torch_concepts.annotations.Concept¶

class Concept(name: str, index: int, cardinality: int, type: str, states: ~typing.List[str] | None, slice: slice, metadata: dict = <factory>)[source]¶

Read-only, per-concept view over a single column of an Annotations.

Groups one concept’s properties into a single named object so callers can write annotations.concept('color').cardinality instead of the index-dance int(annotations.cardinalities[annotations.get_index('color')]). It is a view: the values are read from the owning Annotations’ parallel lists at construction time (no duplicated storage), so the lists remain the canonical representation.

name¶

Concept label.

Type:: str

index¶

Concept-level index within the axis.

Type:: int

cardinality¶

Number of states (1 for binary/continuous scalars).

Type:: int

type¶

Concept type, one of 'binary' / 'categorical' / 'continuous'.

Type:: str

states¶

State labels for this concept.

Type:: Optional[List[str]]

slice¶

Column span of this concept in the flattened (logit) tensor.

Type:: slice

metadata¶

Raw per-concept metadata (escape hatch for extra keys).

Type:: dict

__init__(name: str, index: int, cardinality: int, type: str, states: ~typing.List[str] | None, slice: slice, metadata: dict = <factory>) → None¶

Methods

__init__(name, index, cardinality, type, ...)

Attributes

`is_binary`	Whether this concept is a single binary value (cardinality 1).
`is_categorical`	Whether this concept is a multi-state (categorical) variable.
`is_continuous`	Whether this concept is continuous.
`name`
`index`
`cardinality`
`type`
`states`
`slice`
`metadata`