Skip to content

Dataset Spec

DatasetSpec is the canonical research-assembly contract.

Notable composition surfaces:

  • FeatureRequest
  • FeatureRequestGroup
  • JoinPolicy
  • MissingnessPolicy

Built-in notebook-ready template families live alongside the spec. See Market Templates for the canonical lag-return and rolling-volatility helpers.

alphaforge.features.dataset_spec

DatasetArtifact dataclass

Returned by build_dataset().

DatasetSpec dataclass

feature_requests() -> list[FeatureRequest]

Return the flattened feature-request list used by the builder.

FeatureRequest dataclass

A request to materialize a FeatureTemplate with params, optionally with slice overrides.

FeatureRequestGroup dataclass

Composable group of feature requests with inherited metadata.

JoinPolicy dataclass

How to join feature frames across families.

MissingnessPolicy dataclass

What to do with missing values after join.

SliceOverride dataclass

Optional per-feature overrides. - lookback: extend the data fetch window backwards to support lagged features. - grid: override the grid for this feature family. - asof: override global asof (rare; useful for PIT macro)

TargetRequest dataclass

Target builder (can be implemented as a FeatureTemplate that returns a 1-col frame or Series).

TimeSpec dataclass

Global time/grid settings for the dataset build.

UniverseSpec dataclass

Which entities we are building the panel for.