merlin.datasets.metadata module

class merlin.datasets.metadata.Normalization(method, range, per_feature=True)

Bases: object

Dataset-level normalization metadata.

Parameters:

method (str) – Name of the normalization method.
range (tuple) – Target value range after normalization.
per_feature (bool) – Whether normalization is applied independently per feature.

method: str

per_feature: bool = True

range: tuple

to_text()

Render normalization metadata as human-readable text.

Returns:: Human-readable summary of the normalization settings.
Return type:: str

class merlin.datasets.metadata.FeatureNormalization(original_unit=None, scale_factor=None, offset=None)

Bases: object

Feature-level normalization metadata.

Parameters:

original_unit (str | None) – Original measurement unit before normalization.
scale_factor (float | None) – Scaling factor applied during normalization.
offset (float | None) – Offset applied during normalization.

offset: Optional[float] = None

original_unit: Optional[str] = None

scale_factor: Optional[float] = None

to_text()

Render feature normalization metadata as text.

Returns:: Human-readable normalization summary.
Return type:: str

class merlin.datasets.metadata.Feature(name, description, type, value_range=None, unit=None, stats=None, normalization=None)

Bases: object

Description of a dataset feature.

Parameters:

name (str) – Feature name.
description (str) – Human-readable feature description.
type (str) – Feature dtype or semantic type.
value_range (tuple | None) – Expected value range.
unit (str | None) – Measurement unit.
stats (dict[str, float] | None) – Optional feature statistics.
normalization (FeatureNormalization | None) – Optional feature-level normalization metadata.

description: str

name: str

normalization: Optional[FeatureNormalization] = None

stats: Optional[dict[str, float]] = None

to_text()

Render feature metadata as human-readable text.

Returns:: Human-readable feature description.
Return type:: str

type: str

unit: Optional[str] = None

value_range: Optional[tuple] = None

class merlin.datasets.metadata.DatasetMetadata(name, description, features, num_instances, subset=None, num_features=None, normalization=None, task_type=<factory>, num_classes=None, characteristics=<factory>, homepage=None, license=None, citation=None, creators=<factory>, year=None, feature_relationships=None)

Bases: object

Structured metadata describing a dataset.

Parameters:

name (str) – Dataset name.
description (str) – Dataset description.
features (list[Feature]) – Descriptions of dataset features.
num_instances (int) – Number of instances in the dataset or subset.
subset (str) – Dataset split name.
num_features (int | None) – Number of input features.
normalization (Normalization | None) – Dataset-level normalization metadata.
task_type (list[str] | None) – Supported task types.
num_classes (int | None) – Number of target classes, when relevant.
characteristics (list[str]) – High-level dataset characteristics.
homepage (str | None) – Dataset homepage.
license (str | None) – Dataset license.
citation (str | None) – Citation text.
creators (list[str]) – Dataset creators.
year (int | None) – Creation or publication year.
feature_relationships (str | None) – Optional description of relationships between features.