynamic time warping will be automatically applied server side) or not.	False.
channel	tuple[str, str] | str | None	Used in XpViz. Indeed, if the same channel is used for both input (as a lookback, under a first feature object) and target (as a horizon to predict, under a second feature object), this parameter may be specified to visualize on the same curve both features, (the lookback and its corresponding horizon).	None
name	Literal['UNIVARIATE_TIMESERIES']		'UNIVARIATE_TIMESERIES'

Methods:

Name	Description
to_model	Convert to TimeseriesFeatureTypeInput instance.

Attributes:

Name	Type	Description
name	Literal['UNIVARIATE_TIMESERIES']
asynchronous	bool
channel	tuple[str, str] | str | None

name: Literal['UNIVARIATE_TIMESERIES'] = 'UNIVARIATE_TIMESERIES'

asynchronous: bool = field(default=False, kw_only=True)

channel: tuple[str, str] | str | None = field(default=None, kw_only=True)

to_model() -> TimeseriesFeatureTypeInput

Convert to TimeseriesFeatureTypeInput instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> TimeseriesFeatureTypeInput:
    """Convert to TimeseriesFeatureTypeInput instance."""
    return TimeseriesFeatureTypeInput(
        name="TIMESERIES",
        items=NumericalFeatureType(name="NUMERICAL"),
        asynchronous=self.asynchronous,
        channel=list(self.channel)
        if isinstance(self.channel, tuple)
        # TODO(Tanguy): replace with `else self.channel`
        # https://gitlab.xpdeep.com/xpdeep/xpdeep-client/-/issues/331
        else [self.channel, "0"]
        if self.channel is not None
        else None,
    )

MultivariateTimeSeriesFeature

MultivariateTimeSeries feature.

This class represents a multivariate time series (multiple channels), synchronized (no dynamic time warping required) or not.

The DTW will be applied automatically if this feature is required.

Parameters:

Name	Type	Description	Default
asynchronous	int	Whether the time serie is asynchronous (dynamic time warping will be automatically applied server side) or not.	False.
channel_names	List[str]	List of channel names used for visualization.	required
name	Literal['MULTIVARIATE_TIMESERIES']		'MULTIVARIATE_TIMESERIES'

Methods:

Name	Description
to_model	Convert to TimeseriesFeatureTypeInput instance.

Attributes:

Name	Type	Description
channel_names	list[str]
name	Literal['MULTIVARIATE_TIMESERIES']
asynchronous	bool

channel_names: list[str]

name: Literal['MULTIVARIATE_TIMESERIES'] = 'MULTIVARIATE_TIMESERIES'

asynchronous: bool = field(default=False, kw_only=True)

to_model() -> TimeseriesFeatureTypeInput

Convert to TimeseriesFeatureTypeInput instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> TimeseriesFeatureTypeInput:
    """Convert to TimeseriesFeatureTypeInput instance."""
    return TimeseriesFeatureTypeInput(
        name="TIMESERIES",
        items=VectorFeatureTypeInput(
            name="VECTOR", items=NumericalFeatureType(name="NUMERICAL"), channel_names=self.channel_names
        ),
        asynchronous=self.asynchronous,
    )

BoundingBoxesFeature

Feature that represents a list of bounding boxes, detection in a single image for instance.

Parameters:

Name	Type	Description	Default
categories	list[int] | list[str]	The categories that the bounding boxes represent.	required
name	Literal['BBOX_LIST']	The feature name.	'BBOX_LIST'
channel_names	tuple[str, ...]	The expected channel names for a single bounding box expected format is its class, its position, and the detection score.	('class', 'center_x', 'center_y', 'width', 'height', 'score')

Methods:

Name	Description
to_model	Convert to BBoxListFeatureType instance.

Attributes:

Name	Type	Description
categories	list[int] | list[str]
name	Literal['BBOX_LIST']
channel_names	tuple[str, ...]

categories: list[int] | list[str]

name: Literal['BBOX_LIST'] = 'BBOX_LIST'

channel_names: tuple[str, ...] = ('class', 'center_x', 'center_y', 'width', 'height', 'score')

to_model() -> ListFeatureTypeInput

Convert to BBoxListFeatureType instance.

Source code in src/xpdeep/dataset/feature/feature_types.py

def to_model(self) -> ListFeatureTypeInput:
    """Convert to BBoxListFeatureType instance."""
    return ListFeatureTypeInput(
        name="LIST",
        items=BBoxFeatureType(
            name="BBOX", size=[0, 0], categories=self.categories, channel_names=list(self.channel_names)
        ),
    )

feature_type_from_model(json_response: dict[str, object]) -> NumericalFeature | CategoricalFeature | MultivariateNumericalFeature | UnivariateTimeSeriesFeature | MultivariateTimeSeriesFeature | ImageFeature | BoundingBoxesFeature

Convert json features into the corresponding Feature.

Source code in src/xpdeep/dataset/feature/feature_types.py

def feature_type_from_model(  # noqa: PLR0911
    json_response: dict[str, object],
) -> (
    NumericalFeature
    | CategoricalFeature
    | MultivariateNumericalFeature
    | UnivariateTimeSeriesFeature
    | MultivariateTimeSeriesFeature
    | ImageFeature
    | BoundingBoxesFeature
):
    """Convert json features into the corresponding Feature."""
    match json_response["name"]:
        case "NUMERICAL":
            return NumericalFeature()
        case "CATEGORICAL":
            return CategoricalFeature(categories=json_response["categories"])  # type: ignore[arg-type]
        case "VECTOR":
            return MultivariateNumericalFeature(channel_names=json_response["channel_names"])  # type: ignore[arg-type]
        case "TIMESERIES":
            match json_response["items"]["name"]:  # type: ignore[index]
                case "NUMERICAL":
                    return UnivariateTimeSeriesFeature(
                        asynchronous=json_response["asynchronous"],  # type: ignore[arg-type]
                        channel=json_response["channel"],  # type: ignore[arg-type]
                    )
                case "VECTOR":
                    return MultivariateTimeSeriesFeature(
                        asynchronous=json_response["asynchronous"],  # type: ignore[arg-type]
                        channel_names=json_response["items"]["channel_names"],  # type: ignore[index]
                    )
                case _:
                    msg = f"Feature type `Timeseries[{json_response['items']['expose_data_type']}]` not recognized"  # type: ignore[index]
                    raise ApiError(msg)
        case "IMAGE":
            return ImageFeature()
        case "LIST":
            return BoundingBoxesFeature(
                categories=json_response["items"]["categories"],  # type: ignore[index]
                channel_names=json_response["items"]["channel_names"],  # type: ignore[index]
            )
        case _:
            msg = f"Feature type `{json_response['expose_data_type']}` not recognized"
            raise ApiError(msg)